Yes, I’ve been away for a while. Yes, this is 50K words. It’s worth it. Read on.
Chapter 1, Work and Wages in iWorld:
1. Cowen lays out some facts:
1.A. Real wages for young people are down, and unemployment (or, more precisely, underemployment and the labor-force dropout rate) is up.
1.B. The situation is international
1.C. Meanwhile the very top earners are earning much more.
1.D. The Great Divergence (Or Great Bifurcation, or Great Skewing) is widespread and expresses itself in many dimensions, good things are correlated with other good things, likewise bad with bad.
2. This is because:
2.A. Automation and increasing productivity of industrial robots and intelligence machines (intelligence used in a weak sense, mostly any IT).
2.C. Split of the economic into very stagnant (Baumol services, old tech) and very dynamic (new tech) sectors.
The key questions will be: Are you good at working with intelligent machines or not? Are your skills a complement to the skills of the computer, or is the computer doing better without you? Worst of all, are you competing against the computer? Are computers helping people in China and India compete against you? … Ever more people are starting to fall on one side or the divide or the other. That’s why average is over.
There is now a joke that “a modern textile mill employs only a man and a dog – the man to feed the dog, and the dog the keep the man away from the machines.”
Not each and every one of these innovations will pay off. But let’s ask a few questions.
First, in which major areas do we see ongoing technological advances exceeding expectations from just a few years ago?
Second, in which areas do we see a lot of new and promising technological works in progress?
Third, in which areas can we expect the general forces propelling innovation … to remain powerful?
Finally, can we see evidence that these areas are already influencing economic statistics measuring our nation’s well-being?
I’ll get into more detail on all of these questions, but for now the point is that the areas of the economy identified in the answers to these questions all overlap on one technology: mechanized intelligence.
3. Human beings are more predictable, readable, and ‘exploitable’ by sensors and algorithms than many people would comfortably accept or want used against them. But these statements are true, they are exploitable, and so they are going to be exploited, and the world is going to change a lot because of it. This will creep people out at first, precisely because they will be so effective.
This [Franchise, by Asmiov] may sound outrageous to many. It seems to cross a precious line of liberty and freedom. But perhaps we are not as free as might think in the first place. … the future of technology is likely to illuminate the unsettling implications of how predictable we are and indeed in 2012 political campaigns invested heavily in predicting where to find supporters and important swing districts.
Whether we like it nor not, our sparring partners will use mechanized intelligence during our business contests
Wasn’t that in the movie of Crichton’s Rising Sun (1993)? We’re going to get a world where people actually are under the social equivalent of the Eye of God, where everyone has plausible access to all your embarrassing past sins even more than today, so it really does make people God-fearing and suppresses transgression. Of course, that includes political and ideological transgression.
Eventually it will be commonly understood that such analyses are going on in real time. Negotiators will be trained to fool or otherwise throw off the voice-tracking programs. In turn the programs will be improved to keep up with these tactics, setting up a never-ending “arms race” between technologies of deception and detection. And a new kind of sophisticated social interaction will develop. That is bigger news than any new gadget.
… We may tend to think of mechanized intelligent analysis as primarily useful in judging other people, but it will also have the potential to promote self-knowledge. During a date, a woman might consult a pocket device in the ladies’ room that tells her how much she really likes the guy. The machine could register her pulse, breathing, tone of voice, the level of detail in her narrative, or whichever biological features prove to have predictive power.
The sorry truth is that if we knew all or even some of the bad things about our prospective partners, we might be so cautious that we never take a romantic leap. As it stands, the world is set up to overreact to negative information, as even a whiff of scandal causes us to lose trust in other individuals. We will need some significant cultural changes to make do with an increase in the “warts and all” coverage mechanized intelligent analysis will soon be delivering about virtually all notable public figures and many private individuals too. … The positive illusions (all out children are above average) that help us get through everyday life could so easily wither in the face of sangfroid machine critiques. Not this year or next year, but most likely within our lifetimes
4. Tech is improving faster than other sectors because more lightly regulated.
It is no accident that we are seeing so many signs of significant progress in mechanized intelligent analysis, albeit in varying stated of maturity. First, Moore’s law about ongoing advances in processing speed has continued to pay off, with no immediate end in sight. Second, the machine intelligence sector is largely unregulated. If you compare it to health care as a world-altering, stagnation-ending breakthrough industry, regulatory obstacles are a far greater problem for pharmaceutical companies and for hospitals than for the like of Google, Amazon, and Apple. Health care, with its physician licensing, Byzantine hospital regulations, and FDA approval process, also makes most of its changes quite slowly, for better or worse.
Still, very often entrepreneurs and scientists can do the work behind smarter machines, and develop usable products, without need much special permission from the powers that be. … Technological progress slows down when there are too many people who have the right to say no, but software in general gets around a lot of the traditional veto power point.
Part I, Chapter 2: The Big Earners and the Big Losers
…here is what is scarce:
1. Quality land and natural resources
2. Intellectual property, or good ideas about what should be produced
3. Quality labor with unique skills
Here is what is not scarce these days:
1. Unskilled labor, as more countries join the global economy
2. Money in the bank or held in government securities, which you can think of as simple capital, not attached to any special ownership rights (we know there is a lot of it because it has been earning zero or negative real rates of return)
I. Marketing. Echoing Sailer:
Despite all the talk about STEM fields, I see marketing as the seminal sector for our future economy.
Here’s a recent and related article of his at Bloomberg.
II. Personal Services, people are going to treat the rich like royalty all the time, and constantly try to get access to them:
We can expect a lot of job growth in personal services, even if those jobs do not rely very directly on computing power. The more that the high earners pull in, the more people will compete to serve them, sometimes for high wages and sometimes for low wages. This will mean maids, chauffeurs, and gardeners for the high earners, but a lot of the service jobs won’t fall under the service category as traditionally constructed. They can be tough of as “creating the customer experience.” … All of those people are working to make you feel better. They are working at marketing.
It sounds a little silly, but making high earners feel better in just about every part of their lives will be a major source of job growth in the future.
He uses the metaphor of Calcutta beggars all climbing over each other to try to get the favorable attention of a passing billionaire. That’s a lot of people in the future. Billionaires will both be flattered by that, but also quite annoyed and one can expect them to become a bit calloused to it all and put up, for lack of a better term, their ‘bitch shields’ to the army of pitchmen, or force them to go through gatekeepers. That reminds me of this classic Roosh post.
For high earners, life will feel better than ever before, but at the same time, life will feel more harried and more overloaded with information than ever before.
III. Who is earning? Mostly managers, finance, and law.
If we look at the increase in the share of income going to the top tenth of a percent from 1979 to 2005, executives, managers, supervisors, and financial professionals captured 70 percent of those gains.
IV. Scrutiny and Hyper-Meritocracy.
We are going to be scrutinized at work much more closely, continuously, and much more accurately, than most people are used to, or will be comfortable with. There is a joke. “In the past, they said that in the future people would only work 3 hours per day. They were right. What they didn’t know is that, in the present, we still sit in front of our desks for the other 5 hours, pretending to work.” That joke won’t be funny soon, and it may not ever make sense to future workers.
Another development is this: The better the world is at measuring value, the more demanding a lot of career paths will become. This is why I say “welcome to the hyper-meritocracy” with a touch of irony. Firms and employers and monitors will be able to measure economic value with a sometimes oppressive precision.
V. Scarce and Surplus types of labor:
In any case, the slacker twenty-two-year-old with a BA in English, even from a good school, no longer has such a clear path to an upper-middle-class lifestyle. At the same time, Facebook, Google, and Zynga are so desperate for talent that they will buy out other companies, not for their products, but rather to keep their employees. It’s easier and cheaper to buy the companies than to try to replicate their recruiting or lure away their best employees. Often the purchased product lines are abandoned. … The technology blogs call this being “acqhired,” …
Heh – business opportunity. Out-compete the big boys recruiting talented and undervalued engineers out of college to work at a ‘start up’ and paying them little or just in promises of options or something. Have them work on some project which gets attention but otherwise doesn’t have to make any genuine business sense. Maybe do a culling down to the best half. Then tell Google you’ve got an engineer-plantation they can buy, slaves and all. Google doesn’t care about the land or the cotton, it has its own groves, but it likes the slaves, because you are known to have good taste in slaves. Profit.
VI. Good managers are the scarce input in many operations, which is why their salaries have gone up so much.
He says that top folks simply don’t have enough time to invest in managing more people. He gets offers from good people to work as an additional research assistant for him for free, and he still turns them down, because he has no more time. He would have to get a middle manager to take on more people, even free people, and he can’t do that cheaply.
To hire a risky and iffy worker, without a competent overseer, simply isn’t worth it, no matter how low the wage. And so a lot of workers have a hard time being picked up and integrated into productive teams.
It is precisely that process that managers are paid to make work more efficiently. It is a process that is continuing its long, long trend towards increasing importance. And finally it is why managers are being paid more.
VII. Workers need to be more conscientious and happily obedient to be valuable these days.
Team production makes the quality of “conscientiousness a more important quality in laborers. Managers need workers who are reliable. If you have a team of give, one unreliable worker is wrecking the work of four others. …
It’s not just that the bad workers are lazy or maybe destructive. It’s that low quality workers spread bad moral to many others …
VIII. Conscientiousness not equal between sexes
The growing value of conscientiousness in the workplace helps women do better than men at work and in colleges and universities. At my daughter’s recent college graduation ceremony the awards for the top achievers in all the school’s programs and departments went almost entirely to women, including awards in science and mathematics.
I’ve seen this during my education too. They didn’t necessarily get the choicest jobs, but boy did they rack up the good grades awards.
It is well known from personality psychology, and confirmed by experience, that women are on average more conscientious than men. They are more likely to follow instructions and orders with exactness and without resentment.
You can think of men as the “higher variance” performers at work. That means some men are more likely to be the very highest earners and also the exhibit extreme dedication to the task … Other men, in greater number, will be more irresponsible, more likely to show up drunk, more likely to end up in prison, and more likely to become irreparably unemployable.
Didn’t Larry Summers get fired from his Harvard Presidency for saying something similar?
Here is another, more general way to think about the shifting gender balance of power in education and parts of the workplace. The wealthier we become, the grater a cushion we have against total failure, starvation, and other completely unacceptable outcomes. In such a world, both women and men will indulge some propensities that otherwise might be stifled or kept under wraps or that would not have been affordable fifty or one hundred years ago. For some men, these propensities are quire destructive and this turns them into labor market failures.
And some of the women turn into sexual market failures. Anyway, it’s all about minimizing management and supervision costs, especially through early screening and filtering.
The premium is on conscientiousness, namely whether the worker can follow some straightforward requests with extreme reliability and basic competence. … If you’re a young male hothead who just can’t follow orders and you have your own ideas about how everything should be done, you’re probably going to have an ever-tougher time in the labor markets of the future. [Smoky! -H]
Let’s draw up a simple list of some important characteristics …
1. Exactness of execution becomes more important relative to accumulated mass of brute force
2. Consistent coordination over time is a significant advantage
3. Moral is extremely important to motivate production and cooperation
IX. Bad workers are potentially huge liabilities in lots of ways, including of course disastrous legal consequences. Exclusion of even slight risks is key.
Workers represent a firm to the broader outside work, and the firm faces a higher risk of lawsuits. … It is easier to destroy than to create, and the more valuable and the more precision-based that firms become, the more they will worry about destruction of value coming from workers.
Any time there is a discussion of management strategies, you probably will hear a lot of works like teamwork, morale, and integrity. That’s all well and fine, but what if we substitute exclusion for all those nice warm phrases. They would be the same management strategies merely explained from a different point of view, namely of those who are kept away. There is no high morale without exclusions, no integrity without exclusion, and no corporate culture without exclusion. If the management styles at today’s quality companies seem so nice, so friendly, and sometimes so downright heartwarming, it is possible only because those cultures are so very picky, snobbish, and elitist at the same time. There is no open door.
X. Careers – the brightest, best-compensated, non-STEM people are going into finance, law, and consulting.
X.A. Law: Today, laws are more numerous and more complicated increasing demand for lawyers, at least at the top end.
X.B. Consulting: A global economy means longer supply chains, and consultants can help businesses track and evaluate those complex operations (he goes on to say that many business managers never step back and do real intellectual ‘fresh look / big picture’ analysis of their operations.
X.C. Finance: Growing in part because the promise of bailouts encourages banks to become larger and take on more risk (your mileage may vary with this one)
Working to exercise and demonstrate their general intelligence is in fact the main thing they are good for, and moving beyond this can take quite a few years. …
We tend to glamorize these well-paying jobs. If we can set aside the glamour and perhaps our envy, me might notice that our society does not know what else to do with these people, who are otherwise not always very productive.
I’ve been very busy with work and social obligations in the last few days, so I won’t be able to make a detailed comment for another day or two. However, for now, I’d just like to point out that this stuff appears to be optimized for injecting a dose of reality into the mainstream in a way that tries to avoid treading on any crimethink mines while minimizing the inevitable distortion of truth that follows from that. I think Cowen should be given much credit for doing it so successfully.
Heh – business opportunity. Out-compete the big boys recruiting talented and undervalued engineers out of college to work at a ‘start up’ and paying them little or just in promises of options or something. Have them work on some project which gets attention but otherwise doesn’t have to make any genuine business sense. Maybe do a culling down to the best half. Then tell Google you’ve got an engineer-plantation they can buy, slaves and all. Google doesn’t care about the land or the cotton, it has its own groves, but it likes the slaves, because you are known to have good taste in slaves. Profit.
This might actually be a part of the explanation for all these startups that appear to be wildly overvalued relative to the merits of their ostensible business plan.
I know a Tesla mechanic and he really likes his job. He used to work for BMW, and said it had a truly toxic culture (not one that sounded very traditionally German) and the rats (i.e. other mechanics) were fleeing from a sinking ship. A former BMW maintenance manager was poached by Tesla, and he knew who the good guys were at BMW, and so was given them task of poaching them too.
Which really make you think.
One thing Tesla has is that anyone who can create a new car company from scratch will maintain a permanent advantage over all established car companies, in that it won’t be saddled with all those tremendous pension liabilities to former workers, and established super-powerful unions. Musk certainly has an incentive to get as far ahead on the automation curve as possible to avoid ever having to deal with those problems at anything like the magnitude of burden all the other companies must carry.
That makes it very hard for any established company to eat his lunch by copying simple and widely available tech, while also making it hard for any other new company to overcome the barrier to entry, especially if future subsidies are likely to be less generous than what Musk got to help him get started. That means there is a special, one-time opportunity to pick up this particular $100 bill off the sidewalk. He picked it up.
I admit I didn’t give this particular advantage enough consideration before, and now it seems to help account for Tesla’s unique ability to capitalize on electric cars with big batteries, which, after all, anyone can make. But his timing means that he’s the only one that can make them both with the most generous subsidies and before amassing manufacturing-era labor liabilities and before sclerosis infects his company.
It’s not necessarily regulatory arbitrage as it is also a kind of legacy sclerosis arbitrage. Indeed, this was and remains a considerably portion of the competitive advantage of East Asian automakers in the US market. All else being equal, the Big Three had to make an extra few thousand dollars per vehicle to pay for their liabilities. Tesla gets to start from scratch with a clean slate. That just having a clean slate is such a huge advantage these days is revealing in itself. Combined with ludicrously generous crony subsidies, it makes a strong case for his special, inimitable position.
Furthermore, in addition to not being saddled with the unions and all those pension liabilities to former workers, he’s got another advantage which accrues to any new company in an established sector, indeed one the big Silicon Valley companies have conspired among themselves to avoid by means of forming a labor-market demand-side cartel.
I’m guessing a lot of your work environments are a lot like mine, where compensation is fairly flat and compressed and bears little relation to ones marginal productivity in the short term, despite everyone knowing informally who is really pulling the weight. In the long term high performers are rewarded with promotions, but this suffers from Peter Principle problems, and anyway only works in tall hierarchies. There is a new employee where I work who is getting paid nearly as much as I am, but who is doing 20% of the work, because he is a moron, but he beats everybody in seniority, which is, alas, how the system works. He won’t get promoted, but in a way that’s almost worse, since the good performers will leave the job and people like him will stick around, lowering average productivity.
Everybody I know has lots stories like these.
So that creates another kind of obvious arbitrage opportunity. Maybe “Productivity Correlation Arbitrage.” If one could only pick one good manager in a unit or office, tell him he must fire 60% of people, and that he has unlimited authority to fire anyone he wants, and those he retains will get paid double so long as all the work gets done, then I have no doubt that the company and everyone left will be much better off.
Some seasonal companies actually do something like via over-hiring, automatic attrition, and selective rehiring. I had an uncle-in law who worked a job like this on the Alaskan oil fields and called it something like an “underbrush fire” that left all the big timbers standing.
But most mature organizations, especially those saddled with strong unions, can’t legally or practically manage anything remotely approaching this kind of ruthless culling.
But if a new company can poach a few good managers with the special inside knowledge needed to be future poachers of more good people, then your new company can start off with much better people producing much more value and for only a little more money. Is Tesla doing this too? That’s pretty smart, and it seems to borrow from some insights that may have been gained from Silicon Valley experiences.
Hmm… something to think about.
Part I, Chapter 3: Why Are So Many People Out of Work?
I. The labor force participation rate has been going down for some time.
Cowen includes this picture (though timed to lack the little upward spike at the end)
Well, 67 to 63 in 15 years seems pretty dramatic.
But let’s constrain to prime working age adults:
Not as dramatic. Let me abuse faulty human visual pattern finding here (something between pareidolia and apophenia). What I want is to do a Fourier transform and get out the seasonal signal, but also the business cycle oscillation with a bust in the early 90’s, boom in the late 90’s, bust in the early 00’s, boom in the mid to late 00’s, and then the GFC big bust until, well, now. What I want to say is that the smooth hump that would be left would have peaked around 1996 just above 84%, and in 20 years has declined to over 81%. The questions are whether that 3% is gone for good, and where things go from here.
A big part of the difference is more higher education. Not all the difference, but a lot. A lot of people criticize Cowen, saying he is purposefully ignoring that and so exaggerating the problem. IIRC, he has responded that it still means that our economic structure makes more bodies unavailable for production at any moment, but the issue here is supposed to be ‘involuntary unemployment’.
II. Here is how the step by step evolution of machine intelligence worked in Chess (an obsessive analogy with Cowen, for obvious reasons, for much of the book.)
This step-by-step evolution is how intelligent technology will change a lot of industries.
At first the machine hardly adds anything and it’s really just an investment in building a better machine.
At the second step, experts – the in field of the program’s operation – will be required to work with the machines, to fill the gaps in what the machines can do.
As the programs improve, the next and third step is that the humans understand the programs very well, with a minimum of expertise – but expertise nonetheless – in the relevant industry. These workers will essentially be information processors, albeit with an understanding of context.
The fourth and final step is that the human isn’t needed much at all because the program on its own is so strong.
Computer programs do especially well in chess because it is a totally regularized environment where the right answer can be ascertained, at least in principle, by pure calculation.
… In poker, the very best players are still humans, because the computers don’t know how to psych out the opponent, bluff, or real the “tells” from the guy sitting across the table. The more that an endeavor requires inferences about the mind-states of others, the more than intelligent machines will require human aid. We humans do have out talents.
III. Back to labor force participation
Those numbers on labor force participation are telling us that, for whatever reason, over 40 percent of adult, non-senior Americans don’t consider it worthwhile to have a job. They can’t find a deal that suits then.
… Adult males are seceding from the workforce – or being kicked out – in frightening numbers. Few of these individuals are wealthy playboys. Is it no surprise that popular culture today has this image of the male slacker, a young man who lives at home, plays video games, is indifferent to holding down a job, and maybe doesn’t run after young women so hard.
… People are getting accustomed to an existence where they cannot find satisfying work at a wage they are happy with.
… Ten years ago, 5 million Americans collected federal disability benefit; now the number is up to 8.2 million, at a direct dollar cost of $115 billion a year, over $1,500 for every American household. Yet the American workplace, as measure by deaths and accidents, has never been safer.
The number actually peaked near 9 million in Sept, 2014, and has been mostly flat ever since, with a small but steady decline afterwards. The number of applications and awards (the first derivative) peaked in late 2011 and has declined over 25% of average since then. The Disability Trust Fund was set to go bankrupt in late 2016, and was expected to lose $30 Billion a year indefinitely.
The ‘fix’ was in the “Bipartisan Budget Act of 2015”, which shifted 0.57% of the 12.4% in total payroll taxes from the old age program to the disability program, but only for three years (so, spending 5.4% of ‘pension contributions’ to bail out a formally financially segregated insurance program). DI tax receipts, which had been mostly flat for 8 years, suddenly jumped 35% in 2016, by over $40 Billion extra dollars, pushing off the insolvency of one trust fund, at the cost of accelerating the insolvency of the other trust fund, but in the end only delaying the inevitable DITF bust for about 5 or 6 years, unless there’s another bailout, which everyone knows there will be. This was one of the most under-reported story of that year, and not just because of the election and Trump.
These problems with labor have gone beyond the general problems with our economy, so something has gone wrong with work itself.
… But for men, from 1969 to 2009, as measured, it appears that wages for the typical or median male earner have fallen by about 28 percent.
He admits some people dispute that number, but even some rosy assumption come up with a pretty disappointing and surprising number.
Imagine yourself as an economist back in 1969, being asked to predict the course of American male wages over the next forty years or so. You are told that no major asteroid will strike the earth and that there will be no nuclear war. The riots of the 1960s will die out rather than consuming out country in flames. Communism would go away as a major threat and most of the world would reject socialism. Who would have thought that wages for the typical guy were going to fall?
Ctrl-F for ‘immig’ comes up nothing, and maybe those numbers should have been stipulated as well to our hypothetical 1969 economist. And there were some people on the nationalist, anti-globalist right who thought wages for the typical guy would be hurt by a more open economy with more open borders. But there were not dominant voices, to put it mildly.
IV. The Great Recession. During the boom, firms weren’t paying much attention to granular productivity. After the bust came the microscopes and the firings.
Firms … took some discrete steps to figure out which workers were adding the most value, and once they identified the less productive workers, they let them go.
… Those laid-off workers have been absorbed into new jobs at a rate much slower than is usual following a recession. They can’t get their old jobs back, even though the worst of the crisis is over and corporate profits are back up. Most importantly, the new jobs being created are more likely low wage than mid-wage.
… most labor market polarization is transmitted through the immediate mechanism of recessions, which is when those middle class jobs are disappearing. After the recession is over, the lost middle class jobs do not come back.
V. Putting aside problems with short-term nominal stickiness, lower wages still can’t fix the problems with contemporary low-productivity labor, as in the past, because today’s suspect workers just aren’t worth the trouble.
“Seeking only workman’s wages I come looking for a job, but I get no offers. Except the come-ons from the whores on Seventh Avenue …”
It doesn’t matter how flexible the wage is in the more complex, less brute force jobs. A manual worker who just shows up at your door is probably not someone you want to hire unless it is already part of a preexisting business plan with broad buy-in from your enterprise and your creditors. The worker might say, “I’ll lower my wage demands by thirty percent!” or, “I’ll work for nothing!” It usually won’t matter. The sad reality is that many of these workers you don’t want at all, even if the business plan involves additional labor. Some workers simply aren’t worth the trouble unless the demand for extra labor is truly pressing.
I believe these “zero marginal product” workers account for a small but growing percentage of out workforce.
During the surge and temporary force-builds, the Army and Marines had to lower standards and accept less impressive applicants in order to meet accession quotas for enlistedmen. Usually that involved relaxing each of the many standards each by a little bit. Actually, the system pretends the standards aren’t being changed at all, but that individuals are being granted discretionary ‘waivers’ of a typical standard on a one by one basis by commanders, which is the system ordinarily used rarely in exceptional cases for people with extreme talent or value in some area, but maybe just under the threshold for one of the standards. Well, suddenly these waivers were routine. Still, there is value to keeping the standards ‘in the book’ the same, since everybody still knows what they are supposed to do, and the waivers will eventually go away when the pressure is off.
But eventually you are going to be cutting into muscle and bone and not able to relax some standards any more. And someone is going to discover where you are going to get the most bang for your buck in terms of the greatest numbers resulting from a policy change in the other standards. That turned out to be in background check department, which gave rise to the whole ‘moral waivers’ problem. A lot of these guys were good soldiers, fit enough and smart enough to fit in, go fighting downrange, and get the job done well, but, inevitably, a huge number of them got into serious disciplinary trouble at some point. They were good workers who would get in trouble, which is a very different problem from the obedient and law-abiding ones that just aren’t up to snuff.
In times when men were desperately needed, when those men got in trouble, they’d get slapped on the wrist with minor penalties, or even just a good old-fashioned “smoke the shit out of him” extended painful-exertion session with an NCO. But as soon as Congress announced the numbers had to go down – by a lot, and quickly – then a very different message went out to commanders. Suddenly every little thing was a dischargeable offense, and it was, predictably, disproportionately the moral-waiver guys who were getting kicked out.
VI. Productivity. The labor-productivity statistics following the GFC bust bear this story out.
Cowen says that in a typical recession, especially under simple Keynesian Aggregate Demand models, one might expect to see job losses occur across the whole economy and in each sector in a fairly proportionate way. Everything should just shrink, and average productivity should remain the same. But that’s not what happened. In many firms, hours fell faster than output and productivity increased a lot. That means that this time firms went to effort to identify the gold and the dross, and then disproportionately got rid of dross. What’s worse is that while some of those people weren’t dross, everybody knows that people who lost their jobs and became unemployed were disproportionately dross, and so other firms were reluctant to hire them back, based on this statistical generalization. And that means re-employing those people was a different problem from the one of past recessions.
Some of the JOLTS data bears this out too. Firms were hiring, but disproportionately from pools of college graduates on the one hand and people who already had jobs and were just switching companies on the other. They weren’t hiring in anything like a similar proportional rate from the giant reserve army of unemployed.
One quibble with this data is that something like that might still show up as an artifact in some large firms that are just indiscriminately firing line workers, but which can’t cut overhead in headquarters, where people are paid the most. Still, I’m guessing that’s probably a minor issue.
VII. There are plenty of new, if lower paying jobs. However:
There are plenty of lower-paying jobs in the world, more than ever before, but here are the rather significant catches:
1. A lot of those jobs are being created overseas. If the job does not require high and complex capital investment, the advantage to keeping that job in the United States is lower.
2. A lot of Americans are not ready to take such job, either financially or psychologically. They have been conditioned to expect “jobs in the middle,” precisely the area that is falling away.
3. Through law and regulation, the United States is increasing the cost of hiring, whether it be mandated health benefits, risk of lawsuits, or higher minimum wages.
It is hard to escape the conclusion that unemployed young workers will only slowly be reemployed. And the jobs they get will often have considerably lower wages.
VIII. Freelancers and self-employed and food trucks:
Rono Economou … is a typical story. She was laid off [from] her well-paying job at a large Manhattan law firm … After some soul-searching, she responded by opening Boubouki, a small Greek food stall … she wakes up at 5:30am, lift a lot of heavy bags, and can’t afford to miss a day of work. It’s not clear her project will succeed financially, much less bring her riches, and it also doesn’t seem that her life is freer. A lot of future jobs will look like this – that is, they will look more like the jobs we already see in great numbers in developing countries.
Over time we can expect these categories to blur, and freelancing jobs will become increasingly respectable and indeed normal, if only because they offer a bit of pay and a bit of personal freedom too. More workers will think of themselves as free agents, and more employers will be keener to make hires without traditional benefits packages being attached to the job offers.
If the law lets them anyway, and allows independent contractor relationships without benefits, overtime, or guaranteed regular hours.
However, I have a friend who despite being quite gifted was just a bad fit, personality-wise, for the boring daily grind of a middle-management bureaucrat. Too restless and athletic and temperamental. Not the family man type at all; just a little too much wild blood in there.
He ended up quitting his civil servant job and driving Uber / Lyft full time. He was making good money for a while, and now he makes ok money, enough to get by in an expensive city, though of course with no pension and only the minimum health care plan.
It’s a lot less than he was making in his regular job, and his net per ride is decreasing as the situation gets more competitive and the market-makers turn the screws. Also there is volatility and seasonality and he has had to bust his ass to avoid hitting the wall a few times. Also, if he ever does decide to try and settle down and rejoin the workforce, then he’s got a whole lot of Uber on his resume, and no references.
Still, if he’s got enough money saved up, and he gets an invitation or sudden opportunity or just a feeling and wants to take two weeks off on a whim to go climbing or biking or to pursue some love affair, or just go on a bender, he doesn’t have to ask anybody for permission. He just goes on the spur of the moment. He waits for those last-minute crazy international deals out of his airport and if the price is right he just goes. If he runs low on cash or wants to make some indulgent purchase, he can increase his hours whenever he wants. If he gets insomnia or some chick flakes on him or something, boom, he can just turn on the app and start making money then and there. He’s actually met a girl this way once. He says, “I have no supervisors or clients and no one is responsible for me but me. I am a free man. I love it.”
To him, this kind of life is satisfying, if hard, low paying, and low status, but it feels like it is still full of dignity because of his independence and freedom from having to submit to anybody or any schedule other than his own. I am almost of opposite temperament to him in many ways, but even I’ll admit that there are times when that grass looks a little greener and I slightly envy his flexibility. But I’m fairly certain his standard of living simply must descend, eventually, to that of his most hungry competitor. Which is to say, next to nothing.
IX. Threshold earners, and their culture:
Today, many of these young earners are threshold earners, meaning earners who are content just to get by and who do not push ambitiously for a higher wage or stronger credentials at every step. Williamsburg, Brooklyn is full of young threshold earners, although rising rents are starting to push them out into the other parts of the city …
… it is commonly recognized that a lot of the young denizens simply aren’t striving after very much, at least not in terms of commercial job opportunities.
Overall, these job market trends are bringing:
1. Higher pay for bosses
2. More focus on morale in the workplace
3. Greater demands for conscientious and obedient workers
4. Greater inequality at the top
5. Big gains for the cognitive elite
6. A lot of freelancing in the services sector
7. Some tough scrambles for workers without a lot of skills.
Those are essential characteristic of the coming American labor markets, the new world of work.
Part II: What Games Are Teaching Us
Chapter 4: New Work, Old Game
I. Gaming is a huge part of the economy now, bigger than Hollywood, which is an under-reported and under-recognized fact. They also emphasize rapid processing of large amounts of information, which is what will be especially valuable in the future.
II. Watching how intelligent machines play and eventually beat the top masters in Chess (and recently, Go) provides a rich source of data about the tendencies, strengths and weakness of human decision making.
In the past, without the ability to ‘verify’ whether a decision was really the ‘correct’ one through exhaustive computational analysis, only other top human experts could evaluate a decision by another top human expert. And they would have to use their fallible heuristic intuitions to do so. What’s more, if there is something common about brain architecture that makes these intuitive engines fail in common ways, the expert consensus will evaluate another expert’s ‘wrong’ move as right.
But now we can really scrutinize these things with the help of machines, and as a result we can learn about the power and limits of our own brains, or, at least, the best our species has at the moment before the Chinese pay Stephen Hsu to use CRISPR to create biological super-intelligent super-alpha-Han or something.
III. Computers make a lot of ‘ugly’ moves, that feel wrong or weird to the pattern-recognition heuristic intuitions of most human players, and which full of complexity and mystery.
Chess grandmasters have coined a phrase – “That’s a computer move” – to describe those ugly, counterintuitive decisions made by computers, the moves that surely appear wrong. Yet the machines that produce those ugly moves beat the grandmasters virtually every time.
The moves of the machines show, regularly, how puny and unreliable our intuitions are, even if we spend our decades studying chess.
It makes you wonder if the same is true about the rest of our lives.
IV. Partnering with machine-intelligent advisors and playing to win in broader human contexts.
It may get riskier yet, as the computers are programmed to play an active, tactical game. The computer is programmed to play for a win, not a draw. We can imagine competing intelligent-machine companies offering programs that seek out an active advantage in a typical human situation. No one rises to the top of the business world by breaking even on a lot of deals, and no one successfully woos a lot of women, or marries the right one, by acting “just okay” or neutral. People know that they need to take chances in complex situations, and they will buy tactical computer programs that help them do this. We’re going to generate a lot of hairy, very complicated personal interactions, driven by real-time data analysis and computer intelligence.
Average is over. Some real-world interactions will become a lot simpler and call for conservatism and simple rule-following behavior, while others will become far more complicated and extreme. The case for keeping it simple is plain: Just do what the machine tells you. Avoid mistakes, hang on to your job, your relationship, your portfolio, or whatever it is you are trying to preserve. Defer to the authority of the beast with the intellectual brute force.
Part II, Chapter 5: Our Freestyle Future:
I. A lot more chess analogy stuff. Freestyle (or ‘advanced chess’ or ‘cyborg’ or ‘centaur’) is human+machine teams playing other human+machine teams. If H is human and M is machine, you can imagine the possible competitions as:
H v H (traditional)
H v M (human playing computer, will always lose now)
M v M (software playing itself or another piece of software)
H+M v H (Freestyler versus human, human alone should always lose)
H+M v M (Freestyler versus some software, maybe different software. If it’s the same software, and the human is expert, then the Freestyler should probably win.)
H+M v H+M (Freestyle competition)
I am leaving out the complication of there being multiple humans, or multiple machines, or both, with maybe a team of people trying to figure out how to decide between multiple options when different chess engines give different answers.
And that’s the example Cowen uses with the champion British diversity-poster-children team of Anson Williams (Afro-Caribbean), Nelson Hernandez (looks Spanish to me), and Yingheng Chen (Anson’s girlfriend).
Anson, when playing, is in perpetual motion, rushing back and forth from one machine to another, as Freestyle chess is, according to team member Nelson, “all about processing as much computer information as rapidly as possible.”
Freestyle teams study the opening moves their machine opponents have made in previous games because, as Kasparov has observed, an initial advantage in Freestyle chess usually means an eventual victory. The players also know the weaknesses of particular engines and how one engine can at times offset the weaknesses of another.
Well, I suppose what we have here are kind of a meta level of intuitive heuristics, bolstered by ‘intense probing’ of each engine’s offered move when there is a ‘evaluation flip’ opportunity. How do the humans know about these strengths and weaknesses and make decisions? Why aren’t those intuitions susceptible to the same problems as in standard chess? My point is, why can’t that be automated in a similar fashion by a kind of meta-chess engine. Of course, if you have meta-chess engine software, combined with all the other engines, you have a super-engine. And maybe then we need to go one level higher in the hierarchy of engines and meta-engines, and meta-meta-engines …
II. More chess …
III. Having good and quick memory is important in these games, and in the real world too
Indeed, in plenty of real-world situations the immediate command over factual or analytical material brings a big edge. Discussions in meetings, strategies and reactions during sales calls, lawyers arguing in front of a jury, and managers in volatile, voices-raised personnel situations all try to draw upon preprocessed information at a moment’s notice. In all those cases, it matters more and more what workers have learned from the computer, or not, and how well they remember computer-derived information and advice.
IV. Another real-world ‘freestyle’ combo example. Medical diagnosis.
For over 20 years there have been automated imaging systems for histology and pathology and cell screening, such as in pap smears, blood draws, and biopsy samples. It’s also true in radiology. There’s also been automated EKG analyzing software to help cardiologists for longer than that. (In fact, little Handle once worked in the cold room at a local university hospital where the ‘supercomputer’ did exactly that, reading the data off old cassette tapes and making preliminary diagnosis, often very well, sometimes comically wrong.)
These systems are under the ‘supervision’ of an expert human and try to divide things up by ‘comparative advantage’, but really there is some feedback and complementary cross-checking to compensate for the other party’s weaknesses.
For pap smears, there can be lots and lots of cells, and the computer never misses a potentially abnormal one, but it has a false positive a lot of the time. It sends only these images to the doctor who studies them more closely. But that processing is done over thousands of cells and completed in a flash, which saves the docs lots of valuable time. The ‘team’ works like this:
Machine: Many, Easy. Filters down to:
Human: Few, ‘Hard’
Of course, what is ‘easy’ or ‘hard’ is a matter of the latest software and how smart it is which, in the short term, is probably also a matter of expense. But as machines get smarter and faster, the margin will keep shifting until eventually the human doctor isn’t ‘needed’ at all. That is, he adds zero value to making the results statistically more reliable.
V. On the other things, it will be hard to replace human agents for some diagnostic tasks, because of GIGO problems and that whole poker-like ‘reading the state of a human mind’ problem.
One medical innovation would run a patient’s reported symptoms through a Watson-like software program and see what might be wrong, drawing upon extensive databases. But can the computer ask follow-up questions to the patient properly or guess where the patient might be lying or exaggerating in the description of symptoms? … Not anytime soon, and so we are back to collaboration.
Actually, I think the computers might be pretty good at figuring out if someone is lying or exaggerating, or helping tip off the doctor or nurse.
VI. Implications for credentialed guild professionals:
It is clear that for the collaboration to work, we need to have a very smart machine. But, if the machine is already in place and plugged in, how expert does the human have to be? When the worker has to be a highly paid physician, a collaborative team can be costly, even if it improved health outcomes. The world – not the mention the American Medical Association – is pretty far from accepting this fact, but the person working with the computer doesn’t have to be a doctor or even a medical expert. She has to be good at understanding and correcting the computer’s mistakes, which is a very different skill.
VII. We are already running a kind of experiment of automated diagnosis given symptom descriptions with Google, which never had to ask anyone permission to do it, and which, apparently, isn’t liable for this going wrong. Which it must have done plenty of times. Again, ‘the tech exception’. But lots of people are typing in their inquiries and getting, it turns out, ok diagnoses.
The study did not consider the possible costs of incorrect or misleading results, so we’re still far from evaluating this rather large-scale experiment in new medical institutions. If nothing else, it’s an argument for proceeding with more regularized and authorized forms of the collaborative approach in medicine. We already have computerized doctors, and that illustrates the power of information technology to spread rapidly; the next question is how good and how reliable our mechanical medical servants are going to be.
VIII. Broader Lessons
1. Human-computer teams are the best teams
2. The person working the smart machine doesn’t have to be expert in the task at hand
3. Below some critical level of skill, adding a man to the machine will make the team less effective than the machine working alone.
4. Knowing one’s own limits is more important than it used to be.
We also can use the concept of man-machine collaboration to define the difference between a worthless or “zero marginal product” worker and a potentially valuable worker. The worthless worker is one whose cooperation with the machine makes the final outcome worse rather than better. A potentially valuable worker offers the promise of improving the machine, taken alone.
Clearly, immigration is the elephant in the room that Cowen is doing his best to ignore. Another is that a policy of protectionism of domestic workers against having their wages crushed by overseas competition (except for narrow politically powerful special interests) is no longer really in the Overton window, and he doesn’t even stop to consider if this might change.
Moreover, he doesn’t get to the core of the problem of what’s really so bad about being poor in the New Economy. The only really important aspect of this problem he does mention is real estate. But clearly, the much worse issue is the social Coming Apart that’s taking place along with the economic Average is Over, i.e. the fact that one must be somewhere in the upper deciles of the wealth and status distribution to avoid being stuck in horrifying social chaos and decivilization. Alas, even with Cowen’s skills, it’s just not possible to talk about this meaningfully without going too far in obvious crimethink directions.
Despite all that, I think Cowen is pretty much correct with all his observations, and the fundamental trends he identifies would still be taking place even if their consequences weren’t further exacerbated by immigration and overseas labor competition. As libertarians like to point out when they argue in favor of free trade and open borders, imports and immigrants — ignoring the political and other externalities of the latter — are just like technology. The argument is supposed to work by appealing to the unquestionable goodness of the latter, which supposedly only crazy Luddites could dispute. But of course, properly understood, it actually demonstrates that economic changes brought by technology itself may at least in principle have bad consequences for a majority of the population. Even if that had not actually been the case for the last few centuries, things may be changing now.
No Coming Apart in Japan. 3.5% unemployment. No immigration. No real estate boom. Increased centralization in Tokyo, yes, but nothing too dramatic. You can buy a decent house for 300k in a Tokyo suburb.
Wages have stagnated since the peak, but inequality hasn’t risen that much. No great increase in compensation for managers, CEOs and all that. Certainly no boom in financial services. Salaries for traditionally lucrative guilds such as law and doctors are if anything decreasing, especially law after the bar exam was made easier.
I see a lot of assertions about automation and human+machine teams which sound edgy and fun and make it sound like a sci-fi novel where governments are weak and engineers determine the future; but I’m not seeing much of a thorough argument.
I’ll make the obvious objection: Everything is Politics. Average is Over happened because of the particular political choices of USG. It could have happened differently. It could still change. Nothing inevitable about it.
4. Tech is improving faster than other sectors because more lightly regulated.
It’s remarkable that the word “technology” in its popular shortened form (“tech”) has come to mean “things done in Silicon Valley.” There seems to be a popular assumption, not altogether inaccurate, that nothing else happens any more in the realm of technology that’s not totally stagnant and uninteresting.
As a fascinating look into the past as a foreign country, I recommend the story “The Chief Research Chemist of the Metaplast Corporation” from Richard Feynman’s autobiography.Outside of the computer industry, I don’t think it’s possible to imagine anything like this happening today: a smart guy given the freedom to tinker around and improve and invent things, unrestrained by an onerous and rigid bureaucratic process and superiors trembling at the prospect of regulatory and legal repercussions at every step. It’s a sobering thought when people mention all this economic deregulation and liberalization that’s supposedly taken place since the 1970s.
Spandrell: Assuming your description of Japan is accurate, then according to you, what exactly were the political and other factors that have made Japan different?
I can see the following possibilities (not mutually exclusive) off the top of my head:
(1) Lack of mass immigration combined with demographic shrinkage has made the supply of low- and mid-skilled labor stagnant or falling just in time to compensate for the falling demand due to technology.
(2) As a highly cohesive and disciplined society, Japan implements protectionism and makework exceptionally well, ensuring continued prosperity even for those classes that would otherwise lose out badly in the AOE economy.
(3) Similar to Germans, the Japanese are so exceptionally productive that capital is still chasing them wherever they live, preventing the Great Centralization from playing out in a severe form. (What exactly do people in Japanese small towns do these days that’s competitive in the global economy, rather than being propped up by the state?)
(4) The Average is Over/Coming Apart trends are not just due to economic factors, but also because the old Malthusian farmer cultural and social capital is eroded further with each generation that lives in cushy prosperity, so there’s a bifurcation between the successful upper percentiles who continue with the old farmer ethic, being somehow unaffected by this degeneracy due to genes or lucky circumstances, and the lower classes who are reverting to a savage state. Maybe due to culture and/or HBD the Japanese just have the old farmer ethic ingrained much deeper.
(5) The Japanese have never done much except dull and uninspired (if still very successful) copying of American technology and institutions, and their Asiatic conformity prevents any exceptional individuals from sticking out. So there just isn’t any market for exceptional performers who would form the successful right peak of the bimodal AOE distribution.
(6) A variation on (6): maybe there are exceptional performers, but the Asiatic conformity dictates that they must not stick out in terms of compensation. (This isn’t so far-fetched: for all the frantic talk about rising inequality in the U.S., my experience is that income distribution in the corporate world is, if anything, quite communist, due to factors I don’t quite understand. People who are more productive and have greater responsibilities get paid more, but nowhere near in proportion to how much their contribution and responsibility is greater.)
(1) There’s something to it. Also IT adoption is fairly low. Most Japanese offices haven’t changed that much and are fairly unproductive, but nobody seems interested in changing that. Candide has some insight on how fucked up the local IT industry is.
(2) Legally speaking Japan is hardly protectionist, but the industry is set up so that imports don’t sell well. Part of that is public policy part of that is the complex guild-like structures that dominate the economy and keep makework a thing.
(3) I live in a very, very small town, inside a very small province. Don’t get me wrong; young people are leaving, Centralization is a thing. But a small thing. 80% of young people stay, and they have plenty of jobs. Lots of industry, large and small. All Japanese industry is in the countryside, more or less. A large part of that is due to public policy and sheer pork; the Japanese politician class are mostly rural, to this day.
(4) I don’t think that’s it. Denmark has been prosperous for longer than Greece. Degeneracy is a matter of public policy too. Japanese don’t do drugs because they get life imprisonment. Single mothers don’t get welfare easily. Japanese law doesn’t subsidize degeneracy. I blame France for White savagery, in Asia it just doesn’t happen. Athrelon had a good article about how in Asia the lower class is well behaved.
(5) Conformity is certainly a thing. But I don’t see what those exceptional individuals in the US are doing that justifies coming apart in the whole West. As Sailer puts it most recent Silicon Valley innovations are just regulatory arbitrage (this is funny too). Asian regulators are not into arbitrage. They set up guilds decades ago and they enforce them.
(6) This is absolutely a thing. The Japanese are well aware that some workers are awesome and many are semi useless; yet everybody has to make a living. If anything very productive people are forced to work more than anyone else. They get paid more, but not vastly more. In the end the purpose of compensation is bragging rights. In Japan you get massive bragging rights with a $200k salary. Out of the chart bragging rights. You can look down on everybody. Do you really need a yacht?
So in summary:
1. The rent-seeking structure the state set up in the 1950s still remains in place. No regulatory arbitrage, no disruption. You can’t make rich by destroying a whole industry. They won’t let you.
2. Investor influence on management is nil. Management culture in Japan is collegial, they have their old culture of collective management and there are limits to executive pay. People are *very* sensitive to the difference between a $200k salary and a $190k salary.
3. The State is very invested in the rural areas and won’t let them fall. Money is running out, so they’re going to have to do something, but if Centralization is to happen they’ll take care it happens slowly.
4. No anarcho-tyranny. No drugs. No petty crime. Little immigration, and the few there is happens in Dubai-style semi-slavery terms. So the “flight to civilization” component of Centralization in the West doesn’t exist here.
Basically Japan (and I guess Germany) doesn’t like AIO. To the degree that the information economy makes AIO easy to happen, Japanese public policy buts effective barriers. You can’t stop it completely, but you can slow it fairly well. Seems to me that Anglosphere policy is actively accelerating AIO. And Russian or Chinese policy it’s pushing it even further.
Given the omnipresent status of the modern state, one can’t talk of the economy in a vacuum. Economic forces work inside the framework of state policy. Anglo state policy enables AIO, Asian state policy staves AIO, Third World state policy captures AIO. There’s nothing inevitable about it. There is no vacuum, there is always a choice.
You can learn a lot about an economy if you look at business software. How much it values productivity, and do people really work efficiently or not. Humans will make mistakes and sometimes use the wrong post code, etc. Ideally business/ERP/accounting software – the order processing, invoicing module – would allow you to preview and even print an invoice in order to review and amend it before you finally post it to the accounting books or finalize it, where it become unchangeable. Or it would allow a simple, one-click crediting the invoice and making a new, editable invoice copying all the data if you realize the mistake too late. Now if you look at Denmark, where 90% of the businesses use Navision / Dynamics-NAV, it is best summarized as “LOL WUT I don’t even.” Zero opportunity for human scrutiny after processing. No preview, you have no friggin’ idea how an invoice will look like before you actually post and finalize it, and one-click crediting was just introduced in 2015, before that, for 20 years, they had nothing. Do Scandinavians even wörk?
This may look like a tiny thing, but the sensibility of the software is a signal of how local employers regard employee productivity which should normally be correlated with GDP because what else. This is how I don’t understand why Denmark is (comparatively) rich.
I do not know if the Danes are actually that inefficient. But if they are, my thought would be that destruction is easier than construction. So it takes a lot of efficiency to make up for wrecking things through bad personal traits. If you don’t wreck things, you can be pretty prosperous without too much effort.
This is one reason why professional jobs in the 1950s and 60s were relatively calm. Women were happy working secretarial jobs, so the e-mail hell of today was handled by typists and secretaries handling memos. The memo volume was probably similar in terms of cognitive load to our current e-mail load, but now we expect everyone to handle an inane frenzy that keeps people paying attention to work after hours.
It’s also pretty common for professional women to either do nothing or to do worse than nothing by scheduling tons of expensive meetings. Then, the productive people have to work overtime uncompensated to make up for the slack in productivity.
We also have women being overhired in junior positions, overgroomed for promotion, who then tend to leave the workforce in enormous numbers at various biological breakpoint ages — which is the problem that “Lean In” was supposed to try to address. This makes the entire long term structure of the labor force seriously discombobulated.
We’ve also lost specialization in many office settings — now everyone is expected to type, whereas before, with even less efficient technology, we had specialized typists.
Handle’s corollary to Parkinson’s Law: Behaviors that tend to lengthen the time needed to finish a project increase to the productivity available.
I’m about a third of the way through the book and I don’t like it much. Cowen is too glib to my taste (I have no gift for glibness, so this might just be envy on my part). Content-wise, I feel he got his bearings wrong through an unfortunate focus on one example of man-machine collaboration — namely chess — which is very untypical for having a clear-cut, objective goal (winning or losing) and ways of telling whether you reached it or not, as well as taking place in an extremely circumscribed environment with no unquantifiable factors and “unknown unknowns”. Once these two restrictions are removed, enhanced computation speed stops being nearly as useful (as, indeed, Cowen notes even about chess programs in the opening when the programs are operated without opening databases) and tends to function as a mirror for the operator’s prejudices, like a modern haruspex.
To be fair, he says something very close to this in Part II, Chapter 6,
I don’t know where you read that. He did make remarks about the peculiarity of chess as a field of endeavor in Chapter 7, but his takeaway was nothing like what I wrote above, but rather that we’ll try to make the world more like chess — sliced and diced to be easy for software to work with. He makes a good observation that companies tend to offload some formerly clerical tasks on customers, lowering productivity. Also, I disagree about his take on humans “learning to override their intuition through use of/experience with computers”. This may partly stem from a conflation of “intuition” and “cognitive biases”, parallel and related to the conflation of “religion” and “superstition”. The chess players don’t “override their intuition”, they tune it in a different way. System II cognition is not really separable from System I cognition, the former is built out of and upon the latter.
ETA: Finished the book. The science chapter was rather bad, in my opinion, because Cowen’s philosophy is faulty. I have mentioned his misuse of “intuition” above. In the context of science, he uses “understanding” to mean something on the order of “corresponding to common sense derived from everyday experience”. In fact, it is quite possible and normal for one to pick up intuitions and quantum-mechanical common sense in the course of learning physics, and if this quantum-mechanical common sense does not jibe with everyday-experience common sense then so what*. One can still apply that common sense and those intuitions to solving problems, which is what counts. The same goes for more abstract mathematical structures. Cowen might have made a much better case for himself if he picked up on the problem of proof-by-computer (the map-coloring theorem is a well-known example) where a sizeable proportion of mathematicians feels doubtful whether this constitutes proof at all. If one takes away the “genius machines” and the pro-immigration stance, the last chapters are a passable, if relatively uncontroversial and not novel, argument in favor of the “Brazilification” version of American future.
* To pick up a mundane example, I think and believe that astronauts staying in space for long periods, e.g. at the ISS, quickly develop a separate set of common-sense intuitions related to their everyday experience, where water does not spill, objects don’t fall downwards etc. It would be interesting to know how quickly they adjust between these two sets.
Here is a related paper of his from last year: Economic Development in an “Average is Over” World
I don’t get it. What’s he saying, that the Third World doesn’t need to make stuff because Samsung sells cheap phones?
Well, cool, every single dude in Africa has a smartphone. He still doesn’t have a toilet, nor a proper house, nor a job. Isn’t that what development used to be about? Have they redefined that to mean having a facebook account?
Cowen has been pretty straightforward about this. He believes the world will be composed of a vast tannish underclass living in barrios with substandard infrastructure and shortages of anything that has some supply bottleneck. They will have access to cheap internet media and some trickle down mass produced goods, but won’t have great access to the things you mentioned. Above them will be the elite and perhaps 10-15% of people making up the professional/engineering class to run the robots and create high end goods.
He puts a positive spin on it, but he doesn’t lie about what he’s pushing for.
Why is deindustrialization seen as a point of success? Having a close loop — culturally, legally, physically, economically — between design and manufacturing is helpful to businesses rather than harmful.
Also even though the proportions of manufacturing employees go down, in absolute numbers, they are going up, even those employed effectively by the same company. “Radical insourcing” could only happen with more manufacturing expertise rather than less. If Boston Dynamics can’t even produce a useful robot, the imaginary academic manufacturing engineering community may not be as useful as the actual manufacturing community which is building expertise in China, Taiwan, etc. and not here.
The proposed radical automation stuff assumes that the useless prototypes routinely touted in magazines like Popular Science/Mechanics that never result in any usable products are going to suddenly become useful any day now.
He also misunderstands the nature of ‘Free’ software:
First, measured gdp won’t pick up free goods such as Facebook and Google.
Bzzzzzt. Neither of these things are free. Merchants pay for Facebook and Google. They’re the primary users: the other people are just along for the ride. And that activity is indeed picked up by GDP.
It also may not require much in the way of cultural changes or transformations, as most cultures in the developing world already are sympathetic to higher personal consumption. The almost obsessive pro-saving, pro-education ethics which evolved during the East Asian miracles need not be repeated for this growth path.
More clever-silliness. To consume, there must be savings.
Further, the digital economies are not actually all that digital. Amazon runs on UPS. UPS drives its trucks on publicly financed highways. The homes that it ships packages to benefit from advanced infrastructure and a network of blue collar labor that keeps those homes well maintained. Not even free apps can make all that much money from third world users because unlike in countries like China where there is some infrastructure, it’s awfully hard to ship packages to them. Extrapolating that economies can then be digital-first is profoundly confused.
The few countries that do manage something like this, like Estonia, have some unusual characteristics that aren’t shared by the Philippines or Central Asia. So even attempting to generalize a theory of economic development like this is almost a total waste of time, especially because individual political factors are going to be much more important than an attempt to understand which way the Hegelian groundhog is going to twitch his nose next year.
To his credit he makes some gestures towards this.
Most goods are still physical because humans live in the physical world. Even digital products tend to only be as valuable as the tangible objects that they describe.
So for example, even though Salesforce is a digital product, the services that it provides its customers are, at most at a few hops, about tracking the movements and ownership status of physical goods of some kind. Even if the Salesforce rep is selling software to the guy who sells software to the guy who is the regional manager for John Deere who ships John Deere tractors to the physical retailers, the original value down the chain came from the creation of that tractor. And the value of those digital transactions is effectively capped by the physical goods.
I guess people look at Nike and see that the manufacturing only cost $10 and the shipping costed $2.50 averaged out in bulk and that the pair of sneakers retails for $125, but that sorta ignores that there is no $125 sneaker without the physical sneakers, and you can’t build the cult of belief around the magic powers of those sneakers without the physical items themselves.
That’s a roundabout way of saying that even though non-manufacturing activities often make up most of the value of higher end products, that doesn’t mean that the manufacturing itself is of minor consequence. Especially for higher end products, total control over materials sourcing and manufacturing process is important. You can’t get that with enormous language and ocean barriers.
Shrugging your shoulders at the loss of American competitive advantage and saying “great, maybe the Kazakhs will do it for us, because reform here is unthinkable” is setting the country up to become another backwards polyglot republic at a time when it might have been possible to reverse direction from the 1970s.
The ‘why’ of deindustrialization also seems to handwave away the impacts of environmental regulation and other factors that actually motivate companies to offshore manufacturing. The reduction in industrial employment happens in high-regulation states, while the growth happens in low-regulation states. Rationalizing non-competitiveness as progress seems to be counterproductive.
It’s much more difficult to deliver toilets and proper houses than smartphones and second-hand clothes. The economic value-added of average people has been falling steadily towards negative territory, and as First-worlders aren’t allowed to get their kicks from mission civilisatrice and ruling anymore, there’s no incentive to deliver or maintain services in Third-world-like locations. And I’m not sure how much labor-related policies like Moldbug proposed in “Letter to France” and “Dire problem12</1>” would help, since demand by cognitive elite is limited.
The smartphone production is done in the third world basically by having very capable management who rules workers with an iron fist, maintaining military discipline.
No, no, no. Smartphone production isn’t done in the third world. It’s done in China. The Chinese are serious and disciplined, genetically so; it doesn’t take that much to put a Chinese to work, let alone Chinese women, who are the majority at Foxconn production lines. Foxconn bosses aren’t military types. They just pay well, and enforce lots of small petty rules to break resistance. It’s like a primary school more than like a military institution.
There’s no way in hell that smartphones or anything at that level of complexity could be built in India, the Middle East or Africa.
Content-wise, I feel he got his bearings wrong through an unfortunate focus on one example of man-machine collaboration — namely chess — which is very untypical for having a clear-cut, objective goal (winning or losing) and ways of telling whether you reached it or not, as well as taking place in an extremely circumscribed environment with no unquantifiable factors and “unknown unknowns”. Once these two restrictions are removed, enhanced computation speed stops being nearly as useful (as, indeed, Cowen notes even about chess programs in the opening when the programs are operated without opening databases) and tends to function as a mirror for the operator’s prejudices, like a modern haruspex.
That’s certain to happen if there is no reality feedback from competition. But if there is competition, then the haruspexes get outcompeted by those who are able to recognize when they should trust their judgment and when to defer to the machine. (The latter sort of ability is one of Cowen’s key points, which is in my view a very accurate insight.)
It’s similar to how, for example, statistical methods in the hands of academics usually become a tool for producing ideological propaganda and impressive-looking nonsense that’s good only for padding one’s publication resume. But this doesn’t mean that similar methods are worthless when used in industry, or that one could compete in these industries without using them.
Part II, Chapter 6: Why Intuition Isn’t Helping You Get a Job
You could rename this chapter “[some] Intuition is Overrated,” or even, “Intuition is Over.”
However, as with most things, I think it would be wise to split intuitive instincts into what you might call core social instincts. On the one hand, there are human evaluation instincts such as “woman’s intuition” in sensing underlying discrepancy between her counterparty’s presented self and his or her actual perceived self. On the other hand, there are experienced-based, pattern-perceiving intuitive heuristics. An examples of the latter could be the representativeness heuristic, a potential failure mode of which is said to be the ‘clustering illusion’.
Cowen starts with romance – “we take a lot of wrong turns in our pursuit of a good partner” – but that is of course a very tricky subject fraught with landmines, and so he’s wise to just leave it at that.
And yes, I understand this makes the point more salient to many people in his audience who adhere to a deluded understanding of the mate-choosing process and about the underlying mechanisms of various relationship failure modes. One can’t even get close to subject of discussing human sexual nature without touching the most sensitive and hysteria-inducing triggers.
But to slightly more enlightened readers, blaming modern relationship failures on rationality failures akin to the deviations of chess grandmasters from the best computer programs is a strikingly … misleading … or at least importantly incomplete, analogy.
The issue is tricky, of course, because of the definition of the ‘good’ in this context. That is, the unarticulated end or purpose for which these partners would ‘good’. Of course modern ‘romance’ is a kind of a hybrid or intermittent oscillation between forager and farmer modes of courtship, mating, and family formation, with a serious ‘time inconsistent preferences’ problem, to put it mildly, and one about which it is impossible to have a profitably frank discussion in public these days.
My point is simply that romantic intuitions are not like chess intuitions in ways too important to make this ‘good partner screening’ analogy work for me. And going further, in chess, the obvious goal is to win, or at least not lose. But people lack full conscious awareness of their romantic goals, and have horrible understandings of the situations that will make them happy romantically (e.g. the hundred bullet-point list of partner essentials, ‘you never know what you need until you get it’, and so forth.) Thankfully, most people end up settling into a new romantic equilibrium and forgetting all the nonsense they once ‘believed’ about what was indispensable. That’s a good and important function of ‘Schelling Amnesia’ – socially useful memory retconning.
But if you are trying to ‘help’ people by forcing them to collapse this vague and deluded cloud of sentiments and (mis)conceptions into some concrete set of goals and preferences, and then have a computer produce ‘advice’ as if these goals were stable and permanent things and it were really essential that they be met, then the computer (and by its influence, the person using the computer) are going to take seriously what ought not to be taken seriously.
You risk locking a person into exactly the kind of bad ideas and bad strategies that cause women to hold out too long for Mr. Right, while at the same time not actually ensuring that the priority of effort is placed on maximizing their ‘market value’, so to speak. Now that would be unwelcome but possibly still helpful advice, “the most important thing for you to do right now if you want to land and keep a good husband is lose weight and stop being such a nag,” but if Apple has got to make Siri give officially PC answers, then who is going to be able to capitalize on a RealTalk app?
Part of the old matchmaker system was a more wise and enlightened understanding of the nature of romance and what it takes to make happy marriages (in a supportive, reinforcing, all-encompassing social environment), and of course, a very liberal dose of paternalism. But computer-assisted hyper-individualism is as likely to lead men and women astray as they might soon think ‘faulty intuition’ does in the unaided context.
But Cowen says the matching software could nudge people away from these bad ‘hold out’ strategies too, thus helping them avoid the pitfalls of too much nudge-free, self-reliant individualism. And speaking of injecting a palatable dose of reality into mainstream discussion without triggering crimethink instincts, check this out:
But it is easy to recognize the value of having a matchmaker nudge the indecisive into, well, growing up. You can’t keep shopping forever, though humans have long seemed to want to do so. It is common to continue sampling profiles and dreaming on about a perfect mate but not actually dating anyone. Surely this is not the best use of technology. Perhaps most importantly now, dating algorithm technology can help us realize the errors in some of personally generated choices.
Yeah, but dating algorithm technology like Tinder can keep attractive women in their fertile prime on the carousel until theirs 30’s, constantly exposed to occasional addictively thrilling tastes of five minutes of alpha and bombarded with the lusty attention of thousands of potential suitors, turning the social-media false-consciousness of being an “ultra-hot mini-celebrity” knob all the way to 11, thus ruining them for mellowed stable marriages with ordinary loyal providers but with a comparative paucity of drama and passion.
A final note on modern romance. Who discriminates?
Years later, when doing the research for this book, I read that the scientists at Match.com have discovered that a conservative is, on average, more willing to email a profile listed as politically liberal than vice versa.
Sounds plausible, though I’d like to see the methodology.
II. Some Behavioral Economics / Psychology / Nudge Thoughts.
Biases, such as toward the familiar, are something we have long tried to overcome. In the growing field of behavioral economics, researchers measure the biases behind individual choices as judged by some external standard. We’ve learned – or we think we’ve learned – that individuals overestimate their degree of influence over events, and anchor too much on one piece of information when making decisions, among many other human errors and biases. The last time I looked, the list of cognitive biases on Wikipedia had forty-eight entries.
But even after all this work and all this evidence, nagging questions remain. When it comes to measuring a human bias, are we sure that the researcher is correct and that the individual choice is wrong? I have a lot of funny habits that I think serve me pretty well.
We see similar dilemmas in the more systematic literature. Remember that old rule of thumb “A bird in the hand is worth two in the bush”? Economists often consider it a “bias” that we value commodities we already own much more than commodities we might acquire (this bias is known as the “endowment effect”).
But for all its apparent irrationality, maybe this is an inescapable part of anyone’s ability to be loyal to friend and family. Maybe part of true loyalty is that we can’t always apply it selectively on a moment’s notice. In that case, these endowment effects may be an essential part of the good life rather than a signal of our irrationality. I’m not saying we know this for sure, only that the models we economists device don’t really settle it.
This is very reminiscent of the whole infamous “Fat man and the runaway trolley” problem that we are also told is evidence of an irrational bias. But insurmountable reluctance and hesitation to commit personal and lethal violence against some random innocent stranger in cold blood – and a potentially dangerous large male at that! – and even in such a ‘lesser evil’ circumstances, is probably also an indispensable ingredient in the social recipe of ‘the good life’.
And of course, as we’ve discussed in the game theory threads, many behaviors that may seem irrational from a certain, narrow perspective are in fact perfectly rational when one grasps the role they play in the big picture of managing human disputes.
When economists investigate human rationality, they are often too dependent on arbitrary stipulations about what is rational and what is not, expressed in the form of models. An economist must write down some mathematical axioms and then find that human behavior falls short of these axioms. But how convincing were the axioms in the first place for complex and multidimensional human problem solving? A lot of the research in this tradition isn’t convincing, no matter how brilliant the investigator. Other economists rely on artificially constructed laboratory investigations to try to measure human rationality or lack thereof. They use inexperienced undergraduate subjects, who are not always taking the problem-solving exercises seriously, and the prizes for good performance are relatively small.
One should also keep Szabo’s point about fallacies mistaking short for long games in mind when assessing these studies.
The nice thing about computer chess is that we have clear-cut standards, albeit not perfect standards, for good chess moves and bad chess moves. So, putting human play through the lens of the engines, what do we learn about human intuition?
III. Testing, understanding – and maybe one day exploiting (to whose benefit and detriment?) – patterns of human intuition by data-mining archives of human calculation decisions and comparing them to the computer’s answers. Today in chess, tomorrow … everything?
Ken Regan was a chess prodigy (like Cowen) who became an Oxford-minted Mathematician working on the infamous P-NP problem, but has in recent years been conducting some of the most rigorous investigations in the manner described above.
Some of those observations:
III.A. Humans are ‘contempt averse’ and try to avoid (if I’m interpreting this right) “unfamiliar complications without a perception of plausible paths to victory – or even out of deadlock – within their the horizon of their experience or ability to calculate.”
A player is least likely to make a major error when the game is tight, and if anything, players do their absolute best when they are faced with a slight disadvantage in their position.
III.C. There is not much evidence for a Nassim Taleb “Black Swan” model of cognitive failure.
Most games are decided on the basis of the accumulation of advantages, and the level of error is fairly well predicted by the relative skills of the players.
III.D. Players have consistent ‘styles’. For example, Kramnik is highly accurate ‘for a human’, but he doesn’t try to exploit psychological-based tactics against his opponents to throw them off their games by purposefully generating ‘unfamiliar complications’ where, in addition to the difficulty of the decision, their frustration, emotions, and ego will cause them to make more mistakes than usual. The polar opposite and currently most ‘nettlesome’ player is said to be prodigy Magnus Carlsen, of whom Cowen is something of a fan-boy (though in this hero worship of Carlsen he is hardly alone in the chess world, and this kind of hero-worship or real heroes is fine by me.)
The striking fact about chess is how hard it is for humans to play it well. The output from the programs shows that we are making mistakes on a very large number of moves. Ken’s measures show that even top grandmasters, except at the very peaks of their performances, are fortunate to match Rybka’s recommendations 55 prcent of the time.
Rajlich [Rybka’s creator] stresses that humans blunder constantly, that it is hard to be objective, hard to keep concentrating, and hard to calculate a large number of variations with exactness. he is not talking here about the club patzer but rather the top grandmasters: “I am surprised how far they are from perfection.” In earlier times these grandmasters had a kind of aura about them among the chess-viewing public, but in the days of the programs the top grandmasters now command less respect.
III.F. Humans also seem to prefer thinking about things – especially competitions – in “drama narratives aligning with the story patterns which most humans instinctively experience as particularly salient”. Or maybe a Sapir-Whorf-style constructed conceptualization in terms of the way they categorize (and mischaracterize) what is really ‘going on’ in certain situations. But we’ll revisit that a little later when I talk about Go. But for now, it’s important to note that this may give humans a kind of ‘Schelling Point’ advantage over computers in certain human-to-human match-ups if one’s intuitions align well with the opponents’ intuitions, which allows one to make essential – and more accurate – inferences regarding mental states and motives. This is something the current approaches to machine learning has a very hard time accomplishing up to now.
It might be possible to get programs to learn to be better players of chess or Go by playing other machines – or copies of their own software – over and over. Theoretically, it could use those ‘lessons’ to get ever better and faster at totally crushing humans in man-machine play too. But if you had a program play itself in poker some other simulation of a human competition and it might not actually learn anything very useful about how to get better at beating humans in person. It’s easy to get a machine to play the odds based on the cards on the table, it’s hard to get them to play the humans, based on human factors. So far, anyway.
All this talk about improving on the intuitions of experienced experts reminds me a little of the ‘Moneyball’ revolution, at least in baseball. Or maybe I should say “the story of the Moneyball revolution”, since there is still plenty of debate and controversy about the matter, which is further complicated by the fact that when everybody adopts Moneyball tactics in the arms-race for victory, then Moneyball stops looking as effective as it did when it was newly innovated and used by only one or a few teams as a temporary way to exploit an opportunity in the marketplace.
The “story of Moneyball” is that hero amateur enthusiast Bill James practically single-handedly realized that coaches and managers were making hiring, pay, and fielding decisions based on little more than gut hunches and some crude, individualized statistics (hits, base runs, RBI’s, etc.) that may reflect individual glory but didn’t add up synergistically to achieve the mission of maximizing team wins and championships. So he took a look at all the numbers in all the databases and tried to discover new metrics like ‘win shares’ which were much more statistically reliable and could guide these decisions. When Brad Pitt started actually dumping intuition-based decision-making and using these guides in a dynamic and continuously-refined manner his team suddenly started winning a lot more. Yay, score one for statistics and computers and boo human ‘expert’ intuition.
But in addition to arms-race complications (and, ahem, the steroids era, and also the explosion in player salaries which attracts much better athletic talent on average) there is still a lot of debate about whether all this was really any good. After all, one cannot escape the problem of ‘Garbage In, Garbage Out’, and there are plenty of important human things (like having the right personality to be a good team player who enhances morale) which can be observed and perceived by coaches, managers, and scouts, but which are never measured or captured in the game statistics and so won’t show up in the regression analysis.
Another example could be noticing that a player is one of those athletes who peaks too early and then will wear out fast and get injured too severely, too many times. Should you play him into the ground now, chewing him up and spitting him out, or maybe pace him so he’s more of a long-term investment and maybe stays healthy long enough to really hit his stride later on? Or maybe the guy is just a thug deep down and, whatever the numbers say, is destined to get into serious trouble eventually and in a way that will really hurt the team at a critical time. Worth taking that chance? Will the computer know how to look for thugs? Would it be legal to teach it to, even if it could? I anticipate that the law of ‘discriminatory algorithms’ and what must and must not be done about them is going to be a very hot area in the years to come. Someone should write a legal note on it now (besides Orin Kerr this time).
Anyway, a lot of the articles critical of Moneyball these days have as a thesis something along the lines of, “On second thought, and taking another look, and now that we know the whole human story which wasn’t reported at the time or known outside a small circle of insiders, it turns out that maybe human intuition about a lot of these things wasn’t so bad after all. And is still really important and, at least at present, not easily replaceable with more ‘data science’.”
Remember the example of the doctor who knows how to tell when his patient is lying or exaggerating or omitting important facts, vs. the computer who merely records and processes everything inputted on the form as if they were genuinely presented symptoms of an underlying disease.
And, it seems to me, such confusion and controversy is likely to remain the case in these very human instances with a huge number of dynamic and subtly interacting variables, low signal-to-noise measurement problems, and a huge amount of ‘causal density’. The big trick according to Cowen is how to have the machine and the man be good complements, focusing on their respective comparative advantages and correcting the weaknesses of their ‘teammate’. But teasing out which teammate is better at which function in these complicated circumstances won’t always be straightforward.
However, I think the economic incentive is big and permanent enough that eventually this ‘division of labor’ will be optimized one way or the other – even through just a lot of dumb trial and error – and mostly what the humans will be doing is ‘assessing the human factor’ and “being a pleasant (and effective / manipulative / motivating / inspiring) human interactor when a particular experience of human interaction is desired (thus more useful and profitable).” That’s taking ‘customer service’ and ‘customer experience’ to a whole new level. We are all Geishas now. Geishas to other Geishas. And at this point Cowen might dryly quip, “Those new service sector jobs.” But that probably is the future of labor, and, of course, the qualities needed to do these things really well are both rare and very hard to teach.
IV. Circling around to the point about ‘Nudge’
So what? Haven’t thousands of articles from psychology and behavioral economics outlined major weaknesses in human perception and decision-making abilities? There are the works of Daniel Kahneman, Dan Ariely, and many others. Haven’t we all heard about “nudge,” the concept so eloquently outlined by Cass Sunstein and Richard Thaler? In that worldview, experts know the biases of other decision makers and design the choice architecture to manipulate better human choices, such as changing the default options for which pension plan you will enroll in.
Yes, but the chess result differs. Computer chess is pointing out some imperfections in the world’s experts, or you might say it is pointing out imperfections in those who, in other contexts, might be nudgers themselves.
… It is precisely our reasoned, considered judgments that we should be more suspicious of.
Well, ok, but then why not let Rybka nudge us? And maybe not ‘us’ all at once with some one-size-fits-all society-wide ‘policy’, but individually, with some app that takes our specific, current circumstances and particular preferences and goals into account, and gives everyone the unique and personally-tailored nudge? Isn’t the idea to have the machine in the man-machine team tell the man what it ‘thinks’ the man ought to do? If a human chess expert/scholar like Ken Regan tells us that, if we want to maximize our odds of winning, we should do whatever some computer program tells us, or even ‘nudges’ us into following the computer’s ‘provably superior’ advice, then what’s wrong with that?
So, this needs a big bit of context that Cowen leaves out, which is the whole (largle Libertarian) counter-attack against the Nudge™ intellectual enterprise. Which follows the Cato Clever™ strategy. When the progressives start barking up some novel, dangerous tree, it’s tempting to take the bait and challenge their assumptions and worldview head on. But that never works, because challenging these assumptions is sure to get one tarred as either a bigot, moron, or an eye-rollingly naive geek who is totally out of touch with the latest science of human reality.
So the Cato Clever™ strategy is to posture as if one accepts all the false things which much be publicly accepted – and even embrace, celebrate, and cheerlead for them – but then to find and relentlessly pursue some oblique line of attack.
In the case of Nudge, the strategy path seems to be what one might call ‘Counter-Mandarinism’, or applying correctives to the naive progressive faith in ‘scientific’ experts to rationally guide the populace and the economy towards some kind of optimal equilibrium. The same general road was later taken in the field of ‘Public Choice’ (and now ‘behavioral public choice‘) and Academics, Experts, and Mandarins put in charge of studying, designing, implementing, and enforcing the new Nudge requirements are sure to have less than impressive results – certainly less than advertised – because of the inherent limits to any system of empowered government scientific experts tasked with management of human decisions and behavior.
V. “Signs of Improvement”
An example of ‘bad intuition’ is that, for what are well-understood psychological reasons, male chess players noticeably increase their risk-taking and aggressiveness when paired with an attractive female opponent, “though not in a manner that increases his chances of winning the game.” If they are relying too much on intuition without some
“coldly-calculating disinterested supervision” (such as that provided by an intelligent machine), then they may not consciously realize they are thinking with the wrong head.
And it seems, in part due to the increased competitive pressures and deeper insights brought about by the introduction of chess computers to the scene, that chess players are learning to avoid some bad moves that many players used to intuitively perceive as good moves, and are still steadily getting better, even while the edge of the best chess machines continues to grow. Maybe the best players today could have put off Deep Blue for another year or two, and that’s something, but not much in the big scheme of things. And here we are trying to grapple with that big scheme.
Old respected openings are now understood to be surprisingly vulnerable, and others are understood to have been underrated. “It’s revolutionized our understanding of the game. What else will machine intelligence revolutionize?”
Cowen devotes nearly a whole page on the progress of women playing chess (Polgars are so exceptional that they are in the “prove the rule” category), but this is not much worth discussing even in brief form unless you are already familiar with the real explanations of why fathers are more eager than ever to encourage their daughters to engage in competitive diversions.
VI. “The Future of Intuition”
Ken Regan ran an extremely deep analysis of a game between prodigy (they’re all prodigies, aren’t they?) Alexander Grischuk and Kramnik in 2007. Kramnick lost, but there was a point where the perfect move could have forced a draw.
At the cognitive level, this unexpected depth is also a disturbing result. It shows that we humans – even at the highest levers of intellect and competition – like to oversimplify matters. We boil things down to out “intuitions” too much. We like pat answers and we take too much care to avoid intellectual chaos. Even if you don’t think those flaws apply to everybody, they seem to apply to some of the most intelligent and analytic people in the human race, especially good chess players.
What does all this mean for our decisions, especially in the workplace?
1. Human strengths and weaknesses are surprisingly regular and predictable.
2. Be skeptical of the elegant and intuitive theory.
3. it’s harder to get outside your own head than you think.
4. Revel in messiness.
5. We can learn.
It is both scary and exciting. Human intuition is becoming radically aware of its own limitations.
Finally, I’ll add some personal speculations about the nature of these intuitions when it comes to certain competitive activities.
Like many of you, I’ve been following the news and conversations about AlphaGo and the games with world champions with interest. Or, I should say, I’ve been trying to understand the discourse surrounding these developments as best I can, not ever having played Go, and being hardly competent – not even at a dilettante level – to discern the merits of claims regarding cutting-edge machine learning.
Now, I understand that any activity accumulates jargon and terms of art and turns of phrase that cannot really convey their intended meanings to those who are inexperienced or uninitiated in the underlying experiences they seek to describe. Wittgenstein and all that; ok. If you’ve never surfed or skied, you can’t understand what the hell those people are talking about – except on some theoretical level perhaps – when they use some strange, often childish, terms to describe some kind of ‘disturbance in the force’ of some wave or slope that most experienced practitioners in those sports would recognize as generating a particular set of management problems for even very skilled athletes. If you’ve ever been a captive but ignorant audience to X-people talking about X at this high level and asking them to explain what they mean when they say ‘Y’, you’ll know that more often than not they pause in tongue-tied impotence and can’t figure out any good way to relay the experience.
What I’ve found interesting is that in these discussions about Go and Chess, for perfectly obvious reasons, the phrases used to describe the situations and moves take on a very martial character – ‘dominating that territory’ and so forth.
However, it turns out that many of these intuitive (and I’d guess instinctive) simplifying conceptualizations and heuristic-based judgments are not very accurate ways of summarizing the ‘real’ strengths and weaknesses, or stabilities and vulnerabilities, of the actual situation, at least from the ‘provably superior’ perspective of the victorious intelligent machine.
My guess is that humans like to simplify and generate these imperfect reductionist abstractions in particular and common ways, given the architecture of our brains and psychological tendencies with deep evolutionary roots. Humans like particular story arc patterns and arrangements of dramatic tensions, and they will probably interpret competitive situations instinctively in terms of the emotional framework of these built-in responses to stimuli that resemble these primordial competitive scenarios.
The imperfections and deviations of our intuitions from the ‘provably superior’ moves of the computer – and our over-and-under-estimations of the character of certain situations – might stem, in part, from wanting to cram the square pegs of the reality of these artificial games into the round holes of our instinctively preferred narratives, which probably did once fit the rounder, more natural competitions our ancestors frequently experienced. But then again, that’s just my guess as to one of the reasons why these deviations aren’t random.
My guess is that humans like to simplify and generate these imperfect reductionist abstractions in particular and common ways, given the architecture of our brains and psychological tendencies with deep evolutionary roots. Humans like particular story arc patterns and arrangements of dramatic tensions, and they will probably interpret competitive situations instinctively – maybe subconsciously – in terms of the emotional framework of these built-in responses to stimuli that resemble these primordial competitive scenarios.
Don’t know about chess, but with Go terms it’s a bit different. There are three factors: some terms are not ‘loaded’ (e.g. invasion, approach, pincer, points/territory) and are that way simply because we have to talk about things that happen on the board somehow. Other terms, like shape, influence, strength and weakness of stones, carry much more weight and are much vaguer — it’s a very common complaint from beginner to about candidate master level. Only very strong players really understand their meaning, and even then mostly implicitly, i.e. they cannot easily explain in words why a particular shape is good or why these stones are heavy, but only by demonstration of play sequences until their judgment becomes obvious to you (now that I think of it, it’s the same situation as with proofs in mathematics). And the third factor is that amateurs, or pros entertaining amateurs, necessarily fall into using the kind of easy-to-understand analogies you mention, or give up and discuss trivia.
To add to the board games angle, I remember reading some time ago about an estimate of how far are the best (human) professional Go players from perfect play. The idea was that the stronger a player, the more consistent his results are with his and his opponents’ ranks. The author graphed a measure of consistency against handicap stones, a common and ancient measure of relative strength, and the intercept corresponding to perfect play was at around +4 stones from the strongest Koreans. This is actually an almost unbelievably large distance, as the distance between 1p and 9p dan is usually estimated at around 2 stones. (Incidentally, Takagawa used to say sometime after WWII that he wanted 4 stones handicap to play with God.) Unfortunately I have forgotten where I read this, maybe it was in a tournament booklet and didn’t make it to the Internet.
Yeah, but dating algorithm technology like Tinder can keep attractive women in their fertile prime on the carousel until theirs 30’s, constantly exposed to occasional addictively thrilling tastes of five minutes of alpha and bombarded with the lusty attention of thousands of potential suitors, turning the social-media false-consciousness of ultra-hot mini-celebrity knob all the way to 11, thus ruining them for mellowed stable marriages with ordinary loyal providers but with a comparative paucity of drama and passion.
I think that the feedback from it is a long way from being theoretically processed, even on otherwise relatively analytical manosphere blogs, and it is as significant a development as the anonymity / alienation of the move to the city, away from the rural areas, was a couple of centuries ago. It obviously intersects with the thoughts on the “extreme lightness of [modern ‘progressive’] being” that you expressed recently:
The only sources of real animation or passion in their lives seems to be limited to work, entertainments such as sports and music, and various consumerist and socializing indulgences such as partying, going to restaurants and events … If you ask them about marriage, family and children, it’s not exactly a sore or touchy subject – though slightly more so for the edge-of-the-wall women – but their rationalized response is along the lines of “maybe, maybe not, who knows. I’m cool either way. If things feel right and it happens, great, otherwise, no big deal. It’s not like I’m particularly trying for some particular outcome, just enjoying life as it happens.” But those explanations and ‘justifications’ never seem heartfelt from either men or women …
To these people, it seems to me that the extreme lightness of their being (at least from my perspective) is indeed usually bearable. Actually, bearable is probably the low point of the spectrum, and the young ones with some status, money, prospects, and SMV seem to be genuinely enjoying a life barren of considerations of purposes and major life events that are utterly central to one like myself.”
For most people, until relatively recently, being single was a relatively undesirable state. For most normal, un-game-enlightened men, meeting people outside friendship groups was hard and carried a relatively high risk of rejection contra success rate (‘cold calling’ in a bar, etc). Obviously, with time, knowledge and practice rates could be improved, but many beta-males would much sooner avoid the hurt and settle into a committed relationship which guaranteed them access to sex without the hassle / humiliation / energy expenditure necessary to score new sexual partners, at least in so much as doing so required active effort.
The Internet changed things to an extent with sites like Adult Friend Finder, but openly using these was considered low-status and certainly weren’t the kind of thing you could respectably talk about at work. But the newer generation of apps (Tinder is essentially Grindr for heterosexuals), changed all this and made open discussion of their use commonplace, which in turn attracted many more users who would previously have been turned off by the low-status connotations of (openly) touting for sex on the Internet.
The overall effect of this has obviously been to make being single a much more attractive option to both men and women for a much longer duration than before, since even as your meatspace social groups start to shrink and atrophy as you get older, there is an endless source of strangers out there open to relatively low investment / commitment sexual encounters. The barriers to entry for men (humiliation / rejection) have also been greatly diminished as the initial contact takes place remotely, and for women (social stigma) have been removed because these encounters can take place outside the constraints of their normal social groups.
This in turn places much more pressure on people forming relationships in the first place, since it is always possible for either party to defect / default back to this relatively easy / attractive reset option. Knowledge of this also breeds distrust on both sides within long-term relationships, as the list of previous sexual partners on both sides grows exponentially longer with each hit of the reset option. I shudder to think what the long-term effects of these trends will be on family formation, or indeed preservation.
Cowen highlighted a paper demonstrating an argument that at least one aspect of automation may decrease with an increase in the minimum wage.
The idea is that lots of current automation is not really a replacement for low-skill humans, but a complement and augmenter of low-skill humans that allow them to do what was previously higher skill work.
So, as a hypothetical example, imagine that to make a really good latte, you used to need to hire a really skilled Barista, and the number of people who can do that really well is scarce, so they command a premium in pay over the minimum wage.
Now some company invents a machine that allows any random retail schlub earning minimum wage to make you an equivalent quality drink (or equivalent in utility when price-adjusted downward) at the push of a button and a few mindless “machine feeding” tasks, with the machine even guiding the worker in those dull tasks step by step to be nearly foolproof.
If the price of the machine is sufficiently low, the employer will replace the skilled human with the unskilled-human plus human-needing-machine combo, or “team”
But, if the minimum wage increases to that which the skilled Barista once commanded, then the team is more expensive that the skilled human, and there is no incentive to invest in or develop our intermediate-automation-case machine in the short term, which might stall out of the development of full-automation-cases later on if they must build on the foundation of the successful marketing of intermediate solutions.
Chapter 7: The new Office: Regulation, Stupid, and Frustrating
One of the biggest challenges in the new world of work is dealing with a world designed for the (relatively) smooth operation of machine intelligence.
At this introductory sentence, I was thinking something more along the lines of “staying out of the way of automated floor cleaners, and having to walk awkwardly in rooms and hallways with furniture styles and layouts and even ‘tracks’ that were designed for those machines instead of humans.” Like living in a ‘baby-proofed’ house, which is no longer quite like a house made to match the needs and wants of human adults. Or maybe having to deal with extremely risk-averse self-driving cars.
And yeah, that’s partially what Cowen is leading up to. It’s in contrast to merely putting up with badly designed human-interface or customer-service systems which don’t have much ‘machine intelligence’ at all, but instead suffer from a lack of it. And which may in fact be designed in a subtle effort that is trying to chase those expensive customers away instead of serving them, but with plausible deniability.
That is, the experience isn’t frustrating as an unfortunate but inherent consequence of using technology, because we’ve got to put up with machines which are frustrating because still too dumb.
Instead, in these instances, the (plausibly deniable) frustration is the point, as an instrumental device that organizations (or individuals through passive-aggressive behaviors) use to filter and nudge other humans away from making the most resource-intensive demands. The machine is merely the impersonal tool being used to deliver that intended frustration, so that some poor human doesn’t have to take any abuse or suffer an escalating emotional conflict. It also discourages those emotions in the first place, since everyone knows (or learns eventually after kicking an ATM a few times) that they are useless when deployed again dumb machines.
Now, humans have been deploying intentionally-frustrating interactive experiences at each other for a long, long time to accomplish the same goals for the same reasons. It just so happens we have these powerful tools to do this these days. And it’s socially convenient to have these tools be the scapegoated ‘face’ of the frustration and to provide us with a great excuse for it. As if it the frustration was due to dumb tech, which is forgivable, instead of a human’s intentional design, which is infuriating.
Is there a test for whether the frustration in an annoying experience is intentional or incidental? I think so. Since customers in genuinely competitive markets don’t like frustration, if a company can reduce it cheaply, then they will. I know I pay a little extra to my preferred car insurance company, because their customer service is consistently outstanding compared to the others I’ve known. Totally worth it.
But anyway, in trying to introduce his point, Cowen provides an example that goes in the second – bad interface – direction. Yes, I understand the point he’s trying to make here about the frustrations to come and their causes. And since they will be somewhat novel, the best way to explain them is to use some familiar, current frustration.
So it’s nitpicking to go after a tenuous analogy just trying to get this across to the average reader while delivering some bad news. And yeah, I’m spending too much time on a brief introductory portion. However …
I really, really didn’t like his example since it seemed to be of a category that has very little to do with ‘machine intelligence’ and much more to do with bad institutions, poorly managed organizations, refinement-paralyzing regulation and red tape, and sclerosis-inducing insulation from market incentives for quality service and efficiency.
And it’d be really unfortunate if high-status and influential academics seemed to be lending any credence at all to this corporate excuse in those cases when it’s a bunch of baloney.
Now, maybe one can say those human problems are now such durable or even expanding features of our social and cultural landscape that in any discussion of the future we should certainly try to anticipate what the typical experience of the combination of that organizational context with new automation solutions will feel like.
And yes, of course companies will try to use newly cheap technologies whenever possible in order to route customers, employees, etc. through the less costly, but dumb, computers first, to see if barely-adequate software can handle the issue or direct you to the right place instead of a comparably expensive and potentially trouble-making human. That’s nothing new.
However, maybe you’re like me and you’ve noticed that some of these solutions – especially online or via mobile apps – are well-designed, intuitive, efficient, fast, and in almost all respects so preferable to dealing with a human being that one would never go back unless forced to.
And maybe you’re also like me and you’ve also noticed that, in pretty much the same niches and information-exchange-and-processing spaces, you are occasionally forced to deal with truly awful automated systems that never seem to get better – or worst-of-all-worlds combination of bad humans using worse automated systems on your behalf – and you wonder why certain organizations can’t or won’t adopt better practices and solutions which are obviously readily available. Sometimes even the poor customer service agents on the receiving end of a tirade are just as equally clueless and exasperated as to why their company’s system is so much dumber than their own experiences with other, better companies. They are equally surprised, frustrated, and disappointed that they can’t use their system to do anything more for you and fix obvious issues.
So my ancillary point is that it’s possible to interpret (or spin) the accompaniment of “frustration and technology” as either “frustration because technology” (+economic incentives of course), or as a mere coincidence, with the frustration having a different root cause, with more human culpability attached.
Sometimes technology developed as well as is currently possible (given certain costs) will still be frustrating, because the state of the art just hasn’t advanced to the non-frustrating point yet. But sometimes the state of the art is perfectly fine, and has been for a long time, and through some combination of human and organizational incompetence and indifference, the ‘solution’ one gets is pathetically bad.
And thus my primary point is that maybe a prediction of lots of similar future “interactive technology proximate” frustration is perhaps not as indubitable a proposition as it seems, and it may be somewhat misleading to attribute that frustration to new tech instead of to old human foibles. A lot of frustrations I had in the past have already been remedied with IT, so it’s an ambiguous situation.
So it’s hard to estimate the future path of our ‘aggregate frustration’, since some formerly-frustrating things will get better (and many have improved dramatically in my lifetime) while other new frustrations emerge. Some current frustrations will turn out to be less inherent and more a product of the old-dog-new-tricks problem of stubborn human preferences for the way (older) people had already learned to perform some task. Old retired people in particular are much more patient than young people, have low time value, and try to squeeze some socialization time out of their commercial and transactional encounters. Young people often find that awkward and annoying.
So, younger humans might even prefer the ‘frustrating’ dumb-technology experience to the old human one. This will depend on what becomes socially normal, the kind of training and experiences they’ve had. I recently bypassed a line at a McDonald’s to use a big touch tablet instead, and for me, it was a superior experience in almost every way. My mother would be horrified at the prospect of having to use it.
It will also depend on the human capital of the people who end up allocated to certain service jobs in the future. Because systems using human employees can be just as frustrating too! And sometimes frustrating in the same way and for the same root reasons. And quality humans matter just as much to one’s experience as quality tech.
Frustratingly stupid technology is certainly no more frustrating that dealing with frustratingly stupid humans, or ones insulated from market pressures by captive audiences and job security and de facto government protection. Isn’t that why the DMV or Social Security Office experience is such a common joke? In fact, some commentators have started pushing back against these jokes by noting mild improvements in some local DMV experiences in the last decade. But of course, most of these improvements and lowerings of frustration are precisely due to the introduction of “machine intelligence” technologies, which does goes against the grain of Cowen’s point.
So again, whether or not we should expect more or less frustrating interactive experiences, whether we’re dealing with humans, technology, or some combination, seems to be less a function of the emergence of still-mostly-dumb machine intelligence, than a function of the usual social factors that undermine the incentives for humans and organizations to deliver pleasant experiences to the users.
It’s probably reasonable to expect the state of the art in machine intelligence to keep making progress and to keep getting cheaper, and so the capability-imposed ceiling probably won’t be what’s responsible for frustrating interactive experiences in the medium-to-long term.
Instead, the cause of our frustrations will be the same factors that made it annoying to deal with some Roman or Chinese bureaucracy two thousand years ago.
Cowen’s illustration is a story about trying to call Cox’s help-line to get a storm-damaged cable rehung. He gets in the typical annoying phone tree designed for routine calls and tasks – “to check your balance, press 1, to pay your bill, press 2 …” – and ends up having to ask for human help, which he gets and which solves his problem. He was annoyed at having to repeatedly give information they should already have, like his phone number even though calling from a Cox phone line.
But in my own experience, companies have been recognizing me by caller-id for about a decade, and the copyright on AIO is 2013. And the fact he chose a Cox experience provides a hint to what may really be going on. I don’t think I need to point out that cable companies are infamous for poor customer service, tend to be monopoly-ish, and that this is not a coincidence. I think Cowen would argue for the government leaving Cox alone, so it’s at least a little bit ironic that he would pick this particular frustration – something potentially derived from Cox’s monopoly-ish position – to provide as an example of a more general phenomenon.
Now, was the annoyance of this experience really due to machine intelligence? Again, you can try to spin it that way, but I say no. First, phone-tree machine intelligence now has redundancies in website and mobile app options. If I’m calling anyone on a phone these days, it’s almost by definition because I can’t do what I want without a human, because if I could, then I would have been able to do it online or with my smartphone, and that’s not just cheaper for the company, it’s easier, more convenient, and more pleasant for me.
So, if I’m calling in, I already know my solution is not on the damn phone tree, and I start pressing 0 or yelling “operator!” or “help!” And in the arms race of telephone customer service, that is often not working quickly anymore, so they throw some extra delays and irrelevant options at me.
And that’s because nearly everyone is hoping to get a hold of a human right away too, whether they’ve exhausted internet options or not, and even though they ‘shouldn’t’ need a human in the eyes of the company. For people like me, there will be no alternative or cheaper way to handle my issue than by giving me a capable and authorized human to deal with. Well, no cheaper alternative except hoping that I’ll give up, which I’m sure a lot of even internet-savvy people do.
As an aside, this is apparently also the ‘Kafka’s Gatekeepers” approach to American medicine these days, which may even be for the best, given that many people will eventually get better on their own and have better outcomes without early medical intervention. One hopes that these selective gauntlets are ‘choice architecture’ nudges set up with the thought that the people without real problems will get nudged out of the system early, while the people with real problems with pay a small frustration cost but will keep persisting, which means that eventually they’ll get the care they need.
I worry that neither of these is true. That is, that, (a) this model isn’t true in fact, that stubborn persistence and genuine illness are not strongly correlated, and that there are too many persistent hypochondriacs who suck up resources, and shrinking violet sick who just get sicker and end up costing a lot more later on, or maybe dying. And (b) that the frustration filter wasn’t even set up with any evidence that this model would be true, and instead is just a crude way to try and cut volume and costs.
But see, once again, we’re talking about a human, social problem as the source of frustration. Not a technological one. Some people really do still put up with a call to customer service for ordinary tasks. And right now, systems are designed in a ‘dumb’ one-size-fits-all manner that has to route every person through some master flow chart to figure out what kind of user you are and what you want to do.
Emerging IT is like an growing ladder that extends our grasp of economic fruit higher up the tree. And I’ve said before elsewhere that I think that radical personalization might open up a lot of new, meaningful opportunities for real progress by eliminating some of the one-size-fits-all / lowest-common-denominator inefficiencies that currently plague our lives.
So maybe in the future the cable company or the bank already knows who I am, what kind of ‘customer service user’ I am, and I can ‘earn’ and establish credibility over time as someone who is a ‘power user’ of their internet options, and who should therefore just get routed immediately to a human being if I’m calling in.
A boy can dream. Maybe instead of a credit score I’ll have a customer score that companies can share, as they exchange Yelp-like reviews and notes about my tirades. Maybe high ‘customer scores’ get you ‘concierge service’. That’s something you or your company can pay for, but one can also ‘pay’ for it in terms of an established record.
Actually, I have no way of knowing things like this aren’t already happening. I think some companies are already doing this quietly, as I compare my experiences with those of others and I’m already starting to notice odd differences.
Maybe this kind of comparison will become impolite, like discovering price discrimination by asking the person next to you on the plane what they paid for their ticket. ‘Service discrimination’ will also be something you keep to yourself, while you are always wondering if there is some secret higher tier or ‘lifehack’ way to manipulate and shape one’s ‘presented customer profile’ to get better perks, like ‘comps’ at a Casino.
For air travel, an analogy could be that you know you can always pay for first class, but you’d like to get comped to business class by being a good, loyal, low maintenance customer.
Maybe we’ll have a layer of meta-frustration: frustrated at the system that handles service discrimination, just like air travelers are irritated by the increasingly complicated welter of the boarding process.
If I can impose frustration on you at the right time, once you’re already committed and the cost of exit is high, maybe I can extract a rent from you at the last, impulsive minute in exchange for reducing it. Like overpriced concessions at the ballpark or theme park. Or like a tempting upgrade in boarding tier and seat class. (Though if this is expected, solve for equilibrium.) But in which case, once again, the frustration is both intentional and arguably unnecessary (in the sense that technology could do no better). Instead of being due to the current technological limitations, it will in fact be due to technological capabilities having passed some threshold which enabled such scientifically calculated frustration imposition. This will be in combination to whatever truth there is behind Scott Adams’ Confusopoly and Akerlof and Shiller’s Phishing for Phools.
3. Cowen notes that the future of retail involved the customer replacing a worker and being required to “pitch in” and contribute some work with their whole transactional process, couples with machine intelligence. Many of us already have the experience of pitching in with bagging, boxing, or warehouse retrieval when shopping at Aldi, Costco, or Ikea. At the DC Ikea, the path to the warehouse leads one to a terminal where one can search for parts and assembly kits. Fortunately the software is smart enough to not be Scandinavian special-accents-and-letters-sensitive, so you can just type “bork-bork-desk” on an English keyboard without the umlauts and empty set symbols or whatnot. And of course in online shopping, the customer is putting in some work in partnership with the search function to browse and learn about various options, so no salesperson has to bother with any of that.
This will make things cheaper, which is hard to notice. But it will also feel like annoying work, which is a cost which will “hit you in the face” as the practice spreads into new domains.
4. He goes into a short rant against the frustrating limitations GPS systems. Well, first, does anyone even remember how frustrating paper-maps-and-directions road navigation used to be? I remember what it was like to drive around, at night and/or in weather, in a new town, trying to keep a reasonable speed, while squinting to read the print on the road signs from far enough away that I could signal in time and brake slowly enough to not cause an accident. The truth is, people missed their turns a lot in these circumstances, and circling back, sometimes several times, was a fact of life.
And second, I think he was right in 2013, but wrong in 2017. I use GPS every day, and in my personal experience, it’s gotten a lot better at fixing all kinds of minor issues and in just a few years. That remarkable pace of real improvement I think makes my point about technology in working markets probably being more of a reducer of old frustration rather than a producer of anything but temporary new ones.
In short, when GPS fails, the problem is that human beings have not sufficiently reduced the complexity of the surrounding environment.
But then GPS got a lot better when nothing got simpler with the environment. Instead, it seems that Google in particular was able to observe, with a high degree of precision and granularity, countless humans traversing the same areas and interacting with other vehicles, and to statistically process that growing database to learn not only what ‘real’ paths they were taking, but also, by means of experimenting with many different approaches to directing GPS-using drivers, to learn the optimal way to communicate that reality to the modal driver.
Indeed, here are humans ‘pitching in’ and being unwittingly recruited to complementarily partner with ‘dumb’ machine intelligence and temporarily perform the “navigating messy, complex real world environments” function. At least, long enough for the machines to watch it happen enough times to create some indistinguishable (in the narrow context) imitation of how a genuinely intelligent creature would respond to certain combinations of stimuli. I expect to see a lot more of this kind of thing. Is this how Deep Mind learned to play Go, in part by ‘watching and learning from’ the combined history of all those human-v-human games?
That raises a question about which I can’t even really speculate, being quite out of my depth regarding these subjects. However, if we are teaching machines to be proficient at certain human things in large part by watching and, I guess, ‘distilling the essence’ of how humans are doing it, then it seems like we can make machines act like super-humans in certain contexts. I mean, be in relation to the best humans as the best humans are to second-best humans. Better, but perhaps not different. But does that mean that we are ignoring or have no idea how to effectively explore large areas of potential ‘intelligence space’ that is profoundly different from the way humans would approach certain problems? Will we be discovering human local optimal and missing global optima? Does it matter?
5. He uses another typical example of the self-service grocery store checkout, another example of being asked to ‘pitch in’. Do you remember when you first saw one of these how you thought they would be cool, only to realize on first use how crappy an experience it is? For me, it hasn’t gotten much better in years, which I’m guessing is because cashier tasks implicate Moravec’s paradox and a version of the Pareto principle of diminishing returns that the last 10% of progress will take 90% of total effort.
Cowen says he uses them, but his shopping expeditions have changed as a result, “refuses to buy anything that has to be weighed, named, or otherwise evaluated on a discretionary basis.”
Costco just started experimenting with self-checkout. But in my experience, their cashiers and baggers are much faster and consistently reliable. But nothing is weighed or evaluated. Everything has a prominent bar code. You don’t buy an apple, you buy a flat of a dozen apples, and while the flats might vary in weight, it’s not by enough to matter, so they all cost the same.
Years ago I saw an ‘awesome future’ commercial showing a shopping cart full of merchandise in which each item must have had some kind of RFID tag. The shopper pushed the cart through some magentometer-looking arch and all the items rang up instantly. Apparently we’re not there yet.
I avoid self-checkout unless the extra time it takes to get a human cashier is really severe. When I use them, I’ve learned to just abandon certain items in my basket that have a high likelihood of causing a machine error. If I get an error and no “machine error servicing worker” is able to come immediately, then I usually put up with it, but on occasion I’ll just rage-quit and walk out. The problem is that while the machine is recruiting me to pick up items and scan them, it’s not recruiting me to resolve issues when they occur.
That’s because it doesn’t trust me. It shouldn’t trust me, that’s the right call, since a reliably trusting machine would get played even more than a naively trusting person that never wises up. Margins are tights, and customers will squeeze you a few percent without thinking they’re doing anything seriously wrong, while eliminating your profit. But if the store only wants machines trusting store employees in the event of any ambiguity, then that cuts deep into any potential labor and time saving. And so every trip is a gamble between a predictable wait for a cashier, or a usually faster, but occasionally much slower, self-checkout.
In previous conversations I said that ‘trust issues’ could exacerbate centralization because even when certain tasks need not necessarily be done in close proximity, often times the extra cost would be worth bearing so that sensitive information stays in house among people with longer-term relationships, commitments, mutual dependencies, and under the same jurisdiction for civil and criminal process. My bottom line point is that similar trust issues are going to make some tasks hard to fully distribute to the customer too.
6. Switching gears, Cowen starts a section on Machine Assessments. “Workers will be increasingly tagged with their strengths and weaknesses, expressed in terms of numbers.”
This is metrics working their way up the work hierarchy. Deliverymen have been measured on speed, traffic tickets, accidents, and complaints, and fruit pickers are measured on bushels per hour, and that’s old news.
But Cowen is talking about professors, lawyers, doctors, etc.
No Child Left Behind was 15 years ago, which really kicked off the nationwide fad to try and measure teacher performance. It hasn’t worked out so well, but of course education is a problematic field for the typical study approaches, for null hypothesis reasons.
Overall it’s a big GIGO problem. If we measure academics by number of published papers, we’re going to get a lot of worthless papers. We are still debating whether “Moneyball” really works.
We have metrics at my workplace. Senior managers are obsessed with metrics for performance evaluation, which is what they’re told to be. I’m sure there’s a meta-metric for them that takes into account how many metrics they have developed, how many they are using, and how well they are using them.
But the metrics just aren’t that good as soon as the work gets complicated.
As an example, one workplace I know measures certain expert reviewers by clearance time. But there is an obvious trade-off between speed and the quality of scrutiny, just as with any quality control. And quality of scrutiny is hard to measure, which is precisely why expert reviewers are needed in the first place. So upon implementation of the metric, reviewers start gaming it immediately, and respond with “clear with no edits”, whereas in the past they would have taken a few days to work some of their contacts in the expert community to verify whether one word should be changed to achieve near-perfection. There is also the issue of fairness, as the experts will all be treated as a group of peer competitors, but some of their topics inherently take longer to process than others.
I’m not saying that these metrics can’t be refined to be both more fair and more accurate over time. However, there is little motivation and attention put on that task once the metric is established. And furthermore, there are just enough subjective assessments and judgment calls in the performance reviews that somehow, at the end of the day and averaging over several cycles, people seem to get ranked more or less according to their marginal productivity. And that seems to be despite the metrics, not because of them.
Which I think reveals what is really going on. Most poorly performing people will have at least one metric give them one black spot at every review. Not that high performing people don’t occasionally get a black spot, but no one seems to care much about that, and their managers understand it’s a kind of aberration or ‘system bug’ that can be ignored. But with poor performers, one begins to accumulate the all-important paper trail for Human Resources and the Lawyers.
That being said, maybe the “gimmick metric era” will fade away with the magic of big data and the capacity of machine intelligence to pick up the things that the good human managers were picking up, but without many of their biases. It seems we really can use the best game-playing software to also evaluate the strength of human players of Chess and Go, so how long before those techniques are extended?
We can expect to have this practice spread more widely. The next step is to hire individuals to work with genius machines to assess the performance of workers, most of all skilled professionals. I mean the people we depend on, like doctors, lawyers, professors, and our coworkers too.
The results might not be very politically correct. In two ways. The first is through group statistical disparities, and the second is by more definitively demonstrating Hansonian medicine or Klingian education.
Sooner or later, most professionals, especially at the top end of the market, will be graded by teams of skilled workers cooperating with smart machines. Think of this as more scientific Yelp rating for almost everything, just as we now have such ratings for restaurants.
It had better be ‘more scientific’, because using Yelp reviews to find quality restaurants is very unreliable, as Cowen has himself complained. Of course, using Cowen’s reviews is also, in my experience, unreliable.
Let’s say it is a lawyer. Potential customers can ask their smart phones where the lawyer went to school, what her class rank was, and what kinds of promotions she has received. That information will be accompanies by an asterisk: ‘This information explains only 27 percent of lawyer performance.’
How confident should we be that ordinary clients will be able to process this information appropriately? And anyway, the basic rule of “you get what you pay for” will tend on average to turn price into the best metric available.
Will professionals be able to opt out of this system? Not really.
Many of the lesser lawyers will decline to be rated .. That will hurt their business prospects … Have you ever opened up the Friday movie page and read, ‘The studio declined to make this movie available for screening at press time’? The obvious conclusion is that the film is a dud, and indeed it usually it. … sooner or later most professionals will have to submit to ratings, one way or another.
Of course, currently many professionals are rated by means of subjective online reviews (e.g. rate my professor), and just like Yelp, those reviews tend to be very unreliable except as popularity contests, and I haven’t noticed any improvement over time. That’s in contrast to crowd-sourced product reviews on Amazon, which I think have gotten better and more reliable over time, especially given tools like fakespot.
Cowen says that professionals will compete more intensely with each other and prices will more closely match marginal productivity. That’s going to cause some social problems.
Naturally this is going to both demonstrate some unpalatable facts about the distribution of and persistent gaps in human performance levels, and it’s going to be harder for the political process and social pressure to both put a thumb on the scale to achieve certain numerical targets, while simultaneously denying the true extent of the intervention.
But also, as any union foreman or squad leader will tell you, putting too much of a spotlight on individual performance can be corrosive of team spirit and camaraderie and leads to all sorts of contempt and resentments. In many situations there’s a delicate psychological balance between the justice of giving every man his proper due and the pretense of treating peers as ‘equal’ partners and dissolving a lot of differences in ‘luck’. And managers can only do so much to foster this right balance. The team members must all play their parts joining in the pretense as well. So the best performer can be admired and rewarded, but perhaps the situation calls for the right touch of false modesty and humility, “Aw shucks, I just got lucky or had a good day, and there’s no I in TEAM,” and so forth.
Some professionals will rise to this challenge but many others will be demoralized. Just as chess grandmasters no longer seem so wise and omniscient in light of computer analysis, so will doctors, lawyers, and teachers lose a good deal of their aura. … We will end up with professionals who are less sanctimonious and less arrogant.
One can hope. And this seems to be one of those times in which Cowen is subtle trying to inject a little sense into current thinking on these topics while slipping it gently past the resistance of the usual shields and mental blocks.
Then again, sometimes some people really do need to follow expert advice, and that mystical aura – however disproportionate or unsubstantiated in reality – acts like “legitimacy” and does come in handy for producing the psychological reactions that make these people more likely to willingly obey and comply. Some of you are parents – its own kind of higher-status ‘professional’ role in comparison to young children – and you probably know what I’m talking about in terms of your own experiences. It’s a double-edged sword, and we’ll miss the good edge when it’s gone.
It’s going to be a very different world when consumers feel so much on top, and in some ways it will be more dangerous because consumers do not always know what they are doing. … It will be harder for doctors and lawyers to “nudge” us and control us, because we will become more used to evaluating them, standing above them, and applying the programs to them in a manner that will make them feel small and will make many of us feel more powerful.
7. The Flipside of Ratings
I’ve spoken of patients ratings doctors, but doctors will rate patients too, especially if the doctor’s own rating depends on how well a patient follows a prescribed regimen of treatment. How many doctors will want to treat the patients who do not follow up on instructions and take their medications? In the future we might see doctors turning away those patients, or charging them higher prices, or putting them last in the queue for a visit.
You’ve probably heard of FICO scores, which serve as credit ratings in the United States. The company that created FICO scores … is now working on created a Medication Adherence Score …
Doctors will send away from of the very worst patients, in part to avoid wasting their time, in part to avoid the feeling of failure, and in part to protect their own performance ratings
Well, while I can only hope that my track record demonstrating I’m a ‘good patient’ will get me quick, friendly service anywhere I go, compared to those ‘bad patients’ who get turned away or suffer the stink eye, I have my doubts. Maybe I’ve become cynical about these things.
I suspect that to the extent this kind of thing can reduce my privacy and make my medical experiences even more annoying, it is already happening. And to the extent any of this could improve anything, it’s probably already illegal, or would quickly become so as soon as it started to ‘bite’ just a few typical sob stories. “Why don’t I inject my insulin according to the instructions? Well, it all started in my childhood … ”
How about a “consumer difficult quotient”? … Customers will have to live with the reputations they create for themselves.
The digital panopticon will mean we’ll all be living a neo-village kinds of existence, with very little effective privacy and the social / ‘permanent digital record’ equivalent of the All-Seeing God watching your every move. Except the belief in the existence of this God doesn’t depend on faith. Who knows, maybe the equivalent of village-life norms full of ‘God-fearing’ people will reassert themselves as a new social equilibrium emerges in reaction to these new features of life. Again, a double-edged sword, but in terms of social and cultural capital, we could probably use extra incentives to help lots of people stay on the straight and narrow.
On the other hand, I’ve remarked several times that our culture seems to be producing new effective norms of ‘digital discretion and prudence’ very slowly – or not really at all – even in response to plenty of prominent media reports of people getting caught. The instinctive impulses that create ‘digital panopticon complacency’ (or temporary absent-mindedness, blindness, etc.) and cause people to lower their guard seem very easy to trick into a false sense of privacy, and very hard to discipline. Something about the way people interface with the internet or social media seems to evoke a sense of being free from all the human and electronic prying eyes. I mean, even the former director of the Defense Intelligence Agency apparently forgot that he was subject to surveillance!
Maybe in the future, instead of just burying it in the disclaimers in some EULA, these interfaces will have to constantly warn people over and over that everything they do is subject to being recorded and shared, just like some telephone customer service lines have to warn you that the conversation is being recorded to avoid liability under the wiretap laws. Maybe the “webcam light” should always stay on, whether the camera works or not, just to keep reminding people of this problem. Maybe we’ll need daily prayers (or nudges) to remind ourselves, as a “memento memori”.
One way of looking at this is Brin-type of Transparent Society is also a kind of ‘pan-underwriting’ society, in which every application, request, attempt to match, etc. in which one tries to make the best impression is complemented (or perhaps ‘kept honest’) by the notion that a thorough – and now instantaneous – background investigation will be performed to verify and assess all claims.
Then again, loan underwriting – and the prudent principles upon which it is based – has been around forever, and that didn’t stop Countrywide from handing out loans to all comers with the bare minimum of scrutiny like candy at Halloween. “You don’t look like a little kid, and you’re not wearing a costume, but, whatever, here’s your ARM.”
8. Our (Legitimate) Fears
For lack of a better way to put it, what if too much accuracy and truth about human measurement is harmful to motivation or bound to lead to sub-optimal reactions? Maybe there’s something to ‘growth mindset’ and there’s such a thing as nobly lying about ones odds in order to boost optimism, confidence, attitude, and motivation to optimal levels? Maybe there’s a sweet spot, and you can overdo it in either direction. Maybe that infamous Dunning-Kruger chart tells us something not only about delusional self-regard, but the necessary amount of delusional self-regard and uncertainty to bring about the optimal levels of effort, so that potentially productive people neither get too demoralized nor too full of themselves, nor too gleeful or disappointed if things don’t go as expected.
On the negative side, too much knowledge can hinder achievement. What if a computer had graded Einstein at five years of age when he still was not talking and assessed his change of becoming a great scientist? That truth, albeit some imperfect statistical estimate of the truth, will discourage too many people. Alternatively, maybe a future star will be done in because he is repeatedly told, from day one, that he is the anointed one. [Wasn’t that the plot to Gattaca? -[b]H[/b]] A certain amount of ambiguity may be good for the career ambitions of young people, and in the future we may miss some of the ambiguity we enjoy today.
… In other words, however useful the concept of standardization may be in the workplace, it can be scary when applied to social and economic relations as a whole. This is not a world where everyone is going to feel comfortable.
The discomfort with machines will expand along a number of dimensions. The most skilled man-machine teams will earn a lot, but there will be an issue of societal trust, precisely because their mastery of external environments may outstrip our ability to judge them.
We already can’t judge the performance of good algorithms without … trusting the judgment of special algorithm-judging algorithms or meta-algorithms. One of you Latinophones can make the appropriate modification of “Quis custodiet ipsos custodes?”
He does have a paragraph that again suggests people will try evade personal responsibility and launder socially-undesirable but true / necessary decisions by attributing the culpability to algorithms.
That will reduce accountability both for ugly-but-right decisions, but also for questionable-but-wrong mistakes. Your perspective on that probably depends on your opinion of our current accountability systems. For example, if you think the current liability system is borderline insane, pins far too much of the blame for incidents on the wrong parties, penalizes them in complete disproportion to the harm or degree of hazard, and leads to awful risk-averse / zero-tolerance reactions on the part of deep-pocketed decision-makers who are terrified about the prospect of a lawsuit, then maybe you think that making an algorithm the ‘face’ of responsibility would be an improvement if it disrupts the imposition of crazy judgments on any particular party. Then again, I spent far too much time above writing about the pitfalls of the ‘algorithm as whipping boy’ approach to accountability to trust that will actually happen.
In other contexts we will not be able to avoid making some very direct moral judgments. Let’s say, for instance that you own a driverless car. How should such a car be programmed in case an accident in imminent. Should the car swerve away from hitting a baby carriage, but at risk of running into two elderly people?
Oh Lord. It’s Trolley Problems all the way down. Can the answer be other than what is expected to be cheapest in a lawsuit? If that’s not the answer, the law is wrong. And the law says that the law can’t be wrong. QED.
Also, what about your poor passengers? Pro-tip: To weigh the utilitarian moral calculus in your favor – so your car doesn’t decide to murder you by veering into a wall instead of hitting a group of thugs or carjackers blocking a road – always be sure to carpool and fill up the seats. That way, the Rawlsian logic circuits will decide to just mow those people down. Uh oh, now people are going to hack their own driverless cars to convince the machines that they are full of people. Uh oh, now google will try to measure the number of smartphones in the car to verify. Uh oh, now people will just fill up their cars with extra smartphones … I can do this all day.
Should the car be programmed to crash you into a telephone pole, rather than run a p=0.6 chance of knocking over a pedestrian? It is easy enough to imagine such issues being debated on the evening news.
I don’t imagine that public debate will let the programmed cars behave very “selfishly.” But what if a human driver takes over operation of the car right before the crash occurs? Currently the legal system allows a fair amount of leeway to human judgment. Might our legal standards for human drivers toughen up too? Will the moral calculus of your “driving program” be admissible evidence for your reckless-driving hearing? Probably so. You can see that a lot of everyday morality is going to change.
I don’t know. Humans might be reluctant to shift norms in a direction that has a good chance of biting themselves in the asses later on. The leeway we give other people is often the leeway we hope to get ourselves. It might just be the case that human liability stays the same, but machines are held to a higher standard. Or maybe humans will fall so far behind the machines so quickly that human drivers will be outlawed faster than we think. It’ll be like Pinker’s Better Angels, and future generations will think the level of harm we blithely tolerated for the sake of having our human-driven automobiles is practically inconceivable.
I think my TLDR on it is that capitalism works, but state-micromanaged capitalism is pretty crappy. The awful experiences you described earlier (bank, doctor) were both interactions with two of the most regulated industries there are. You mention GPS and yes, it’s pretty great now. I just drove to the vehicle emissions inspection station having google direct me via voice on my phone. No need to look.
There certainly is the question about exactly why interacting with uncompetitive corps is typically vexing. I expect there are lots of explanations, but to some degree they all boil down to one thing: providing anything — a good customer experience included — is not easy to do well, and unless there is some pressure coming down from on high, in general it won’t be done well.
In particular, I think that strategy of “frustrate a paying customer so much he gives up” is something that basically will never exist in a free market. Indeed even the suspicion of such a thing strikes me as a nearly infallible flag of state intervention of some kind. Paying customers are gold; you piss them off at your profit’s peril.
Some specific comments:
I don’t generally have trouble with self-checkout at grocery stores. I do buy produce. Yes, it took a little effort to learn how to do that; it’s not just “wave the bar code around until the machine beeps”. But I did learn. It is indeed more work, but you get something for it: shorter waiting time. And in any case the time I used to spend standing in grocery store lines was not really quality time for me. Certainly I’d enjoy it more now with a smartphone than I used to with … not much, people magazine? But “losing” time that I wasn’t really using well to begin with, presumably for slightly lower prices and slightly less time in line seems reasonable to me.
To me, the biggest problem with supermarket checkout is getting clueless people in front of me. Variance is higher.
One more thing is this vein is even more automation of stores. If you haven’t already, check out Amazon Go. Amazon Go is a brick and mortar store with no checkout. It doesn’t really exist yet save for a test store in Seattle which apparently serves only Amazon employees. Still, they are almost there; they claim to want to open to the public this year. Assuming they prevent the entry of random bezonians, then they can probably afford a significant proportion of unintentional “shoplifting”. Customer discrimination FTW.
Metrics: fine for sports. (Even there, require expert judgment to apply in a sensible way.) Useful ones not obtainable in real life. Too easy to game.
The panopticon: coming. I doubt this will create a significantly useful “customer rating” because of disparate impact; we all have the right to “privacy” which is to say that only a conviction in a court of law can be used against you, and probably not even that. (In a non-democratic state, though, very likely.) Indeed, as you suggest we saw the exact opposite play out in mortgages, which were big money enough to have a “customer rating” even before serious computerization of everything.
It will be interesting to see how the impact of the digital panopticon plays out in different legal regimes and social systems. I can see a whole spectrum of approaches, from maybe a ‘European’ and perhaps quasi-futile prohibition on collection and use, to maybe a ‘Chinese’ overt encouragement or insistence on it.
This is metrics working their way up the work hierarchy. … But Cowen is talking about professors, lawyers, doctors, etc.
Overall it’s a big GIGO problem. If we measure academics by number of published papers, we’re going to get a lot of worthless papers. We are still debating whether “Moneyball” really works.
… But the metrics just aren’t that good as soon as the work gets complicated.
It’s the classic problem known as “Goodhart’s Law.” If the metric itself is all you care about, it’s pretty straightforward; we’re in the realm of classic piecework. But any metric that is only a statistical correlate of the thing that’s actually important will rapidly degenerate into uselessness, because it incentivizes people to target the metric rather than the really important goals.
I am probably an outlier with my experiences, being in an industry where things are still done in far more unregulated and less bureaucratized manner than anywhere else, and moreover also working in a team that has these characteristics to an exceptional degree (despite existing within a large corporation). Still, in all of my working experience, successfully done projects have always rested on informal sorts of judgment that basically boil down to whether someone truly knowledgeable and reliable — a reputation established by an informal but convincing track record, and only somewhat correlated with formal rank — is keeping a close eye on things. Whatever metrics I’ve seen used have always seemed to me utterly worthless unless backed by this sort of informal assurance that, so to say, real adults are in charge of things.
Chapter 8, Why the Turing Game Doesn’t Matter
It’s the bumps and delays that will make the rise of smart machines a livable process. It could well be destabilizing if the technologies of mechanical intelligences – two hundred years’ worth of progress, plus their complementary applications were placed in our hands overnight. A lot of people couldn’t get any jobs whatsoever because they couldn’t work with the advanced machines, and it would take them a long time to learn. We deal with machines today as well as we do because our progress has been gradual, allowing us to learn along the way. When it comes to technology, progress is usually good, but gradual progress is usually better.
This seems plausible to me. One of the theories of the Great Depression compatible with Alexander Field’s Great Leap Forward is that period of invention in the decades around the turn of the century was so radical that it completely disrupted established modes of productions and patterns of specialization and trade. This was the ‘Edison-Diesel era’ with new electric devices and internal combustion motors opening the door to all sorts of new applications and possibilities.
In the ‘technological deterministic’ Great Depression-causation story, the adjustment to these new realities was completely traumatic, not only affecting employment levels and investment volatility and yields, but also having an enormous and rapid impact on trends in politics, society, and culture.
We may be seeing something like that again. Here’s an excerpt from Cowen’s The Complacent Class
When it comes to ordinary, everyday American life, how quickly will matters turn chaotic or disorderly again, and what forms will that implosion take? .. The biggest story of the last fifteen years, both nationally and globally, is the growing likelihood that a cyclical model of history will be a better predictor then a model of ongoing progress. .. All of this can happen even if you think the majority response will be a great and greater love of peace.
I wonder if slow and steady growth is not just preferable in terms of personal and social stability, but if to some extent it has been ‘caused’ by human and social limitation. Perhaps the ‘drag’ of learning curves and the constrained pace of human adoption of, and adaptation to, new production technologies has been like a governing mechanism which, to some extent, may have been responsible for a kind of disruption-reducing smoothing-out or slowed-rollout of what would otherwise be sudden, radical shifts and booms.
That’s mere speculation, but it there’s anything to it, then we might have some reason to expect “this time is difference” and things will be worse in terms of adjusting to new social equilibria “in time”. When new capital technologies and machines acted more as complements than substitutes for labor, then major expansions of production required training up large portions of the population, which was a slow process. But in the future, if fewer and fewer specially talented humans are needed to partner ‘freestyle-like’ with machine intelligence, then there will be both the ability and incentive to select for that subset with the highest adaptability and shortest adjustment times. As for everybody else … it’s not a pretty picture.
2. The Long Run
Cowen covers some of the most prominent ‘futurist analysis’ in this area.
First he mentions Yudkowsky’s bleak ‘foom’ prediction and unfriendly AI-takeover, but dismisses it, with an only-somewhat subtle dig at what might be the underlying psychological impulses that encourage people to focus on this particular type of scenario:
But the evidence so far doesn’t suggest this kind of unstable cascade to be very likely. Even the strongest programs need human assistance every step of the way. …For all their practical abilities, there is no reason to expect these [chess] programs to move down the corridor of self-awareness … The truth is there are no real vampires, no dragons, and no HALS. Let’s not worry about them appearing under the bed or in our hard drives.
Hanson’s next, naturally. I’m guessing he got an early look and probably objected to the word ‘dystopian’.
In another dystopian vision, the proliferation of computer intelligence will bring about a Malthusian world where human laborers will struggle to warn subsistence. Economist Robin Hanson, my colleague at George Mason University, considers this scenario in the fascinating and influence paper “Economic Growth Given Machine Intelligence.”
… It’s even possible that the machines will be cheaper than the level of subsistence wages. In that case either workers must live off charity or the population will shrink rapidly or some mix of both. … but eventually machines substitute for intelligent labor and wages can fall rapidly.
But “who will own the machines” and “what will government even mean”? Maybe we’ll spread the wealth.
The machines are still owned by someone, and the owners of machines are very wealthy … Alternatively, perhaps the government owns a share in the machines and it uses that wealth to support the remaining poor, who did not but machines in time and who now cannot find jobs because of competition from the machines. They will become wards of the state, much as many people live off of oil wealth in some of the less populated petro-states.
But we have this thing called taxes. Taxes on income and property and sales, oh my. And those taxes are already like the government owning a share of all companies and output. And pretty much all governments collect a lot of taxes for the explicit purpose of redistributing that wealth.
And, despite the differing nominal titles and official, de jure structures of formal distribution of power, we have also seen a de facto international convergence of systems of political and economic organization in which private parties are allowed to own and allocate capital and perform entrepreneurial functions (so long as they do not upset the political apple cart.) Over time, we can probably expect competitive pressures to select against any severe deviations from the basic outline of this economic model.
So I expect the mixed economy and redistributive taxation to continue, and the question is in what form and amount. There are also the questions of how big the ‘ward of the state’ class will grow, and how the inherent tension between an increasingly concentrated machine-owning class and a growing but voting welfare class will be resolved in democracies with universal suffrage.
Still, Cowen doesn’t think we need to worry too much about this vision:
Robin’s paper is thought-provoking. Still, I am looking across a more modest time-scale and more modest set of changes. Robin’s analysis might apply to the very long run, perhaps hundreds of years from now. But for the next fifty years or longer, the Freestyle model is more application. Most AI applications still require human support, and those applications, even if they spread considerably, will not come close to displacing all human jobs. Instead, intelligence machines will replace some laborers and augment the value of others in a slow and halting manner.
I think this is more far sanguine than warranted. One doesn’t have to believe in foom or ems coming soon to see a higher probability than Cowen of fast and disruptive change. It will take time for entrepreneurs to discover “those new service sector jobs” for all those displaced laborers. And the discovery process may not be able to keep up with technological and economic change. Not just for a little while, but maybe from now on.
I think major cultural and political shifts will be necessary to establish a new social equilibrium well-adapted to the new – and still evolving! – realities. And I am not optimistic that our institutions are up to the task, or that out present ideological environment is at all conducive to it.
Finally, how did we get to page 137 without mentioning ‘singularity’ yet. Ah, here it is. Cowen is really writing about ems, and could have stuck with Hanson since Robin did literally write the book on the subject. Or I guess he had come far along into his draft by the time AIO was written. But Cowen is rule-of-three futurist name-dropping, so it’s Kurzweil of course.
The most radical hypothesis about future technology is Ray Kurzweil’s vision of a machine intelligence “Singularity.” Kurzweil argues that mankind will obtain the capacity to scan brains and upload them into computers. There will be many copies of each “person” and presumably these entities will exist for a long time, with the multiple copies making the “person” hard to wipe out, even in the event of a system crash. I’ve heard some of Kurzweil’s followers claim this scenario will happen within the next fifty years, and Kurzweil’s writing seem to encourage such speculations.
… I suspect this will never be viable, if only because the human brain is so intimately connected to the human body, and because it so relies on the body for inputs and nourishment. … That means “brain emulation” requires building a whole working body (or significant parts thereof) not just an abstract, digitalized “brain in a vat.”
This seems to me to be Cowen’s weakest argument in the whole book. Even if true in its strongest form, why not just emulate the endocrine system too? Even the entire body would probably be profoundly easier than the brain. He has debated Hanson on this point, occasionally with some indirectness, and, well, Cowen is highly gifted in responding to criticism in a way that makes him appear as little like a debate loser as possible.
This is one of those times where I suspect Cowen knows perfectly well what’s up, and has some ulterior motive for sticking to an obviously weak position, perhaps to be palatable to most of his audience while expecting close readers of a Straussian disposition to pick up the real message. I know brothers, many of you will say I’m giving him too much credit. But I think he deserves it, and this case it seems clearer to me than most.
He returns to the critique in which he claims that the real reason these narratives are believed by their proponents is not because of the compelling case of the argument, but, ahem, because of the instinctive psychological appeal of the dramatic contexts they describe – like a bildungsroman plot line following Joseph Campbell’s The Hero with a Thousand Faces. He’s not subtle at all this time around.
All of these extreme scenarios, whatever their differences, share some features. They are about worlds that we cannot today control, influence, or even visualize clearly. They tend to be about worlds we cannot investigate empirically or find direct parallels for today. They are somewhat like religious allegories, involving some mix of concepts of deliverance, resurrection, destruction of the current world, and immortality. Given the appeal of such tales, I’m suspicious that these extreme scenarios are living off of their emotional appeal, and they are a kind of religion for computer nerds. Let us set that devotion aside.
3. Man-Machine Convergence.
Cowen discusses the Turing test and some of the chat programs. Occasionally I like to check up on the latest programs that run online. They are more powerful than in the past because often powered by a search-engine or database, Watson-like, but by now they all seem to have long settled on the same set of what are little more than gimmicks. These are designed to cover up obvious deficiencies and fool the poorer judges who are unserious about making any real attempt to uncover the presence of a program. These gimmicks include pretending to not be an adult fluent in the language of conversation, and framing a bunch of non-sequiturs or nonsensical responses as ‘jokes’. The irony – kind of a meta-joke really – is that these programs don’t get your jokes, and testing for a sense of humor is one of the easiest and quickest ways to determine whether you’re really dealing with a human. But it’s hard to maintain interest in the progress (or lack thereof) of these programs when the whole effort seems to have entrenched itself into mere gimmick-improvement.
He has a pet theory about Turing and his namesake test.
Turing was gay. He was persecuted for this difference in a manner than included chemical castration [Diethylstilbestrol] and led to his suicide. In the mainstream British society of that time, he proved unable to consistently “pass” for straight. Interestingly, the second paragraph of Turing’s famous paper starts with the question of whether a male or female can pass for a member of the other gender is a typed conversation. The notion of “passing” was of direct personal concern to Turing and in more personal settings, Turing probably did not view “passing” as synonymous with actually being a particular way.
The movie The Imitation Game (which sounds like The Crying Game, probably not by coincidence) didn’t quite go into the embarrassing, rough trade particulars of how and why Turing got caught in flagrante. Lots of people thought Turing was a real genius and also a very strange bird. It’s not clear how many people knew or thought he was gay, but it seems he was careless about hiding it.
Turing did not make this point, but many human beings, intelligence and of mature age, could not pass what we now call a Turing test. This includes many human beings who would do well on IQ tests or other traditional measures of intelligence.
I get his point, that ‘passing’ as a normal human requires a mental module which automatically and correctly infers norms and cultural conventions from experience and then conforms ones speech and behaviors to what will be accepted as ‘normal’ by ones interlocutors. And maybe better than merely ‘normal’, this module also seems to be a kind of “mental deep state” pulling strings behind the scenes and shaping one’s consciousness, preferences and emotions to present an effortlessly sincere character role, one that is calculated to produce the optimal social result for the speaker: an actor who doesn’t know he’s acting. And this social calculus module is obviously defective or deficient in many otherwise talented people, disproportionately males, some of whom try but fail to compensate for it by disciplined effort and training, for instance in the basics of ‘Social Interaction Artistry’.
But I think that his ‘many’ might be carrying more weight than it should. I’ve known a few of those people, and one can still tell right away from written conversation that one is dealing with a ‘human with issues’, which is a totally different experience than dealing with a human-imitating machine.
His larger point is a kind of argument from comparative advantage, which is that the incentives in place will ensure that, in the beginning, machine intelligence – not just automation of simple tasks but higher-level information processing – will evolve in the direction of complementarity instead of substitution. The machines will be doing what machines do best, and the humans doing what the machines can’t do as easily or cheaply.
What does this say about the collaboration between humans and computers? Convergence isn’t needed. In his question to take down the notion of imitation as the standard of intelligence, Turing notes that a machine will be most effective, cognitively speaking, when it does something other than imitate or try to imitate a man.
If the Turing test is cracked, how might the asking and answering bots change things?
“Face time” will become all the more important as a signal of actual interest and caring, because “computer time” will be too easy to replicate through the bots. Maybe you’ll use Skype to prove it is really you, and that will work for as long as bots cannot replicate your facial expressions and voice patterns through a streamed image”
Which, might be closer than we think. These videos are freaky-impressive:
On machine-enabled cheating and “symbiotic complementary specialization”.
That all said, a true systematic cheater will usually leave telltale marks on his or her games, at least if the cheating is repeated rather than one-off. Ken’s method is likely to catch the chronic crook, but it will not detect a grandmaster who cheats only a tone critical turning point in the contest.
In any case, rather than converging, man and machine are likely to become more different in some ways, including cognitively. Most of this book is about the evolution of the machines, but people will change too. I’m not talking about longer-run chances in the genetic code, but rather more simple changes in how we live out lives, and which skill we decide to acquire or not. To put it bluntly, we are outsourcing some parts of out brain to mechanical devices and indeed we have been doing that for millennia, whether it be to writing implements, books, the abacus, or a modern supercomputer. In response to all these developments we have focused more on the skills that the machines can’t bring.
Maybe a good analogy for what will happen to both humans and machines in the evolution of machine intelligence is what happens very early on in an instance of biological symbiosis. Originally you have two very different organisms that only occasionally make low-value ‘trades’ and are otherwise independent and able to survive well on their own. But eventually comparative advantage divides labor up so that the partner more efficient at some particular task is the only one doing it, and the other partner eventually completely abandons the capacity to do the task itself as ‘deadweight’, increasing the gains from trade, but making both creatures completely co-dependent.
4. Memory and “Search”
The Google crutch, if I may call it that, influences how we think and how we learn. There’s now good systematic evidence about how Google changes out mental capacities, and I think most of us have experienced this personally as well.
Definitely. I’ve often been amazed with accounts of the past in which, for example, story tellers were expected to memorize entire complex narratives verbatim. I’d guess that the transition away from intensive focus on memory began with the spread of the printing press and literacy.
In earlier times there was a prominent “science of memory,” in which was taught the skills of remembering numbers, people’s names, sequences of numbers, and so on. This science goes at least as far back as ancient Greece, and it flourished during medieval and Renaissance times. Journalist Joshua Foer recently wrote eloquently about this kind of memory discipline in Moonwalking with Einstein and returned some prominent to it. He focused mostly on one particular technique known as the “method of loci” by which elements you wish to remember are given a position in an imaginary place along with some surprising or colorful attributes. Foer recounts his adventures in the USA Memory Championship using his newfound skill of memorizing decks of playing cards.
The ancient arts of memory, in their most general form, and techniques to improve your mind. … The point was to make your ideas “searchable,” … Google is making a lot of the memory arts fall away altogether. … Second we have become much better at searching for answers, and that too is a skill.
Being a real power-user pro at the art of internet search and having a good instinct for how to use particular search engines most efficiently in the case of hard-to-find-but-know-its-there items is still a respected and important skill in my workplace, and in any law office paying out the nose for still-often-poor Lexis or Westlaw search results. It’s always possible the search engines for those services are intentionally ‘crippled’ so that they generate some profit through unnecessary extra searches. I’m complained about Amazon’s once-lousy search engine before, but I wonder to what extent the company was hoping that they could tempt you with some unrelated items. Hmm.
Two different effects are operating here, but we can tease them apart for a look at where humanity is headed. On one hand, many successful individuals will learn how to think like smart machines, or at least enough to understand their operation, in order to become wealthy, high-status earners. In that way we will become more like computers – well, a large number of high earners will become more like computers anyway, cognitively speaking.
That raises some questions of how we should try to educate and evaluate people in these useful machine-intelligence-exploitation-artistry skills. Will professions try to prevent that, to protect incumbents?
That said, when it comes to our private lives, we will become less like computers, because we rely on computers for many basic functions, such as recording numbers, helping us with arithmetic, and remembering facts through internet search. In these ways we will become more intuitive, more attuned to the psychology and emotions of everyday life, and more spontaneously creative.
I’m not sold on that claim. Thoughts?
5. The Machine’s Place.
It seems even when the machines are a lot better than the human competitors, we’re not so interested in watching them. For instance, when it comes to chess we humans don’t seem to care when the machines play each other. Hardly anyone is watching or talking about the computer vs. computer games. … It seems we care more about drama than about perfection.
Well, yeah. But wait until we have huge battle bots! Seriously though, of course few people care about machine contests in competitions the purpose of which is to rank humans by relative status. Playing a game serves a different purpose than its study, and watching humans play is exciting because that purpose is a major factor.
What does that all say about us as spectators and human beings? We don’t want to treat human beings and computer programs the same way, even when the latter become extremely skilled, or perhaps especially when the latter become extremely skilled. We wish genius machines to serve out practical ends, but we don’t want to turn over to them the spheres of life that structures out narratives , drive out emotions, define what our lives are all about, and help us separate right from wrong., We’re determined to “keep them in their place.”
… We won’t always listen to business or negotiating advice from the genius machines, and maybe we won’t be as interested in the music they compose, buying it only if a human composer pretends to have the creator or co-creator.
Heh. We already use celebrity performers to present the products of the real creative workers behind the scenes. Now humans will be playing Christian to Cyrano for the affections of Roxane, and taking the credit and providing the ‘brand’ for decisions and products made by programs. “Automation with a human face.”
1. The Great Depression was a monetary problem. Boom/bust. Exacerbated by poor policy once it started. This is accepted fairly generally, I thought. I don’t see the need for more causes, and I am very skeptical about the idea that (i.e.) the existence of the automobile was so challenging that it caused any real cognitive fails and subsequent economic effect.
2. As you say, modern governments already de-facto own a share of every company. This is what income taxes are, viewed operationally. Indeed, this is something I would formalize were I promoted to Fnargl. No corp taxation, but when you form a company or expand it, 20% of all new shares are allocated to the government.
In any case, as Moldbug said years ago, the problem with disemployment is not that everyone will starve. We already have extensive disemployment and nobody starves. It’s that people need work to find meaning in life, and when robbed of meaning they tend to live lives of vice, crime, squalor, and alienation.
We move on to Solution B, which I think is the solution most people believe in. Work? Who the hell wants to work? Work is anti-hedonic by definition. If it didn’t have negative utility, it wouldn’t be work. So, it’s supposed to be a problem that in the future, work will be obsolete, and we’ll be able to produce goods and services without any human labor at all? That doesn’t sound like a problem to me. It sounds like a victory.
The problem with Solution B is that we’ve already tried it, quite extensively. You see Solution B every time you go to the grocery store. Next to the button marked “Debit/Credit” is one marked “EBT.” Ever pressed that one? Even just by mistake? It’s the Solution B button. America has entire cities that have moved beyond anti-hedonic labor disutility and entered the gleaming future of Solution B. One of them is called “Detroit.”
Regarding Cowen on singularity, first, the “human ems go foom” seems to me less likely than “pure AI goes foom”. Does he not address the latter? Wrt human ems, Cowen’s argument there is totally retarded. Your “Straussian” explanation is I guess most plausible, because writing anything as stupid as that is implausible.
3. The Turing Test was not restricted to the detection of neurotypical humans. Certainly not as proposed by Turing. (His sexualization of it is interesting… nothing like that would happen today.) It is simply whether one can detect a computer attempting to pass as human. As such, if the computer decides to try to appear to be non-neurotypical, and that fools the interrogator, that counts.
I don’t think the symbiosis metaphor will fly here. The problem is that evolution only hill-climbs. If you’re a symbiont and you’re getting all the X you need from a partner, then immediately you will be under evolutionary pressure to cease making X yourself. There is no path to you becoming X independent. In the analogy, it is true that man-machine symbiosis will largely work this way, because most of it happens within the context of the market. And markets are profit-optimizing; they are hill-climbers. But there is still a path to further AI development: people can act outside of markets. I.e. we do research that is not profit maximizing. We can “jump” around problem spaces, including going “downhill”.
4. For the claim that “we will become less like computers”, it is perhaps useful to analogize computers:humans::nerds:normals. Are normal people getting less tech-capable due to nerds doing that stuff? I dunno. Certainly people work on cars less than they used to, because we can’t as much. But OTOH the massive increase in computers everywhere means a lot more people interacting with them in at least some capacity. Those facebook pictures don’t take themselves.
5. It’s retarded thinking people care about machine vs machine “battles”. By contrast, as we saw recently, we care about human vs. human “battles” even if they are as stupidly trivial as rock-paper-scissors.
I think a genius machine that somehow assigned people purpose in lives would be very popular. Very, very popular. In fact if a decent human-friendly AI ever gets going I am quite certain it will be worshipped as a god. Imagine you pray to God to stop being so lonely. The next day God suggests that you go to the autostarbuck at 9:06… standing there alone waiting for her mocha is this blonde in a “mysteries of science” t-shirt… and you work on the “mysteries of science” veeyarcast!
Re: Turing test, people tend to get into quasi-profound philosophical arguments about it, entirely missing the forest for the trees. The simple fact is that there is a wide range of common-sense topics where any adult human not suffering from severe mental retardation can make direct, clear, and accurate judgments, but which are presently utterly beyond the capacity of any present machine intelligence.
Just start asking a series of trivial common-sense questions. Can a mouse hide in a coffee cup? What about a horse? What’s worse, getting run over by a truck or a bicycle? There will be an immediate striking contrast between complete nonsense output by a computer (except maybe where it gets lucky and returns a direct Google hit) and sensible answers given by even the feeblest (within normal bounds) human intellect.
This clearly shows a vast chasm between currently existing computers and what we take for granted as common sense in humans, which we haven’t even begun to bridge, and we don’t have a good idea how we’d even approach the problem. When (if ever) computers start passing such rudimentary Turing tests, it will be a real game-changer, compared with which all these subtler philosophical issues are insignificant.
Part III: The New World of Work
Chapter 9: The New Geography.
…we have seen a great stagnation of wages in the United States since about 1973. … Should we blame the foreigners for our difficulties? And how will foreign competition shape jobs and wages going forward.
The three common dimensions of the ‘blame foreigners’ argument are:
1. It’s competition from foreign firms in the same spaces as US industries under free-ish trade.
2. It’s the immigration.
3. It’s the outsourcing.
Cowen dismisses 1 and 2 as negligible, but as for 3:
Some of my economist friends will hate this: It is increasingly hard to deny that outsourcing is playing some role is stagnant American wages and slow job creation. …
Economists use the forbidding phrase “factor price equalization,” … a similar mechanism operates for labor. If workers in India or China are much cheaper, and not correspondingly unproductive, either the workers will move to the capital or the capital will move to the workers.
I’ll leave the immigration claim to the side. But it seems to me that “foreign competition”, if benefiting from more favorable local conditions like lower wages or looser regulations, is really a lot like outsourcing, except with formal ownership and incorporation being abroad instead of in the US. Of course, a US company isn’t actively managing the migration of operations from domestic locations to abroad, but if the foreign company has comparable organizational capital, then their products will be cheaper, and eventually the US company will be forced out of business. And that’s assuming no other special advantages or subsidies that those countries might give to their domestic exporting companies.
Anyway, what about workers displaced by outsourcing?
The reemployment of displaced workers has been relatively weak since the late 1990s for reasons that are no fault of the Chinese. The rising demands have been for very special kinds of skilled workers rather than for workers as a whole.
He then reviews Michael Mandel’s argument that a lot of productivity growth (meager as it’s been!) is confusing gains in nominal productivity (i.e. lower prices for the same amount of inputs) with real productivity (i.e. innovation opening up the possibility of a new production function which needs less of an input, at whatever price, to make the same output).
In terms of labor productivity that’s a big difference. Operations can become more “productive” in nominal terms in this way by finding cheaper workers to do the same things in the same way with the same kind and amount of equipment. Operations become more productive in real terms via new automation solutions or labor-augmenting capital tools, even with wages and interest rates held constant. Sometimes one can source abroad to places where wages are so much cheaper that they compensate for drops in real productivity.
But Cowen says outsourcing isn’t enough to explain it either:
… but outsourcing just isn’t big enough to be driving the problem of stagnant wages. It more like one of several reasons why a lot of low-wage American workers aren’t reemployed as rapidly as in times past.
So what’s actually happening?
During these periods of prosperity, we were world leaders in education – K-12 and university. There was a closer match between the skills required of workers at higher levels of the value chain and the skills that American workers actually possessed. Nowadays, the demands of machinery – including of course computers – are rising at a faster rate than are human capabilities. The machines are getting better education, more rapidly and more cheaply, than are their human teammates and potential teammates. That the root of the problem for a lot of workers.
What a conspicuously bizarre way to say it.
Many millions of people can turn a screw on an assembly line, work a lathe, or handle a telephone switchboard. Not so many people can team up fruitfully with Rybka or, more generally, work with intelligence machines … To blame the Chinese, or outsourcing, is to point to a mere shadow play …
But we see manufacturing wages in China rising a lot. Why doesn’t the same logic hold these workers back? A lot of them are turning screws, and there are billions of cheaper screw-turners. Maybe they’ll stagnate now, but what about that recent long, substantial rise?
Back to chess:
By the way, the very best US chess players are often foreigners in a sense – most of them having come from Israel and the former Soviet Union. You might think they’ve “taken slots” from US players, but the deeper reality is that having lots of good players in your country, foreigners or not, helps – not hurts – your chance of becoming world champion.
Yeah, but what about your chances of becoming national champion. To the extent that factor price equalization has not quite run to completion, preserving excess domestic wages is obviously in the interest of people currently receiving those wages.
And if only there were some kind of technology that allowed these Americans to play and learn from these masters, even if they were abroad. Maybe they could send letters with moves via the postal service. Yes, I know that would be slow and cost stamps, but there must be some way, right? Maybe they’ll invent some kind of technology one day …
And what about our poor child chess prodigies abroad, when all their nations’ grandmasters have been moved to the US for our own selfish motives in ‘our’ quest to achieve global chess domination? Will they not be deprived?
When it comes to outsourcing, three truths need to be observed.
First, there is no realistic way to stop Americans from investing abroad. …
I think the government could figure this out if it was really interested.
Second, at a fundamental moral level a job for “a foreigner” is every bit as worthy an outcome as a job for “a real American.” If Chinese wages are going up and American wages are somewhat flat, as has been the case, I say bravo to them and let’s try to meet the challenge. … It may ring hollow with the American electorate, but it is still true.
Maybe we should be more cautious about things that ring hollow with the American electorate, because three years later … well, let me put it this way, it looks like a lot of them were favoring those “real American” jobs over those equally morally worthy Chinese jobs.
I mean, think about it from the perspective of an owner of Pepsi stock. If Pepsi market capitalization falls by the same amount that Coke’s rises, then it’s a moral wash. Why should that Pepsi shareholder vote for board members expected to obey a fiduciary duty to vigorously pursue the policies that maximally increases his own wealth, with total indifference to what happens to Coke owners?
Well, we know what Pepsi is supposed to be for vis a vis their stockholders, and we know that most economists would agree. Milton Friedman, at the very least, would have argued against any other goal. It seems many members of that American electorate were asking themselves why shouldn’t countries be for them in the same way.
Third, if you’re worried about outsourcing, you should probably have a more liberal rather than a less liberal attitude toward immigration. If the United States takes in more immigrants, the areas in which those immigrants work are less likely to see jobs outsourced abroad. Immigration makes it possible to keep those jobs at home.
So long as they are being performed by immigrants for low wages and supported by the redistribution state. Hooray! Local J. O. B. S!
That’s a clever (and sneaky) rhetorical trick. People love jobs, and maybe they’ll forget that what they really love is the idea of there being more opportunities to do more lucrative things for the existing local labor force, or at least the idea of preventing existing patterns from evaporating. They might not be as enthusiastic about the mere existence of jobs for their own sake, regardless of what is being done or who is doing them.
Well, ok, maybe your manufacturing workers are out of luck, but there would still be plenty of local complementary / support jobs. Which, for some reason, would not be done by cheaper immigrants too? And I guess that’s better than watching the rest of the local economic ecosystem evaporate and go abroad. Still, is that what actually happens? Not exactly.
Here is another way to think about the problem. A lot of economic activity wants to locate itself in those regions that are the most populous, wealthiest, most important, and most central to the nerve center of our world. Not everywhere can be a center of the global economy and not many people and investors wish to focus their activities in Nebraska. … Part of a competitive solution, for the United States, is to build up its own economic and cultural clusters of importance. That means more people and also more immigration, both high-skill and low-skill.
A Global Geographic Trend:
Just as labor market outcomes will move toward the poles of either “very good” or “very bad”, so will the same be true for a lot of cities, states, geographic regions, and countries. What we see is that individuals with college degrees are gravitating to areas where a relatively high percentage of the other individuals also have college degrees.
… Circa 1970, the most educated and least educated cities differed from each other by about sixteen percentage points … Today the difference … has about doubled in terms of percentage points … Ambitious and talented young people today are more likely to want to live in a relatively small number of cities and regions, rather than spreading themselves out as much as they used to.
The internet has made some things nicer for some people away from the hubs, however
But if you wish to be a high earner, learning from other well-educated people, geographic proximity is growing in importance, whether in companies or in leading amenities-rich cities or most likely in both.
It’s happening everywhere, despite linguistic and other barriers. For instance, from Spain to Germany.
I hardly expect Spain to be emptied out, but soon observers will start to realize that “economic integration” isn’t exactly working out as advertised. What people expected from economic integration was a wealthier and slightly more ethnically diverse version of what they had ten or fifteen years ago. What they will be getting is a dramatic shift of labor resources into the most highly valued firms and also into the most highly valued business regions. There will be lots of “hollowing out” of various regions – at least in terms of well-educated high earners – and not everyone will be on the winning side of this process.
… A lot more of Mediterranean Europe is going to look like southern Italy and Sicily: somewhat empty in terms of economically successful enterprises on a large scale. It will rely more on tourism and more on retirees and government transfers, but overall the fiscal problems of these regions will worsen and they may lose some political autonomy.
What about protectionism in the US?
I don’t fear a big surge in protectionism in this country, or in the world more generally.
Three years later …
About those new service sector jobs (echoing The Great Stagnation):
The security of these non-tradable sectors is nice for many of us, but it also means that people in most newly created jobs in the United States aren’t facing so much of a daily market test. Most of our job growth is coming in what I call low-accountability sectors. People get paid to produce things, or offer services, and we’re never quite sure how much value they are putting on the table.
Keeping America Great Still
The good news for Americans is that the longer-term trends seem to favor the relative position of the United States in the global economy. The United States is likely to continue as a leader in applying artificial intelligence and this will likely cement long-term American economic growth.
One reason for this is simply that America can sell its artificial intelligence products to the rest of the world, but a deeper mechanism is operating too. If there aren’t many workers in a plant or project, hiring in a wealthy country brings less of a wage penalty.
So, for example, if we built giant factories with enormous capital-to-labor ratios and full of robots, then there are only a few people around, and one can be choosy and select the ones least likely to be costly troublemakers. Aggregate wages are such a small part of total operational expenses that siting will be based on other considerations, such as being closer to the customers when maximum speed if essential.
But of course wages are just one reason why outsourcing is attractive, and there are plenty of other things which can make a location an easier and cheaper place to do business. Can any place in America compete on electricity costs? What about regulatory burdens, or the liability system? On which of those other measures is the US also globally competitive outside of certain industrial sectors where institutional and human capital is most important? Um…
After all, Mexico is right there. I have a feeling new ‘gigafactories’ will only be set up in the US because of regulations, protectionist politics, and subsidies.
No one really cared about Chapter 9 I guess.
Chapter 10: Relearning Education:
Education is a touchy and politicized subject about which typical beliefs are so stubbornly distinct from anything approaching an accurate view of the real state of affairs that it’s nearly impossible to have a sensible conversation. It’s also why we’re so far out on the diminishing marginal returns curve that education expenditures are draining the budgets of local governments and student-loan-encumbered-youth dry, and why we tolerate forgoing enormous opportunity costs for such pointless wastes of time that there is probably plenty of worse-than-useless Hansonian Education to go along with Hansonian Medicine.
But it’s sacred. So, if he’s going to get progressives to pay attention, Cowen can’t charge into the barricades like Kling raising the Null Hypothesis banner, or Caplan firing his pure-signaling-model canons, or Hanson quipping that “School isn’t about Learning“.
Speaking of Caplan, he said in mid ’16 that his education book would be published in March 2017, but I haven’t seen any sign from him or anyone else of the typical pre-publication marketing blitz. His publishing company wanted to avoid the election period, and I’d guess it also wanted to deconflict schedules with Cowen’s The Complacent Class. That was smart call, since Cowen took up all the GMULE marketing bandwidth for a while.
Anyway, we’ve talked about the Cato Clever™ strategy before, and specifically applied to education reform advocacy. It’s like Judo. instead of “No, because …” it’s “Yes, and therefore …” That is, instead of resisting the movements of your opponent, which they are prepared for, you unexpectedly move to the side and give them room to keep charging – and even help them do so – so that they make absurd overshoots. You then apply a little unexpected pressure in a strange direction, which causes them to stumble.
For policy matters, instead of disputing certain sacred assumptions directly as Hanson, Kling, or Caplan might do, one adopts a posture of accepting these assumptions in earnest (i.e. not saying you accept them merely provisionally or just for the sake of argument) and then use that very assumption (plus some convenient theory and studies) to argue that the implication of the logic leads to one’s desired course of action.
So, instead of saying that public education doesn’t make any difference to outcomes and is so wasteful and ideologically obnoxious that the huge government subsidy can’t be justified and should be abolished, say that of course public education could be making big differences, but is failing to do so for some reason – probably too much ‘uniformity’ or bad public choice incentives – and so school choice, vouchers, and charters are the way to go.
One can claim perhaps that school choice gets ‘better’ results. When that proved untrue in terms of test scores, one can move to goal-posts to utility and talk about parent and student satisfaction.
One way I think Cowen frequently does this is by using a variant of motte-and-bailey tactic on his own readers regarding key words. See here and here for Arnold Kling’s take on Cowen’s, um, unconventional and ambiguous uses of the word ‘Complacent’ in his latest book. (Kling preferred ‘pathological’, which, again, is much more direct and frank, but thus inescapably more provocative of the mind-closing psychological resistance Cowen seeks to avoid.)
In this chapter, Cowen seems to be playing the same game with ‘education’. Which, fortunately for him, is already such a confused term susceptible to a variety of different meanings and uses in ordinary parlance that his word game is much more defensible. Do we mean “what happens in school” or even just in the university or “any kind of learning experience” or what?
He tends to slide between meanings, but at least he is being clearer about a particular purpose for the sake of analysis. But right after an opening paragraph which introduced us to that particular analytical perspective, he injects an unexpected and gloomy vision of the future. It’s like when the pediatrician distracts the child to lower the impact of a shot. He starts the chapter this way:
We as a nation have been thinking about education without knowing what we really want from it. Do we want well-rounded young adults to emerge? Or good citizens? Role models? These goals seem reasonable but what do they mean? For the purpose of this chapter, and indeed this book, I’ll keep the goal simple. [b]One goal of better education is to procure better earnings[/b]. How we might achieve that is the question.
Whether we will remain [b]a middle class society or not[/b] depends firstly on how many people will prove to be effective working with intelligent machines. One percent of the population? Ten percent? Fifty percent? Secondly it depends on how many people will find work either as [b]personal servants or as more distant service providers to the high earners, [/b]and what wages they will be able to negotiate.
Ah, the power of juxtaposition. He doesn’t actually say that fiddling with pedagogical techniques, at least, those for the masses, will raise or lower that percentage of elite machine partners. More likely, as we’ve seen in athletics, the naturally gifted elites can be coached better than they are at present to be even more productive and more formidable global competitors. And the vision of the Neo-Downton Abbey future of formerly middle class people becoming less well off personal servants to the rich is concisely articulated. And since everyone already knows Cowen’s punchline that we won’t remain a middle class society because, well, ‘Average Is Over’, then, like Charles Murray has said many times over the last few decades – for example in Real Education – America’s future depends on how it chooses to teach its most talented kids. And as for everyone else, well, it’s not going to be pretty.
The answers will have a lot to do with education. In particular, how many people will receive world-class or at least satisfactory educations in the years to come so they can the best, or at least acceptable wages?
I would recommend a very careful and lawyerly reading of the above excerpt. It seems to be diverging from the Null Hypothesis and GMULE consensus, but I would suggest it’s written specifically to trigger that patellar reflex in his typical reader: of appearing to be drawing the mainstream, approved-orthodoxy causal line from “good education” to “higher earning”, but without actually doing so.
He then talks up the potential of online education, MOOCs, Khan Academy, etc. This was an interesting line:
Circa 2013, no one is surprised when a new foreign aid program consists simply of dropping iPads into rural Ethiopia and letting children figure out how to work them.
Well, I guess I’m not surprised that there are foreign aid programs like that. We do all kinds of stupid things with foreign aid. What do the fine effective altruism folks at Givewell think about these iPads? In fact, we have completely domestic programs like that which (at egregious expense) put iPads in the hands of ordinary students, and those schools are not exactly flooding the ranks of the software development labor force. On the contrary, there seems to be a lot of porn watching and game playing, i.e. enablement of modern, bad-habit vices.
Learning how to ‘work’ (that is, merely operate the user interface of) an iPad is also quite a bit easier than learning how to program for one. Or if that’s at all controversial, then let me put it this way. Anything you can teach nearly any child by means of “Ethiopian Airdrop” is practically by definition not a skill that will remain scarce in the way that could justify higher than typical levels of compensation. I mean, I think when automobiles first hit the mass markets, learning to drive them was challenging and something young people did better, mostly for analogous reasons of motivation that makes the Ethiopian country kids want to learn how to use their iPads. But truck drivers weren’t scarce for longer than a historical blip, and if a Teamster ever got paid well it wasn’t because of his special, hard-to-reproduce-in-quantity skills.
Anyway, the internet revolution is more than just substitutes for classes:
In my own field of economics, what is the most common and regular form of contact the general public has with economic reasoning? It’s no longer the Econ 101 class but rather it is economics blogs, which are read by hundreds of thousands of people every day. I submit that “cross-blog dialogue,” as I call it, is for many people a better way or learning than boring lectures, PowerPoints, and dry, overly homogenized, designed-not-to-offend-anybody textbooks.
He lists a few other obvious examples, maintains they are qualitatively different – and clearly superior to – the old “classes on videocassettes” model, and that they are well-suited to study and refinement.
The world still awaits systematic, rigorous (randomized control trial) studies of all these methods of learning, and it is too early to say what is working and what is not. Nonetheless, we do know two things for sure. First, very often the online methods and much cheaper and also more flexible than the previous alternatives. Second, some learners – quite possible a minority – love the online methods.
“Quite possibly a minority” is an interesting turn of phrase. Notice also that “loving” certain methods doesn’t necessarily translate into “flourish under and enjoy superior results for the relevant target outcomes.”
Here’s his argument for why online education will really matter.
The first is that online education will be extremely cheap. Once an online course is created, additional students can be handled at relatively low cost, often close to zero cost.
Ok, sure. But let’s think about why we’re even worried about education costs. For one thing, we know education used to be much cheaper and a much better value until fairly recently, by almost any metric one can devise. Also, we’re reading constant complaints about the bloat in both overpaid administrators and completely extravagant buildings and amenities on university campuses. For some reason these institutions have been behaving as if prices in education don’t matter much for a long time. If they start mattering all of a sudden, then there’s plenty of fat to cut out.
Another obvious example. You what has close to zero marginal cost of reproduction? Textbooks Pdfs for sure, but to the extent they are necessary or preferred, printed texts and photocopies could be negligible costs.
But look at the college bookstores – they aren’t even close to cheap. We have all heard the many specious arguments for why textbooks assigned by the professors in the typical American university are so expensive in the typical campus store, but these fall flat the vast majority of the time.
In a discussion about this topic over at Kling’s place, I looked up a price at the local college which I expected to be better than average, and discovered:
… let’s take a look at the book for George Mason’s Fall 2014 offering of Math 315, “Advanced Calculus I”: The Way of Analysis by Robert Strichartz. Buy New for … $224.65! Whoa. Talk about sticker shock.
Yes, there are the now standard ways to reduce the cost to students. You can ‘rent’ it, or get it used, and also try to resell it. But those coping mechanisms have emerged precisely because the book cost is so outrageous.
No disrespect to Strichartz or his publisher or the professor who picked this book, but in this field the same learning objective could be achieved with a free book that entered the public domain a half century ago.
We could be cheap right now if cheap mattered. Again, everyone is acting like it doesn’t matter. Universities are acting as if they are operating under Blagojevich logic.
Cowen admits that sometimes people are willing to pay high prices not just for some marginal extra value but precisely for the high prices – what they accomplish in terms of selectivity, and what they signal about prestige and wherewithal.
I don’t think the price will fall all the way to $200, because good schools won’t want to look too cheap, and maybe they don’t need the money, but still I expect the price for a class to be much lower than its current level, especially at institutions below the top tier.
We’ll see. It’s been four years since the book was published and things still seem to be heading in the opposite direction. One could guess that the economy will eventually evolve in a direction in which there is clearly little to no status or life-outcome premium to be gained by an average American by moving from, say, the 200th ranked institution to the 50th ranked one.
Online education can thus be extremely egalitarian, but it is egalitarian in a funny way. It can catapult the smart, motivated, but nonelite individuals over the members of the elite communities. It does not, however, push the uninterested student to the head of the pack. Here is yer another way in which the idea of a hyper-meritocracy will apply to our future.
… the rewards will flow more readily to top talent, not the socially well connected.
Again, this is hyper-meritocracy in terms of overall market-measured productivity, which is raw cognitive talent, learned proficiencies, plus CUTS (Complementary Uncorrelated Traits and Skills). If the student isn’t a striver – isn’t motivated, curious, diligent, perseverant, resilient, conscientious, emotionally stable (or well-adapted), scrupulous, and hard-working, then he isn’t going to get very far.
In terms of absolute levels of performance, machine-driven education will boost strivers at many difference levels. In terms of [b]relative [/b]performance, it is actually many of the non-top performers who will rise the most, because many of the very top performers would attract a lot of attention and instruction under any system, with or without computers. And then many individuals will not rise at all – they won’t sit down at the screen.
That’s going to be a lot of students.
Will online education spread at rapidly as computers have revolutionized chess teaching and learning?
… most likely it will not. One major problem is simply that universities are for the most part bureaucracies. Faculty often fear online education because they sense it will either put them out of a job, lower their status and importance, or force them to learn fundamentally new methods of teaching, none of which sound like pleasant prospects, especially for a class of individuals used to holding protected jobs that involve a certain amount of autonomy and indeed coddling.
Perhaps a preview of The Complacent Class. What about accreditation?
And the faculty are by no means the only obstacle. One central question is how quickly accrediting bodies will move to grant full and transferable credits for good online courses. I do expect to see some progress in this direction, but accreditors serve in part to prop up a higher education cartel. At some point they might think twice about allowing so much competitiveness into the market …
Chess vs. Harvard:
The chess programs are the very center of the teaching, and for the best students they end up being the most important teachers themselves. The computer becomes the center and – dare I say it? – the human teacher becomes the add-on …
The kind of machine-based learning is driven by a hunger for knowledge, not by a desire to show off your talent or to “signal” as we economists say. If you’re not a good player, the fact you studied with a top teacher doesn’t mean a thing. No one is impressed and no one will want you to play for their team. There is nothing comparable to the glow resulting from a Harvard degree … The company selling Rybka tries to make its product replicable and universal, whereas Harvard tried to make its product as exclusive as possible. Now, which model do you think will spread and gain influence in the long run?
Maybe. Harvard’s had a pretty good long run. To wit:
Of course Harvard, MIT, Stanford, and other schools of that ilk may end up being the ones providing the online education. … Yet there is a fly in the ointment, and it remains to be seen whether schools such as Harvard can excise it. The current business model of Harvard and Princeton is to market the quality of exclusivity and to raise money by encouraging alumni to donate to such a wonderful and exclusive institution. …
Given that business model, will Harvard and Princeton really be the ones to award credits for online courses, say to several thousand very good students in Bangladesh?
II. Face-to-face Instruction
What are the human teachers going to be for?
In part, the human chess instructor teaches the pupil how to use the computer. [ie. with mastery -H] The human instructor has also become more important for motivation, psychology, teaching pacing, and teaching the psychological foibles of potential human opponents. With younger and less experiences players, the skills include keeping one’s composure, maintaining concentration, and not getting psyched out or intimidated by older or better opponents.
Sounds like sports when taken seriously. I.e. Texas. In other words, the future human teachers of the very talented will be like today’s super-coaches to elite professional athletes. One third experienced expert mentor / one third zen-level trainer in the arts of meta-cognition / one third Tony Robbins-class confidence booster. Or something like that. Elites are going to get coaches along with their machine-based training to help them reach their potential and be their ‘personal best’. CEO’s, Surgeons, and so forth.
The machines will help teachers and students overcome the bias that sometimes emerges in what become very close relationships and could cloud judgment. Many a coach or trainer has become fond enough of a student to start to overlook their shortcoming at critical point, and then with remedies that are too little, too late.
The next step is that human instructors will consult the machines to better understand the mistakes their students are making. Even if the machines do the work measuring the mistakes, as discussed earlier, the human instructor may still be the one to interpret and deliver that information (in inspiring fashion of course), and the one to outline a course for improvement.
This probably demonstrates two things. One is that if “what is well-defined will be automated”, then what’s not yet automated is what’s not yet well-defined. The art of inspiring coaching is probably hard to define, and hard-to-define ‘arts’ will be left to humans as our comparative advantage. Like AI’s will encourage us to babble better.
But even if it’s not that hard to define, or for some Alpha-Go neural nets to extract the patterns from a million coaching transcripts, there is the question of whether humans can be motivated in the ‘inspiring coach’ way by non-humans, and what kind of status-based relationships are necessary to generate the kind of psychological states one is trying to produce. My hunch is that coaching is a job that has real benefits, but that humans won’t respond well to coaching by non-humans. They don’t even respond to humans who aren’t right there in front of them. The trouble is, there can’t be many more coaches worth their salt than there are elite performers, so while these may be some of those “new service sector jobs”, there will not be enough of those jobs to employ most people who will fall into the servant-class or underclass.
For all these reasons, chess lessons on Skype, as you might commission from India, have not become popular, even though they are cheaper than face-to-face instruction. The programs have forced chess instruction to evolve, in largely beneficial ways, and – here is a key point – in ways that make the job harder to outsource. The instructor who teaches human qualities like conscientiousness and who motivates his student need to be there.
Echoes of The Great Centralization and Handle-Baumol stagnation theory. The jobs that will be left are the ones that can’t be automated or outsourced. That means they are jobs that require humans beings to be close to one another. That leads to geographic centralization through tightening the physical spider web of economic relationships. And things humans do close to each other can only be done at human speed, which for ordinary workers is limited, so productivity growth stagnates. These are also tasks the productivity of which can’t be augmented much more by adding capital, so there is low demand for investment, but high demand for central real estate, in a slow-growth, low-interest-rate, rent-is-too-damn-high environment. The mid-century manufacturing era of the everyman was one in which there was plenty of remaining opportunity for high-yield complementary uses of capital to augment labor. Now there are plenty of substitution opportunities, but not much complementary opportunities beyond a small, saturating amount.
So, take coaching as an example. Coaching must be done in person by humans, and coaches can only coach so many players at a time, and perhaps only one at the top levels of certain fields. The machines will help the coaches coach, sure. But they won’t be that expensive. And past that small amount of money invested in software or hardware, there’s not really any additional opportunities to invest lots more money in a coach that will really pay off in terms of his becoming such a better coach that he could pay back high interest loans with additional income.
In the long run, professors will need to become more like motivational coaches and missionaries. The best professors have understood this for years and have been serving that function from the beginning. What’s less well understood is that improvements in AI will make these the remaining roles of what we now call “professors.” The professor, to survive, will have to become a motivator and coach in essence and not just accidentally or in his or her spare time. … We could think of the forthcoming educational model as professor as impresario. In some important ways, we would be returning to the original model of face-to-face education as practiced in ancient Greek symposia and meetings in the agora. … Let’s treat professors more like athletics coaches, personal therapists, and preachers, because that is what they will evolve to be.
Of course, educational institutions aren’t ready to admit how much they share with churches. These temples of secularism don’t want to admit they are about simple tasks such as motivating the slugs or acculturating people into the work habit and sociological expectations of the so-called educated class. As it presently stands, we are losing track of a college educations real comparative advantage.
This is the argument that Cowen uses against Caplan’s pure signaling model. Ok, maybe most of the kids aren’t learning much content in college according to exit exams. But they are maturing and learning the kinds of meta-cognitive skills, self-discipline, and SJW orthodoxy that will help them fit in with members of their likely future scenes and succeed in workplaces full of other ‘educated class’ individuals. And, ok, maybe we’re failing at that too now because, but it’s only because we’re ‘losing track’ of what college could be doing, as it was doing in the past.
Maybe educators are thought too highly of:
We like to pretend our instructors teach as well as chess computers, but too often they don’t come close to that ideal. They are something far less noble, something that we are afraid to call by its real name, something quite ordinary: They are a mix of exemplars and nags and missionaries, packaged with a marketing model that stresses their nobility and a financial model that pays them pretty well and surrounds them with administrators. It’s no wonder that this very human enterprise doesn’t always work so well.
How good students will be treated vs. not-so-good students.
What does the resulting model of education look like? The better-performing students will be treated much as chess prodigies are today. …
The lesser-performing students will specialize in receiving motivation.
“specialize in receiving motivation” is another great Cowen-ism and quasi-euphemism. It actually made me stop and laugh out loud when I read it. Usually we say that people specialize in doing something, not having something done to them. So do renal failure patients now ‘specialize in receiving dialysis’? Anyway, what kind of motivation are we talking about, exactly?
Education, for them, will become more like the Marines, full of discipline and team spirit.
Ah, that kind of motivation. “I WILL motivate you, Private Pyle …”
And what about the drill sergeant of the hearth?
Not everyone will adopt the so-called “tiger-mother” or Asian parenting style, but its benefits will become more obvious. A lot of softer parents will hire schools and tutors to do this for them. The strict English boarding school style of the nineteenth century will, in some form or another, make a comeback.
First Downton Abbey servants, now this. Talk about Vickies.
If your eleven-year-old is not getting with the program, you will consider sending him away to the hardworking, whip-cracking Boot Camp for Future Actuaries. Neo-Victorian social ideals may not triumph, but they will become a much stronger force among lower earners.
Will we have star tutors for ordinary subjects like Tony Robbins for motivation in general or perhaps some the minor-celebrity PUA-trainers? They’re already here. Asia, naturally. Home of cram schools and grind culture.
Especially charismatic teachers will surely have their place – and probably a very well-paid place – in the new world of work. Hong Kong already has glamorous celebrity tutors, called “tutor kings,” … It is rumored that Richard Eng, one of the leading tutor kings, pulls in $1.5 million a year; his face is on billboards, he drives a Lamborghini, and his license plate reads simple “Richard.”
High-skilled performers, including business executives, will have some kind of coach. There will be too much value at stake to let high performers operate without a steady stream of external advice, even if that advice has to be applied rather subtly. Top doctors will have a coach, just as today’s top tennis payers (and some of the mediocre ones) all have coaches. today the coach of a CEO is very often the spouse, the personal assistant, or even a subordinate, or sometimes a member of the board of directors. Coaching is already remarkably important in our economy, and the high productivity of top earners will cause it to become essential.
Coaching is not going to be formalized for high-status jobs, excluding athletics. In all coaching it is the coach’s knowledge that is superior to that of the performer. I.e. knowledge-wise, he is superior; he has higher status. But you cannot lead an organization if there an obvious other leader with higher status. It makes no sense. The Board should hire the coach if he’s better than the CEO.
Athletics is an exception proving the rule. In athletics, the athlete has superior physical ability, whereas the coach presumably has superior knowledge. Since it is the physical performance that people want to watch, the coach can be lower status and yet still offer something to the coachee. This won’t work for mental performances — i.e., almost all work.
Furthermore, if coaching many (most?) high-skilled performers really will make economic sense in the future, it should make economic sense now. Perhaps not yet for more marginal performers, but surely it should for those whose performance affects billions of dollars in value. But who is Bill Gates’ coach? Warren Buffett’s? Etc. Who is Trump’s coach?
Again, notice the contrast with athletics. Today, top athletes uniformly have (team-shared) coaches; many have personal trainers and personal coaches. In fact “coaching” as a thing extends all the way down to tee ball levels. (I’m a coach!) And personal trainers exist and are hired by individuals fairly routinely, simply to help improve health and personal appearance. If you can get volunteer coaches in little league, and “personal training” is a consumption good, why don’t we have paid coaches for every Starbucks owner or McDonald’s franchise? I expect because these hypothetical coaches offer no real value.
There is such thing as a CEO coach, and many famous CEO’s have discreetly had coaching. Bill Campbell is a well-known example, he coached many Silicon Valley CEOs — See this article
Sales coaching is also a thing, companies routinely hire an outside firm to give their sales rep coaching to achieve higher productivity. Though most sales coaching is done in house, either via the managers or by hiring full-time coaches/trainers.
The reason why you hire an outside expert in business is because they have outside knowledge and more expertise because they have solved similar problems you’re likely to encounter.
While it’s better to bring as much knowledge as you can in house, it’s not always economically possible to do it. Trump obviously has tons and tons of political ‘coaches’ but they don’t exist to give them affirmations or to be his boss.
Also it’s funny to see predictions about automation that are almost entirely based on tech company PR pieces and that have not all that much to do with the reality of 21st century office work. I mean, the PR people at these tech companies don’t even talk to the engineers at the same companies — why do you think that predicting the economic future based on PR puff pieces is a sound strategy?
There are many hidden flipside costs and risks to automation that may not be readily apparent from a surface analysis. It doesn’t result in a straight line of productivity growth.
SNORLAK said: We’re actually only 1-2 years away from computers being able to consistently beat the best humans in no-limit Texas hold ’em. Already, if you’re dumb enough to play online poker for money, you’re going to lose to a computer (you are not one of the best humans). Assuming the unscrupulous online poker site owner isn’t fixing it, of course. Poker was actually considered an easier problem than Go, Go just had more resources thrown at it.
That’s not quite what I was getting at. As far as I know, those poker-winning algorithms are not connected to cameras scrutinizing human faces for revealing tells of intentions and states of mind. Those algorithms are combining a near-perfect statistical understanding of the odds of various hands, with some game theoretic considerations of the best possible brinkmanship strategies, and a huge repository of typical human master-level games and frequency of triumph given certain plays and betting behaviors.
So, the algorithm can ‘infer’ (kind of) what the human is really trying to do from the observable data of the cards, the track records, the betting patterns, and so forth. But not from actually physically observing the human, listening to variations in the timbre of his voice, noticing breathing, etc. Certainly not from carrying on a indirectly probing dialogue.
Now, those inferences from that limited set of game-based data are, apparently, good enough for the purposes of winning at poker. But similar approaches may not be good enough for other activities, for example, using human intuition, wisdom, and judgment to determine how seriously a doctor should take his patient’s story of symptoms, health history, and compliance with medical advice.
A algorithm for a ‘diagnostic computer’ taking down and processing symptoms at face value is probably not going to be able to be as good as the ‘walking intuitive polygraph machine’ ideal for a human doctor, or sophisticated customer.
Come to think of it, Cowen should have used the debreifer-polygraph combination as an almost ideal human-machine partnership, able to do things that no human or machine can do on their own.
Which is not to say that computer’s won’t get equipped with the rest of the observational suite of sensors they require, and will then get better at those sorts of tasks, and probably eventually surpass human-level skills too.
Imagine complaining to a future AI doctor about your back pain, and then being told in the Siri-voice, “I’m sorry Dave, I’m estimating a 98% chance that you’re exaggerating to get access to narcotics. Please go home and take an ibuprofen.” And how would anyone know if those systems are reliable and haven’t been manipulated by health insurance companies or the government to be overly ‘cynical’ so as to deny care to save money?
Get ready, this world is coming fast!
(To SNORLAK, attributing my bank customer service problems to old, indecipherable code)
I don’t think that’s it. Whenever I make almost any payment or purchase, whether by card in person, mobile phone app, or on the web, the user interfaces on all those platforms update within seconds and put a “(pending)” by the transaction for some amount of time. Certainly the system knows right away to reduce my available credit, and will decline subsequent charges over that amount not a minute later.
So, obviously some of the (no doubt quite old) communications systems and databases are updating right away. But something about the payment system puts these transfers ‘on hold’ for some amount of time, and makes them somehow tentative and subject to revocation or charge-back or detection of fraud during the window, only after which they are deemed to have been somehow confirmed, verified, and cleared.
To me this all smells of having more of a legal, regulatory, and corporate policy nature than having anything to do with the capability of the information technology.
I understand what you’re saying about the half-century-old programming languages and probably some antique hardware and procedures being involved as well, all invisible to me. But I don’t think that’s the root of the problem when the features I can see work so well and fast. We’re trying to explain the sluggish dark matter, but attributing the latency to dark technology doesn’t fit with the technology I can see – which would interact and probably depends on that dark technology – working so well.
Right, those ancient COBOL back-ends actually seem to be the best-functioning part of modern banking from the perspective of an ordinary user. In my experience, all the awful frustrating problems are either in the human parts of the bureaucracy (which should be straightforward to reform for a competent management) or in websites and other modern components of their information technology.
Riffing on the potential of ‘coaching’ a bit, Cowen barely touched on what I regard to be one of the low-hanging fruits of computer-guided learning, which are gains from radical personalization and custom tailoring. Most of us can appreciate the advantages of ‘tracking’ and segregating students by ability so that the more homogeneous group can be taught at their appropriate ‘level’ and pace. Of course, American education has officially moved away from this practice for ideological reasons(while unofficially reinventing it at the High School level with classes advertised as ‘Honors’ or preparatory for AP or IB tests).
But tracking taken to its logical extreme is like the one-on-one coaching / tutoring / apprenticeship model and the most important reason it isn’t done in general is that it’s economically infeasible to provide so many expensive human instructors. (Well, at least for normal kids. Because of various recent education-related laws, disabled children are often allocated these kinds of enormous per-capita resources, for very little gain, as education realist occasionally complains of.)
But computer are cheap, and algorithms have effectively zero marginal cost.
I took the ‘adaptive’ version of the GMAT on a computer many years ago and, I must say, even back then it was very good at escalating and quickly determining my ability levels and then keeping the questions within a narrow range of that level.
So it doesn’t seem implausible to me that, whatever their weaknesses, educational algorithms could be very good at providing highly personalized instruction, determining a student’s potential and comparative advantages, and optimizing for a custom tailored curriculum that will help the student reach his potential in a quick and efficient manner.
This data would be a goldmine for an actually competent counselor, and perhaps also for future admissions personnel. Why, after all, would you even need an SAT, if one had access to the cognitive potential estimates derived from many years of the student’s learning experiences?
Chapter 11: The End of Average Science.
The opening appeals to the “self-actualization” value of work among today’s highly-educated upper classes. Cowen’s idea of his own target readership is most evident in these few sentences. In this milieu, one isn’t a working class deplorable or a money-grubbing, penny-pinching merchant. One can’t be high-status and a lazy aristocrat inheritor living off land rents or dividend checks. You don’t necessarily have to be rich, but you do have to either be a celebrity, or have a job with a sexy and/or world-improving mission.
We also see a preview on emphasis of the themes “The Great Stagnation” and the usual mantra, “In the long run everything depends on the economic growth derived from innovation.”
Many of us have striven to work at something that is not only well paid but is meaningful and important. We want to contribute something substantial. Some of us wanted to be teachers, some medical doctors, some particle physicists. In the vast array of career choices it is easy to overlook the fact that modern professions all depend on scientific discoveries to one degree or another. So what about science? Is average over for science?
The basic idea is that the low-hanging fruit has been picked. Too many of the world’s greatest minds have been attacking too many of the greatest problems in too many areas of inquiry for too long and with increasingly powerful and expensive equipment to leave anything left that could be discovered in an easy, quick, and cheap manner by even an extraordinary, genius-level mind. At some point, the world is mostly mapped, and the remaining corners remain only because they are so hard to get to.
The argument is that we are way, way out on the diminishing marginal returns curve, which means from now on, minor discoveries will take teams of geniuses years to produce even with huge amounts of resources. And since the simpler, easier-to-understand patterns probably have been uncovered already, many of these discoveries will barely be comprehensible, if at all, even to the most elite minds in some specialized sub-corner of a discipline. One won’t be able to lead even a talented novice to the coal face anymore. Maybe to the mine mouth, but the vein lies a mile or more beyond.
We should not, however, take that state of knowledge as fixed. I’m not talking about decline in literacy here – science itself is, in many areas, moving beyond the frontiers of ready intelligibility. For at least three reasons, a lot of science will become harder to understand:
1. In some (not all) scientific areas, problems are becoming more complex and unsusceptible to simple, intuitive, big breakthroughs.
2. The individual scientific contribution is becoming more specialized, a trend that has been running for centuries and is unlikely to stop.
3. One day soon, intelligent machines will become formidable researchers in their own right.
The overall picture is a daunting one for the ability of the individual human mind to comprehend the science of how our world works.
We’re already at the point where there is not always common agreement as to what it means to “prove” a mathematical theorem. … no single mind knows if the theorem is true and instead a group of mathematicians goes over the theorem, divvying out the parts to the appropriate specialists.
He writes of HP Labs researcher Vinay Deolalikar’s stab at P≠NP seven years ago. My understanding is that the an expert consensus emerged fairly quickly that the approach was “fatally flawed”, but Cowen wrote it up this way:
As I write, the matter is still unsettled, though the mathematical community is learning in a negative direction against the proofiness of the supposed proof.
Specialization is also reshaping applied science and invention. Formerly, a researcher or potential inventor could learn the entirety of a scientific or applied area in a few years’ time, master it, and produce an innovation rather quickly, often working alone or in a very small group. The major inventions behind the Industrial Revolution, for instance, were often driven by amateurs. That’s become a lot harder because there is so much knowledge to master in the mature fields. It can take ten years or study or more to get to the frontier of a lot of area, and by the time you get there, and figure out something new, your contribution is a marginal one or maybe a little out-of-date. The frontier moved on while you were trying to master it.
Implication to rapid innovation:
The ability to “go it alone” is conducive to rapid innovation and innovation by amateurs. Lone individuals and small groups can make major contributions, and that limits the stultifying effects of bureaucracy and regulation.
Still, as the accumulated total of human knowledge increases, those breakthrough sectors become just a small part of our scientific understanding of the world. Science tends to look more like bureaucracy, and in standard bureaucracies no single mind has much of a grasp of the whole. In my own field, economics, coauthored pieces are already becoming much more common, and with a greater number of authors, as they are in many other fields of science as well.
Eventually, dare I say it, science will also look more like religion and magic because of its growing inscrutability. The working parts will be hidden, much as an iPhone functions without showing you its principles of operations.
In more recent times, in many particular areas, the hopes for comparably simple major breakthroughs have been dashed on the rocks. There have been plenty of scientific advances, but the world seems to be a messier place conceptually than before. Genetic explanations for human behavior continue to grow, but the connection between genes and outcomes is growing messier and more complicated all the time. Even the height of a person – a clearly heritable characteristic – seems to involve dozens of distinct genes, with more being found all the time. We’re not going to find a “gay gene” or an “autism gene,” even though genes play major roles in both homosexuality and autism.
I think Steve Hsu might have a few things to say about this.
We simply may have reached the point in some key scientific areas where we are working with levels of explanation that our human brains – even those of Nobel laureates – cannot handle. The top scientists might end up being people not who “know,” but rather who hold shadowy outlines of the truth in their heads.
He uses the example of the Wikipedia article on string theory, which was drafted and edited to be as accessible and intelligible as possible, and is anything but. A market metaphor:
Just as Adam Smith and Friedrich Hayek and Michael Polanyi stressed that a market economy evolves to the point where it is very difficult to understand the overall interrelationships of production, so can the same be said for many branches of science.
Cowen says that the age structure is lengthening in many fields, which is rough on innovation because:
… with some age we acquire wisdom but we lose some of the sharp conceptual edge and the willingness to overturn established ways. The innovators we end up with tend to be less revolutionary …
These developments may prove problematic for areas such as mathematics, which have relied heavily on prodigies.
Implications for management, allocation, and policy:
It will be increasingly hard for scientist administrators, philanthropists, and also government bureaucrats to get a handle on what is going on in a lot of scientific areas. The inscrutability of science will place an increasing burden on trust, whether it be trust in particular institutions, scientists, or reward structures such as the Nobel Prizes. How about trust in Google?
Part II: Machine Science:
Most current scientific research looks like “human directing computer to aid human doing research,” but we will move closer to “human feeding computer to do its own research” and “human interpreting the research of the computer.” The computer will become more central to the actual work, even to the design of the research program, and the human will become the handmaiden rather than the driver of progress.
We’re going to have to find a better word for this than handmaiden if we care about recruiting. Of course how many researchers simply throw data at Stata, Statistica, or Minilab and wait for some magic numbers to pop out and then spin an interpretation of them already? Or tweak a few variables to get there? At some point they learned about the logic of manipulating matrices, but then they forgot all that and learned to stop thinking and trust and rely on the machine.
An intelligent machine might come up with a new theory of cosmology, and perhaps no human will be able to understand or articulate that theory. Maybe it will refer to non-visualizable dimensions of space or nonintuitive understandings of time. The machine will tell us that the theory makes good predictions, and if nothing else we will be able to use one genius machine to check the predictions of the theory from the other genius machine.
Sounds like a mash-up of influences from Stephen Wolfram and Douglas Adams. Wolfram once pitched his company’s Alpha aspirationally as maybe becoming capable of just such analysis one day. And if one genius computer tells us the answer is 42, another genius computer will be needed to tell us the question, and perhaps a chain of genius computers needed to design the next level of genius computers.
The incentives for producing better science will encourage this broader unintelligibility … That’s the division of labor and complementarity, both of which can push scientific results away from general intelligibility once those genius machines enter the game.
For the public:
Still, as a general worldview, science will not always be very inspiring or illuminating. The general educated public will to some extent be shut out from a scientific understanding of the world, and we will run the risk that they might detach from a long-term loyalty to scientific reasoning.
In favor of what? (… hopes the answer isn’t “Islam” …)
We will see, more and more, the relatively mundane data-gathering sides of science. The bureaucracy and data gathering of science will be visible, as will be the magic of the devices we use. But that middle layer of knowledge – science as a general means for educated laypersons to understand the world through theories – will peak sometime in the twenty-first century.
Part III: Whither Economics?
The ultimate test of any theory is a market test. It doesn’t matter if your dog food is delicious in theory if the dogs won’t eat it. Do people in the business of building, say, planes for a living actually use your theory of aerodynamics, or do they ignore it and do something quite different?
If any field should understand this insight most deeply, it ought to be economics itself. So, the question is, do the dogs with options eat economics-theories brand dog food? Not really. They eat data. Lots of data.
In the last ten years there has been a big shift in emphasis and it has come largely from web companies, not from academic researchers. When web companies are figuring out their business models, and trying to market to their customers, they tend to use a lot of raw, relative unfiltered data. Quite simply, they do this because they can. Facebook, Google Amazon, and other companies have a phenomenal amount of high-quality information at their disposal, more than more academic economists are used to having.
Which is why it was such a big deal when Chetty got all that juicy, albeit redacted, IRS information. Unfortunately no one else gets to look at it to use it to rigorously scrutinize Chetty’s more suspect claims. The info-trove possessed by the big web companies probably puts whatever the IRS has to shame anyway.
And when they process this data, they go a relatively atheoretical route. They “crunch” the data, and we now have “Big Data,” as we’ve come to call it, as the next business revolutions, which refers to the use of statistics on the data generated by electronic communications.
These companies, in their approach to this data, are fairly suspicious of structural theoretical models. … they’re not trying to start with “the Jonesian model of why people use google,” …They go straight to the numbers and try to find power where they can.
Economics as a research area, in recent times, has been following the same path as these web companies: lots of data and relatively weak theoretical structure. Powerful data crunching, and careful data gathering, is pushing out theoretical intuition.
Pattern-finding algorithms + huge amounts of granular, intimate data = new results. But algorithms won’t be biased in favor of socially desirabiltiy, so, will we like those results? Clearly, given the war on noticing patterns, the usual suspects will hate some of those algorithmically noticed patterns. They will especially hate them because of the inapplicability of the usual accusations typically levied against human researchers. So they will probably try to pin blame on the coders somehow and, predictably, no one’s really going to check whether there’s anything wrong with the code, because obviously there won’t be anything wrong with the code. As a consequence, socially disfavored but also, you know, “true” results, will probably either be taboo open secrets, or they will simply have to be kept in house as proprietary information. We’ve already seen the beginning of this trend.
These programs will confirm some connections we already believe in, see come connections that we currently do not grasp, and perhaps generate some hypotheses that we do not suspect. Economics is not yet there, but perhaps in the next fifty years such endeavors will supplant the economist’s reliance on theoretical models.
What about understanding?
We will know how to feed the machines with data, and how to test them against each other, and we will know how to use their results. But at some point we will cease to understand all of the component parts of the science and we will cease to understand how the predictions are put together. On the machine will, in its own way, be able to encompass the entirety of the theory in its tests.
This makes me think of a metaphorical parallel to the Prime Conservative Insight, that presumptively one ought to be faithful, obedient, and deferential to traditional rules, cultures, and institutions, since these developed through an evolutionary process and we don’t – and perhaps can’t – fully understand their functions, their benefits, and the mechanisms by which they produce those benefits.
Anyway, how will all of this code-reliance change the character academic “discourse”?
Some notion of publication may still exist, but the important outlet for research will be in standardized, machine-digestible form.
I don’t know. That’s not very NYT-worthy now, is it?
That’s the single biggest change in economic science we can expect over the next fifty years. When it comes to “the new paradigm,” a lot of people are expecting the next Marx, Keynes, or Hayek. The changes to come will be more radical than that and they will challenge the very relationship that the scientist has to his or her craft of study. The real change will be the subordination of the individual scientist.
Back to economics:
Overall, the profession is producing more first-rate empiricists than before, yet theory hasn’t progressed much in twenty years or more. Theory is increasingly ignored.
He clearly wants to shout that something went fundamentally wrong in the development of modern, mainstream economic theory. But as for empirics:
We’re not far away from having a single de facto, more or less unified, empirical social science. In that social science, researchers invest a lot in learning empirical techniques and then invest some marginal energies in the simpler theories that surround their chosen field of study. Finally, they spend their research time looking for new data sets, or looking to create that data, whether by detective work or by lab and field experiments.
This keeps reminding me of Chetty’s Magic Dirt.
In addition to empirical researchers, another kind of specialist will specialize in understanding the machine results at the “meta analysis” level in order to interpret and explain them. Some of that will be for the consumption of specialist experts in the fields, but some will be for public consumption. You know, like Vox explainers, or Bloomberg view economist popularizers / journalists. Like Cowen. Or maybe like Scott Alexander for Psychiatry. But also like Paul Krugman and Noah Smith. Which … is not exactly good news.
These Freestyle researchers will be pioneering a fundamentally new way of “doing” economics and a fundamentally new sense of what it means to be an economist and indeed a scientists. They will earn good money and a degree of public game, and their numbers will multiply, even as their daily routines become increasingly estranged from the practices of normal everyday science in their fields.
At least for a while, they will be the only people left who will have a clear notion of what is going on.
… we are way, way out on the diminishing marginal returns curve, which means from now on, minor discoveries will take teams of geniuses years to produce even with huge amounts of resources. And since the simpler, easier-to-understand patterns probably have been uncovered already, many of these discoveries will barely be comprehensible, if at all, even to the most elite minds in some specialized sub-corner of a discipline. One won’t be able to lead even a talented novice to the coal face anymore. Maybe to the mine mouth, but the vein lies a mile or more beyond.
This chapter seems to me by far the weakest and the most detached from reality. Cowen is not describing our world, but an altogether imaginary one in which the enormous labor and resources poured into the official science since WW2 have been utilized in an ideal way towards genuine scientific advance. Whereas, of course, this has only created a bureaucratic system full of perverse incentives, whose output is overwhelmingly — without exaggeration, I’d say well over 99% — just make-work and nonsense.
In most fields, possibly the overwhelming majority of them, it just isn’t that hard to get to the coalface if one uses some minimum of common sense to sort out what’s really important. A very smart person with a good general overview of the field — something that can be acquired by someone who has talent, motivation, and opportunity for quality instruction no later than early twenties, possibly late teens — shouldn’t need more than a few months of focused study to get there in most cases.
And that’s for complex STEM subjects, where the distance to the coalface is undoubtedly the greatest. In “social science” we see the spectacles such as Sailer vs. Chetty, where an amateur blogger, applying only some straightforward common sense, can rip apart the supposedly cutting-edge work of an ultra-elite academic genius.
I had similar thoughts, but isn’t it somewhat surprising that someone like Cowen and in his position with his (apparently) intimate familiarity with the cutting edge output and literature of his field, would make claims like these?
With Cowen, it’s always difficult to tell what he really thinks, and what’s just the “anti-bait” and filler that may be silly and wrong on the face of it, but serves as sugar-coating for what he sees as the important points for which he wishes to open the minds of his high-status progressive readers.
Maybe he asked himself how his fundamental AOE points would apply to an idealized world of which most academics like to imagine they’re a part, and then concluded that such an account, however detached from reality, would nevertheless be a good way to make them appreciate these really important points.
Or maybe it’s simply that he is also drinking the same kool-aid. (Even Hanson often seems to me like his social calculus module is putting some serious barriers between his basic insights and their straightforward application to some of the uglier aspects of the modern academia.)
This final chapter focuses on his guesses regarding the political implications of these trends. It seems almost inevitable in a social democracy that with most wealth and productive activity highly concentrated in small fraction of the population that the temptation to use the state to grab and confiscate the surplus and redistribute it to clients in exchange for votes – but under some cover-story narrative and rationalization of just deserts – will continue to be completely irresistible. Maybe the only thing keeping up from a basic income is that it’s too basic if everyone gets it, and not just clients. Then again, we can always argue forever regarding just how generous it ought to be.
The forces outlines in this book, especially for labor markets, will force a rewriting of the social contract, even if it not explicitly recognized as such. We will move from a society based on the pretense that everyone is given an okay standard of living to a society in which people are expected to fend for themselves much more than they do now.
Depends what he means by “fend for themselves.”
I imagine a world where, say, 10 to 15 percent of the citizenry is extremely wealthy and has fantastically comfortable and stimulating lives, the equivalent of current-day millionaires, albeit with better health care.
A Complacent Class.
Much of the rest of the country will have stagnant or maybe even falling wages in dollar terms, but a lot more opportunities for cheap fun and also cheap education. Many of these people will live quite well, and those will be the people who have the discipline to benefit from all the free or near-free services modern technology has made available. Other will fall by the wayside.
The Dire Problem and Kahn Academy and Joyboxes and Opioid epidemics. And welfare and “disability” and “unemployment insurance”. And everyone who can struggling to distance themselves and their children from “waysiders”.
That’s for developed countries. What about in developing countries?
Since the self-motivated will find it easier to succeed that ever before, a new tier of people from poor or underprivileged backgrounds will claw their way to the top. The Horatio Alger story will be resurrected, but only for those segments of the population with the appropriate skills and values, namely self-motivation and the ability to complement new technologies. It’s in India and China that the risk of a new middle and upper class is reflecting this trend most clearly.
IQ+CUTS+Tech Complementarity. The Horatio Alger narrative of bootstrapping success through persistent and determined exercise of bourgeois virtues via a combination of self-motivation and merit / talent / gifts seems to depend for its popularity on the developmental context of the economy at the time. Also on there still being a lot of unharvested peasant talent that hasn’t yet been gleaned from the hinterlands and thrown into the IQ shredder cities.
But as many have already pointed out, some of the big problems with efficient meritocracies is that eventually the hinterlands get severely creamed like fresh milk. Not just drained of brains, but of the whole natural local aristocracy capable of judicious community leadership, and willing to perform it in a way where they can use their status to advertise for more civilized standards of behavior. Furthermore, those who don’t succeed are left without any socially acceptable excuse to explain away their pathetic condition, and so must contend with the psychological burden – and the hit to ego and self-esteem via sociometer – of knowing definitively that they are real hopeless losers, and of knowing that everyone else knows that about them too. This seems like a recipe for all kinds of trouble, especially the channeling of these negative emotions into potentially explosive resentments. I guess that’s what the tranquilizers will be for. Not exactly what they meant by, “insure domestic tranquility.”
This framing of income inequality in meritocratic terms will prove self-reinforcing. Worthy individuals will in fact rise from poverty on a regular basis, and that will make it easier to ignore those who are left behind. The wealthy class will be increasingly self-motivated, will be larger over time, and -precisely because we are selecting ever more for self-motivation – will have increasing influence. It is their values that will shape public discourse and that will mean more stress on ideas of personal ambition and self-motivation. The measure of self-motivation in a young person will become the best way to predict upward mobility.
Yikes. Highly motivated young people aren’t always motivated to do good things.
But while Cowen is emphasizing cases of rising from poverty, I imagine that in first world countries most of those stories will be, well, stories of first, maybe second generation immigrants. The two prior generations of pretty much my entire extended family were just such stories.
But on the domestic scene, the Coming Apart trends will tend to produce the opposite observation of hardening castes, and more current success descending from past success. Unless by some magic there is more widespread adoption of belief in the reality of assortative marriage and the genetic inheritability of IQ and other marketable skills, then it’s not going to seem very meritocratic, and more like a privileged aristocracy acting as a country club, cartel, and conspiracy for keeping their secret stash all to themselves. Indeed, we see some progressives trying to spin this narrative, seeing what traction they can get from it. And once again one is reminded of Chetty’s bogus research into the magic dirt theory of intergenerational mobility.
Speaking of Charles Murray.
We’ll also see a lot more of some of the hypocrisies common today. For instance, it’s pretty common to hear tenured economics professors at establishment schools espouse the relevance of liberal democratic policies, such as the social safety net. These same individuals, if asked to explain their choice of academic hires, or their choice of which students to push in the job market, often respond in rather harshly meritocratic terms. If a graduation PhD student does not have his job market paper ready by his fifth year of study, it’s because “that student didn’t have a strong enough work ethic,” or something like that. That same professor will be very shy to apply the same kind of rhetoric to discourse about the safety net, for fear of sounding like a non-liberal crisis such as, say, Charles Murray. When it comes to a lot of values issues – and what people really believe in their daily lives – the gap between conservatives and liberals isn’t as large as it might first seem.
It’s pretty obvious how one reconciles and rationalizes away this apparent ‘inconsistency’, which is to say that the meritocracy frame is valid for intra-class (or iso-privileged) comparisons, but not for inter-privilege comparisons.
But anyway, here we have a good illustration of “political consistency vs. analytical consistency”, belief signaling with personal hypocritical deviation, and maybe some Conquest’s First Law.
What does that mix of values mean for actual social choices? We’ll pay for as much of a welfare state as we can afford to, and then no more.
Well, how is this different from the social democracy we already have?
Part II: The Fiscal Crunch
Everybody already knows many governments are going to need to raise taxes and lower spending, and will probably avoid doing so until absolutely compelled by near-crisis conditions. See: Illinois 2017, for just one of many good examples this year. Low wage growth for most current taxpayers and all that “waysider welfare” is just going to make this worse / accelerate the arrival of crisis moments. And if bond rates for new debt ever tick up faster than inflation and economic growth, then that would makes the fiscal situation even worse.
So, “Who Is Going To Pay?” Well, who’s not going to pay?
It’s a common view that “the top 1 percent” can or will fund these forthcoming expenditures by paying higher taxes. I don’t think that is likely, for a few reasons, even though I do think the wealth will end up paying somewhat higher taxes. But why can’t they pick up the whole tab? … The wealthy will grow in numbers, and that also means the wealthy will grow in influence. Imagine that today’s millionaires comprised 10 percent of the citizenry; that would make for an extraordinarily influential and politically potent group, much more so than the wealthy today. Can you imagine that group funding the entire future by raising taxes on itself? I don’t see it.
The very rich also have ways to protect their money, and the wherewithal and incentive to hire clever people with the special knowledge of how to do so. That’s always been a good business for lawyers specializing in that field.
He also uses Laffer Curve logic. And there is a discussion of tax incidence – where any attempt to tax high earners will just get passed on to other parties.
Cowen says it’s too hard to cut the big entitlement spending programs too, because the geriatric vote too consistently for them.
To balance the budget right now through spending cuts, we’d basically have to come close to getting rid of Medicare, Medicaid, and Social Security altogether. And I’m talking about the complete elimination of those expenditures, not shifting them into somewhere or something else.
I haven’t looked at the most recent numbers – or what they were when he wrote those sentences – but that does seem like a bit of an exaggeration. The expenditures for these programs combined are much larger than the budget deficit, so maybe he means it in some other way.
I submit that the aggregate amount of aid given to the elderly, the needy, and other groups is unlikely to decline, whether we approve of that outcome or not. … But the total expenditures on the health of the elderly will rise both in absolute and per capita terms, not fall. This is the general trend of Western societies since the late nineteenth century and no reformers, including Margaret Thatcher and Ronald Reagan, have really taken on entitlement spending through government and beaten it back. It’s simply too popular.
You know, it’s funny. The word “populist” gets thrown around a lot these day. A Lot. And nearly always in some derogatory sense. You know what’s really ‘populist’? Entitlements, that’s what. “Keep your government hands off my Medicare!” But no one uses it as an epithet and means that, now, do they?
In percentage terms, relative to outstanding need and vociferous claims, the altruism of the public sector will have to fall. … It’s not about ideology; it’s a question of making the numbers add up.
I am forecasting a few particular changes, starting with the most obvious and ending with the least obvious:
1. We will raise taxes somewhat, especially on higher earners.
2. We will cut Medicaid for the poor (but not so much Medicaid for the elderly) by growing stingier with eligibility requirements and with reimbursement rates for Medicaid doctors, who will impose queuing on program beneficiaries.
3. The fiscal shortfall will come out of real wages as various cost burdens are shifted to workers through the terms of the employment relationship, including costly mandates.
4. The fiscal shortfall will come out of land rents; in other words, some costs of living will fall as people begin to live in cheaper housing.
5. We’ll also pay off growing debt by spending less of our money on junk and wasteful consumption.
That’s operating from a pretty simple theory of US politics that says “old people get their way.”
Gerontocracy, kind of. More like Geron-Satisfice-ocracy. (Is that already a thing?) Old people will go along with whatever so long as they get their benefits first and are otherwise left alone.
He says financially-stressed workers will have to move to cheaper places to make ends meet, and he uses Texas as the prime example, which is kind of a crappy place to live with, “C-grade public services,” but has, “… very cheap housing and a decent record of job creation…”
He attributes cheap housing to laxer zoning rules, naturally, but not with as much unqualified celebration as it typical in libertarian circles.
For instance, Houston doesn’t have traditional zoning. You might find an office tower, a used-record store, and a whorehouse all right next to your home. Houstonians live with that, and since home prices are reasonable the relatively wealthy can insulate themselves from the less pleasant consequences of mixed-use neighborhoods.
People like having money more than what the government would give them in exchange for their taxes.
Since there is considerable net in-migration to Texas, I conclude that a lot of Americans would rather have some more cash than better public services. Not everyone wants that bundle, as you will see if you poll the wealthy upper-middle-class residents of Brookline, Massachusetts or my own neighborhood in northern Virgnia. Nonetheless, on the whole, we as a nation are moving in that direction.
Are we now? I’m not so sure. What’s going on in Austin lately? What’s that Houston mayor up to lately? We’re about to see what happens when Texas gets demographically transitioned and more and more areas go increasingly Blue. And what’s Land call it again, “ruin voting”?
Which are the states with the highest-quality public services? On the basis of measured expenditures this would typically be California and state in the Northeast, but in general those regions are seeing outmigration.
I lol’d. Sometimes Cowen just goes way over the top to tip off his readers to the Straussian signal. The “basis of measured expenditures” is about the dumbest basis imaginable for measuring the quality of public services. California!? Where retired fire chiefs of tiny nowheresvilles are getting six figure pensions for life? The highest quality! Jesus.
Anyway, the point is, if people choose cheaper, uglier, messier, more chaotic, less-taxes, lower quality public service Texas, then the future is going to look a lot like .. well … Latin America. With lots of actual Latin Americans too. Favelas, tiny homes, delicious beans with freshly ground cumin, all that stuff.
When I visit Latin America, I am struck by how many people there live cheaply. In Mexico, for instance, I have met large numbers of people who live on less than $10,000 a year, or maybe even on less than $5,00 a year .. they have access to cheap food and cheap housing. They cannot buy too many other things. They don’t always have money to bring the kid to the doctor or to buy new clothes. Their lodging is satisfactory, if not spectacular, and of course the warmer weather helps.
The altitude too. The point is, if the choice is between that and barely subsisting in some big Blue US city, you might prefer and choose the Mexican lifestyle, if you could live in the US. Texas beckons as a future “Mexico+”.
We could designate reservations for these untouchables and build them cheap favelas far, far away from gentrifying urban neighborhoods. For instance, in Texas. Texas is, like, super far away from New York and DC and San Francisco!
We also would build some makeshift structures there, similar to the better dwellings you might find in a Rio de Janeiro favela. The quality of the water and electrical infrastructure might be low by American standards, though we could supplement the neighborhood with free municipal wireless (the future version of Marie Antoinette’s famous alleged phrase will be “Let them watch internet!”). Hulu and other web-based TV services would replace more expensive cable connections for those residents. Then we could allow people to move there is they desired. In essence, we would be recreating a Mexico-like or Brazil-like environment in part of the United States, although with some technological add-ons and most likely with greater safety.
I think that parts of Texas are already kind of like this theoretical Mexico+. It’s been a while since I visited, and that was Austin.
Most people will be horrified at this thought. How dare you propose we stuff our elderly into shantytowns? Maybe they are right to be upset, although recall that no one is being forced to live in these places. Some people might prefer to live there. I might prefer to live there if my income were low enough.
There’s your basic income, similar in some ways as for the Indians. You can always do nothing and still get by ok on the reservation. It’s sad and depressing and depraved and socially pathological in every way, but it’s always an option if you assess that everything else is even worse. If you don’t want to live there, then you’ve got to work for your daily fry bread.
If we don’t build the nice shantytowns on purpose, they’ll still emerge through economic evolution anyway. Not so different, just not as nice as they could be.
El Paso is America’s twenty-third largest city (using data from 2000) and it would jump to fifth largest if we combined it with the population of its sister city, Ciudad Juarez, right across the Mexican border. You can think of it as one consolidated city with a very large attached shantytown.
I get what he’s saying, but you know … to nitpick, the shantytown is run by a totally different government, and there are border controls, there’s influences of NAFTA and smugglers and day trip medical and pharmaceutical tourism, and .. well, maybe thinking about it that way doesn’t make 100% perfect sense. I’ve been to Fort Bliss, and Juarez has been pretty much permanently off limits for military for years, and for good reason.
Indeed the shanty is essential to the success of El Paso. El Paso lives off of the manufacturing across the borer, as it lost its own manufacturing base some time ago and it has substandard levels of education.
“Lives off of.” How does El Paso live off manufacturing in Juarez, exactly? Do the workers in Mexico send remittances to the American relatives across the border? Ordinary trade connections in economic relationships don’t qualify as “lives off of” without serious violence to the language.
The city also benefits from an army base, from border-related law enforcement efforts, and from the drug trade. Howard Campbell, an anthropologist, notes that El Paso is parasitic off of Juarez rather than vice versa. El Paso has flourished by hooking up with an adjacent neighbor with much lower rent and much lower quality infrastructure. Despite all the problems that can cross the border, and despite Juarez being one of the world’s drug-cartel and murder capitals, few people in El Paso wish for Juarez to go away.
I doubt anyone actually asked them. I’m just going to let this pile of nonsense speak for itself. Cowen is just operating in provocation mode.
He then mentions Berlin as a low-rent city because of overbuilding when builders thoughts it might become the business capital of Germany.
It is easy to rent an acceptable apartment in a non-peripheral part of Berlin, not too far from a subway to streetcar stop, for a few hundred dollars a month. Food, too, is much cheaper than in the rest of Western Europe – cheaper than in most of the rest of Germany even. There are many thousands of people in Berlin simply living, on low rent, to “get by.” It’s the ultimate slacker city.
You may have heard, it’s a little more crowded now.
The point is, poorer people are simply going to have to move to where they can afford real estate, which isn’t anywhere near the expensive hubs. Unless the government subsidizes them to do so, one way or another. Which it is doing actually, and which it might continue to do, but I guess we can infer that Cowen thinks it won’t be able to afford to do so for much longer. I’m not so sure.
We’re going to get lower land prices one way or another. Not Manhattan, not West LA, not Fairfax Country Virginia and not the whole country, but some parts of it. Some version of Texas – and the some – is the future for a a lot of us. … People will respond to stagnant or shrinking entitlements by moving to cheaper areas.
Not giving out local COLAs for any benefits would help a lot. COLAs for employees are basically essential recruiting and retention tools, but the case for adjusting welfare benefits is weaker. Are there not <strike>workhouses</strike> reservations in Mexico+, Texas?
What about consumption patterns? Will we just get used to being poorer and consuming cheaper stuff? Like the now infamous chalupas and beans with cumin?
There is one final way we will adjust to uneven wage patterns and that is with our tastes. Many of society’s lower earners will reshape their tastes – will have to reshape their tastes – toward cheaper desires. Caviar is an expensive desire and Goya canned beans is a relatively cheap desire. Don’t scoff at the beans: With an income above the national average, I receive more pleasure from the beans, which I cook with freshly ground cumin and rehydrated pureed chilies. Good tacos and quesadillas and tamales are cheap too, and that is one reason why they are eaten so frequently in low-income countries.
… Citizens faces with financial pressures will shift into cheaper consumption, and a lot of them will do so without losing very much happiness or value, precisely because there is already so much waste in what they buy.
Seems legit. At the unprepared bulk commodity level, tasty calories are ridiculously cheap with not too much preparation. For people living in expensive areas, it seems to me that halving or, hell, tripling, food expenditures wouldn’t make much difference at all to the monthly budget when compared with rent (and often, student loans.) But on the other hand, for those who live in cheap or subsidized housing and spend a much larger portion of their budget on food, there is a lot of room at the bottom, especially if your time isn’t very valuable and so you can use it to cook. Still, it’s hard to resist the conclusion that almost all the gains to be had are in real estate, and everything else pales in comparison.
Here’s a bit of ugly reality:
This process of economizing won’t always go so well when it comes to poor women. A recent Pew Research Center study examined exactly who in modern America falls out of the middle class, and it found that women who are divorced, widowed, or separated are an especially vulnerable group. And children don’t help single mothers’ incomes. Taking care of one’s children can be thought of as a very expensive preference but it is a preference that, for a lot of people, is not going away. Younger women in the lower end of the income distribution will probably be some of the biggest losers, especially is they have a strong “baby lust” that induces or compels them to have lots of kids early in life. Many of these women will also find it harder to move to cheaper areas with lower-quality infrastructure because they may still desire good school for their children, especially if those kids are not self-starting learners from the internet. To top off all these problems, the desire for cheaper preferences and lifestyles may induce more lower-income men to abandon their children or at least to scale back financial support, a development that is extensively cataloged in conservative critic Charles Murray’s book Coming Apart.
That’s a lot of “may” for something that has been extensively studies and cataloged. It’s not exactly “Whoa, that future sounds scary, I sure am glad we don’t have to deal with those problems now.” These have been obvious, universally known, and growing features of American underclass reality for half a century at least. The Moynihan Report was in 1965! Oh well, that’s how these books get published these days I suppose.
Cowen goes on to mention that the marginal value of healthcare dollars is pretty low, so cutting back may not hurt people very much. And to the extent it does, as far as anyone can tell, the negative effect could be balanced by the positive consequence of a few behavioral changes, e.g., in diet and exercise.
Not everyone will respond in this way. We’ll end up with a society where the people with decent self-control win back a lot of the lost health gains by better behavior. The people who don’t have good self-control will lose out much more. They’ll lose a chunk of their health care and they won’t respond by getting on that exercise bike.
Personal qualities of character such as self-motivation and conscientiousness will reap a lot of gains in the new world to come. We can already see this in the numbers. The individuals falling out of the middle class are more likely to be divorced, to have low levels of formal education, to have low test scores, and to have a history of drug use.
Part III: The Politics of the Future
Politics follow demographics. And demographics in America means a lot of old whites, and not many more young people, though the young we have now come in every kind.
As I’ve mentioned, right now about 19 percent of Florida is over the age of sixty-five. By 2030, 19 percent of the United States will be over sixty-five years of age; in other words, we’ll be like Florida in terms of age structure. We’ll then get older yet.
Sounds like parts of Japan. Geezers everywhere.
We aren’t going to get revolutions.
It seems that, whether we like it or not, increasing inequality and growing domestic peace are compatible. Very often I read warnings about how income inequality will lead to a society where the poor take by force what they cannot earn in the marketplace. Yet these predictions run aground on the simplest of empirical tests, namely crime rates.
Well, that was 2013, and crime ain’t exactly fallin’ as fast as it used to be. But that’s got little to with economic inequality per se. Anyway, the left is denying that crime is rising, and it’s hard to do that while simultaneously saying inequality is rising, and rising inequality should make crime worse. Political consistency vs. analytical consistency, again.
There are many other historical periods, including medieval times, where inequality was high, upward mobility was fairly low, and the social order is fairly stable, even if we as moderns find some aspects of that order objectionable.
I wonder if this “threat of revolution” argument isn’t a substitute for actually making a good case for a feasible reform. I’ve very often heard commentators from the Left suggesting that if we don’t “do something” about income inequality, citizens will take matters into their own hands. There is a vague insinuation of a threat of violence, yet without any endorsement (or condemnation) of that violence. The commentator or writer doesn’t want to suggest that the violence is in order, yet still wants the rhetorical force of having that violence on his or her side of the argument, as a kind of cosmic punishment for the objectionable inequality.
Something like that, yeah.
The better guess is that Americans will become more conservative, now returning to both the political and literal senses of that word. They will become more enamored of low or falling taxes, whether or not such tax rates prove possible to maintain.
Which is why Ted Cruz won all those primaries. I suppose he could respond that maybe it just hasn’t happened yet, just you wait and see. But also, taxes have been going up. In Kansas, Illinois, Oregon. Where are all the new tax cuts?
They will look more toward local communities and tight local bonds, to protect themselves against economic risks. Unlike the predicted breakdown in the social order, these trends are already significant and observable in today’s America.
Political conservatism is strongest in the least well-off, least educated, most blue collar, and most economically hard-hit states. If you doubt it, know that of 2011, the politically conservative states are, as measure by self-identification, Mississippi, Idaho, Alabama, Wyoming, Utah, Arkansas, South Carolina, North Dakota, Louisiana, and South Dakota. As Richard Florida puts it, “Conservatism, more and more, is the ideology of the economically left behind.”
Those states have becomes outposts of Tea Party support. Their electorates are not out there leading the charge for higher rates of progressive taxation or trying to revive the memory of George McGovern. The most liberal areas tend to be urban or suburban, with lots of high-earning professionals. My own residence – in Fairfax County, Virginia – was strongly conservative in the early 1980s when I first lived there. … Circa 2012, Fairfax County is now in per capita terms the wealthiest county in the United States. It broke cleanly for Obama in the 2008 and 2012 elections and it is somewhat more Democratic than Republican in terms of party support.
Without getting too deep into these distracting weeds, this is also missing quite a bit of the local story and context, especially with regard to demographic change and certain related local policies. But with that paragraph doesn’t it seem like Cowen was this close to predicting something or someone like Trump? Not close enough though.
If we extrapolate these trends into the future, we can expect the higher earners to identify with the values embraced by today’s moderate Democrats. They will believe in progress, diversity, and social justice, although they may not be huge fans of radically progressive taxation.
However, they will be huge fans of the latest radically progressive gender identity orthodoxy. Or else.
Some of them will be “small L libertarians,” but those libertarians will like the same jokes and TV shows as the moderate Democrats among the high earners.
And go write articles for the Niskanen Center and The Atlantic or something.
The lower earners will be split into two groups, the more extreme conservatives versus the individuals who receive transfers from the social welfare programs supported by the moderate Democrats. I’m not suggesting that they will cynically vote to line their pockets, rather that the moderate Democrats will offer a worldview that embraces those individuals and offers them higher status and respect, thus winning their political loyalties. The more extreme conservatives will embrace religion and nationalism to a higher degree.
I think it’s clear we’ve passed “Peak Embrace Religion” for social conservatives, unfortunately. The implosion continues apace, and Dreher’s The Benedict Option was, if anything, published far, far too late. I think Cowen was just way out of his competency window here.
He said we won’t get revolution because most envy is local – the guy down the street, not the billionaire on Wall Street.
Sometimes I wonder why so many relatively well-off intellectuals lead the egalitarian charge against the privileges of the wealthy. One group has the status currency of money and the other has the status currency of intellect, so might they be competing for overall social regard? And in that competition, at least in the United States, the status currency of intellect is not winning out. Perhaps for that reason the high status of the wealthy in America, or for that matter the high status of celebrities, bothers out intellectual class most. That intellectual class, however, is small in number, so growing income inequality wont by itself lead to political revolution along the lines many intellectuals have imagined.
Getting close to the end now.
We might even look ahead to a time when the cheap or free fun is so plentiful that it will feel a bit like Karl Marx’s communist utopia, albeit brought on by capitalism. That is the real light at the end of the tunnel. Such a development, however, will take longer than I am considering in the time frame of this book.
It’s not even that good of a light, and then he just snatches it away. Dismal science indeed!
In the meantime, get ready. The basic look of our lives, and the surrounding environment, hasn’t been revolutionized all that much in forty to fifty years – just try viewing a TV show from the 1970’s and the world will seem quite familiar.
Tell that to my kids. They’ve never seen a pay phone. You want to feel old? Wait until one your kids asks you with genuine confusion why we say “hang up” to end a call.
That’s about to change. It is frightening, but it is exciting too.
It might be called the age of the genius machines, and it will be the people who work with them that will rise. One day soon we will look back and see that we produced two nations, a fantastically successful nation, working in the technologically dynamic sectors, and everyone else. Average is over.
Chapter 12 does seem to be the Cheap Chulupas plan. Its obviously monstrous. The line between “I predict this will happen” and “I want this to happen” is always rather thin with that group. At least the World Controllers in Brave New World lamented the flaws of their system. The GMU crowd seems to be excited by them.
It’s monstrous if you assume — as most people do on both left and right — that the mid-20th century golden age of the common man would just continue indefinitely unless the government actively ruins it, and that we only need to be “smart” in some quite obvious ways to restore and keep it.
However, if we discard this assumption as unjustified (which it clearly is), and look at the range of realistic possibilities, the Cheap Chalupas scenario starts looking not that bad. This is not to say that stupid and crazy government measure aren’t making things worse, of course. But it’s a big mistake to take for granted that the 20th century-style universal prosperity is just what happens by default, or that its continuation would be possible for even a very good government.
Cowen’s Future sounds like an elaboration of what we have called Brazilification; although he makes it a function of economics instead of demographics. He also assumes no big decline in the quality of governance, massive ethnic and plain family nepotism and the hinterlands organizing themselves along violent gangs, among all those other little things we all love about Brazil.
I think it’s obvious that actual Brazil is more likely than the milder Cowenzil. And that’s assuming that Islam is contained and that 2 billion Africans don’t invade Europe and North America.
All you really needed to maintain something mid-20th century is a different immigration policy and mildly HBD aware public policy. Some of it isn’t even public policy, how about just not celebrating degeneracy as a virtue in the cultural arena.
It seems to me that you can find societies around the world that adopt parts of this platform, and that they don’t exactly represent any kind of radical program with no history of successful implementation.
I don’t know if I’m convinced that things were inevitably going to go this way. While we are all aware of the factors leading to the current situation, if we are to believe that even mild tweaks to the timeline, most of which just would have been preserving the status quo at the time, were in fact impossible, then I don’t see how we can hold out any hope for any positive changes going forward whatever, including even maintaining the current situation.
A counterargument is that lots of low-skill immigration has bought us a little bit more time since it kept the capital-labor substitution rate lower than it would have otherwise been, because lots of cheap, reliable labor (and externalizing many of the costs of that labor pool on the taxpaying public) undermined the incentives to develop and deploy automation technology. You can think of a lot of “cheap labor” (whether native, immigrant, or outsourced) as a kind of rival “technology”, and if the price is low, then actual technology won’t be competitive until algorithmic efficiency rises and hardware costs fall, which takes time. One wonders where things would be today had the US stuck with the 1920’s-1940’s immigration policies, or gone with a Japanese or even Canadian/Australian selective immigration model.
On the other hand, I’m not sure how much that counterfactual would have goosed technological development in this area faster than it happened. I’m guessing a little, but not much, but it’s hard to know. Still, eventually we’ll be living in a country full of robots. The Japanese will just have robots and Japanese descendants. The US will have robots and descendants of all the newcomers too. The newcomer descendants will vote, but the robots wont.
Robots aren’t a problem. Surely, people would prefer to live the way they did in the 1950s while having a robot do the hard labor they used to do. My Dad hardly liked waking up at 3am to do hard labor.
When my Dad retired he simply became a more complete human being. He produced musicals, developed athletic talents, got more involved with community and family, and took all his hobbies to a new level.
Let’s say robots really could make all labor obsolete. Imagine lots of well behaved white people getting a cut of their robot produced UBI to make ends meat while they live in their 1950s style communities.
I don’t view robots as being the primary cause. Bad genes and bad culture is. The third world was still a shithole before robots came around, and will still be a shithole after robots.
Well, I can indeed hear Lennon’s “Imagine” playing in the background while reading this!
For even the most civilized population imaginable, if you think its average (let alone below-average) people could live on handouts and nevertheless avoid descending into complete degeneracy and savagery, I think you’re completely out of touch with reality.
Recently updated pseudoerasmus post – tech and inequality – related to this thread topic.
Ironically, as production becomes more brutally efficient with labour-saving technology, consumption becomes more ‘inefficient’. The hallmark of consumption by the rich has always been its labour-intensiveness. Think of aristocratic dining halls as recently as the Gilded Age, with one liveried footman for every guest at the long table in the dining hall.
That’s why ‘hand-made’ has snob appeal. Bespoke fetishists may think of it as “valuing timeless artisanal quality”, as does one London financial journalist who apparently has not only suits and shoes custom-made by hand, but also socks, neck ties, and (!) pocket squares. (When those silks stick out of the breast pocket, woe unto those rolled edges sewn with plebeian machine-neatness…) This tailor-blogger with a cult following makes suits by hand, or ‘deconstructs’ famous brands, and blogs about every lovely stitch. But in reality such sartorial Epicureanism is about deriving more and more marginal utility out of sillier and sillier quality ‘improvements’. And such things point to the niche consumption fantasies of the merely upper-middle-class.
The “dream of the 1890’s is alive” in the era of LARPing as a Victorian artisan, servant, or dandy. In the future everyone is Beau Brummell for 15 minutes on instagram.
I would not have used the phrase the “emptiness of life” to make the case about class-based status-signaling conspicuous consumption. See what you think about the rest of his or her argument.