Followup to The Great Stagnation, AiO takes the same format awkwardly straddling the territory between overgrown Marginal Revolution blog posts and full-length books (AiO can easily be read in an afternoon and could be edited down further without much loss). AiO rehearses some of the background of TGS like the stagnation in median incomes and wretched income growth for most educational brackets. Americans, in 2013 and 2016, feel tremendously insecure; the absolute standard of living may be higher than before, but an iPhone doesn't pay the bills, and YouTube doesn't replace having a sense of self-respect or a stable job.

The Autor 'wage polarization' thesis argues this is due to the economy splitting between garbage jobs paying low-wage for unskilled but currently un-automatable jobs, and highly skilled and productive jobs, which benefit from globalization & technology. The unskilled and automatable jobs have been increasingly eaten by outsourcing to China or by technology (Cowen cites robotized factories, Netflix, dating sites, crime-predictive software for policing; he is skeptical in chapter 9 that outsourcing is the majority contributor to American trends). For the latter, technology & capital 'complement' the highly skilled, enabling them to produce ever more value (which is where their increasing salaries are coming from). This leads to some fairly dire forecasts: the banana republicization of America, with a self-regarding meritocratic class of wealthy white-collar workers continually concentrating into the metropoles and wealthy suburbs with their servants, leaving in the hinterlands the working poor, and the nonworking poor.

What is this complementation that robots or AIs help with? Financial trading and investment, technology tasks like enabling a Google-scale titan to run without collapsing instantly, drone strikes and organizing interrogation & imagery to decide who to drone strike, or just in general management to efficiently organize and run all the highly-paid specialists and keep them on track towards goals. More ordinary people get shut out; they cause too many problems, there's too much overhead and inefficiency in trying to use them, they hold up deadlines or spit in the food & post the video to Facebook. Such zero-marginal product workers can't be usefully used by specialists. Cowen finds himself perplexed to how he would use a person to help him even at a wage of $0:

As a professor, I am given a research assistant each year. Over the last twenty or so years, I have received some extraordinary assistance from some very good workers, students, and eventually, peers and coauthors.
About once a year I receive an offer, usually by email, from someone who wants to work as my research assistant for free. Typically the offer is accompanied by a resume, and for the most part these resumes appear quite good. The emails sound reasonable and friendly.
I turn such offers down. I don't think the applicants are lemons, but still I find that one research assistant is for me the right number, at least if I have a good one, as is usually the case. Even when it comes to the assistant whom I have the time to manage, I am most of all concerned about having a conscientious person at my disposal.
The work with an RA is basically a team relationship, and the core problem is that I don't have the time to build another team, even if it doesn't cost me any money upfront. I don't have the time to work with and manage another person. To put this point in a broader business context, until another good manager is hired, there is no point in employing another assistant. It's the manager who is the scarce input, and that is one way to think about why managerial salaries have been going up so much. Managers play a role of growing importance in coordinating complex, large-scale production processes.
...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 toward increasing importance. And, finally, it is why managers are being paid more.

(As AiO is fairly light on citation and referencing for a book advancing such broad theses, I think maybe Cowen should try to figure out how to manage more than one research assistant.)

Cowen's central case-study of this complementation is chess, and Advanced Chess in particular: a human playing chess with the assistance of grandmaster-level (and not long after its founding, super-grandmaster level) chess AIs, which began in 1998 at Kasparov's proposal. Cowen is an avid chess player, and these parts of the book are by far the best part of it. He describes the rapid progress of chess AIs after Deep Blue and the consequences for human chess playing of the availability of superhuman chess AIs. The chess AIs can see so far past the humans that Cowen, watching two play each other in a match and able to see each's evaluation of their winning chances by using his own chess AI to follow along, became certain that Stockfish would lose despite the evaluations insisting it would win, because Stockfish was in just too horrible a position; but as the inhuman moves pass, suddenly a Stockfish win started to look not so implausible, and by the end, Cowen could confirm with his own AI that the evaluations from almost 30 moves before were correct. Cowen notes that even grandmasters have difficulty understanding, after the fact, the moves that the chess AI play and why they work despite being apparently insanely risky and chaotic - paradoxically, though the best chess ever played is being played now in computer chess tournaments and chess AIs are arguably approaching perfection, humans have hardly any interest in playing, watching, commentating, or analyzing those games! Optimal chess moves, apparently, often strike benighted humans as ugly and risky, for all that they are the correct moves. (One thinks of what the Go players said about some of AlphaGo's moves during the Lee Sedol match.) What do 'AI moves' look like in life, or dating, or business negotiation, Cowen wonders? It might look like matching up people who are apparently antagonistic like conservative men and liberal women, but who might work out well anyway (Cowen cites one Match.com demonstration of a black/white couple where each violated the other's 'requirements' for a match but they married anyway, and his own marriage through a dating site to a liberal women.)

However, as astoundingly excellent as chess AIs playing each other are, as of Cowen's experience before the 2013 publication, a few humans are able to provide some sort of edge, overriding the chess AI to make a better move, and win. Oddly, this does not apparently require one to be a grandmaster or even a master chess player, but some sort of instinctive mechanical sympathy based on having an idea of where the chess AI is 'weak' and watching the evaluations in realtime (along with better preparation like gathering large chess game databases); indeed, being a GM may be a liability, as at least two GMs, Nakamura & Naroditsky, appear to have harmed or at least not helped their chess AIs with their lack of deep humility. (As chess AIs show, GMs arguably make mistakes on almost half of their moves.)

Cowen (as well as some other authors in 2013 like Clive Thompson) takes Advanced Chess as an optimistic paradigm for technological changes: it need not lead to unemployment if people can learn the skills which render them complements to new technology, instead of being substituted. One of his primary solutions is MOOCs and online education. I'm not sure MOOCs are so positively regarded in 2016 as they were in 2013. And like most authors who present education as a nostrum Cowen also doesn't explain why we would expect more education to solve anything when the existing steep education/income penalties/correlations have not managed to motivate the general population. Computerized education has been great for chess education, certainly, with grandmasters minted ever younger; but that didn't reverse Deep Blue's victory.

I think Cowen knows that MOOCs and other band-aids aren't going to reverse these trends, and the Advanced Chess example is telling: very few people can contribute to Advanced Chess, and the very best Advanced Chess players are adding ~100 Elo points, or a few % towards victory. 100 Elo points is not much. It's about as much as chess AIs improve in 2 years. At what point will Advanced Chess stop 'being a thing' as the chess AIs will have become so good that Advanced Chess players can no longer make a discernible positive contribution? Oddly, I'm having a very hard time figuring that out. Advanced Chess is not mentioned much online after 2013. Some extrapolating suggests that Advanced Chess may already have become moot in 2013, and if not then, is probably finished by 2016; so at the most generous, Advanced Chess could be said to have only existed 1998-2016 (so 18 years, hardly enough time for a kid to grow up), and then only for the tiniest fraction of the population.

So he finishes up pessimistically with forecasts of current trends: the American governments, federal/state/local, are going to face the anvil of healthcare inflation and unfunded Medicaid/Social Security promises. These programs are politically untouchable because old people know what side their bread is buttered on, so they will paid out, one way or another. Which will involve systematic rises in taxation and decreases in services. What does the lower half of the polarized economy do to cope with this? They will have to flee to jurisdictions with smaller governments and less taxation and less goldbrick regulation of housing jacking up rents, however unpleasant such places are, like Texas (but which nevertheless has constant inflow of migration, compared to California). American standards of living will decrease: beef burgers will be replaced with bean burritos, houses will downsize. Alternately, this inevitability of lower incomes could be embraced and deregulation and reductions done deliberately rather than implicitly: "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." This constriction won't be as bad as it may sound. Just as most healthcare expenditures in the USA are wasted so getting health insurance doesn't make much of a difference to health, many Americans (rich or poor) have extravagant spending habits (consider who buys all those lottery tickets and tobacco): "The bad news is that there is a lot of waste in American consumption-massive amounts of waste, in fact. Everyone has their favorite story about what the other guy spends his money on and could do without. But also the good news, oddly enough, is that there is a lot of waste in American consumption. Citizens faced 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." I could hardly disagree. If I had a buck for every boat or in-ground pool I've seen people pay a fortune for and then never use, or use once a year, I could buy a bundle of burritos; or not take even a few seconds to shop around online; and one can go to Walmart and simply watch people shop as they buy the smallest unit grocery (despite having a large family or it being something which never goes bad), or buy a brand-name food which tastes exactly the same as the generic but costs 50% more, or buy food they'll let rot before they can be bothered to eat it... (Nor do I exempt my relatives from this criticism.)

In this section Tyler also says something that particularly amused me in this election season: "Most American voters are fairly moderate, disillusioned with both political parties, and looking for someone who can fill the proverbial niche of "getting something done," or "unifying the nation." Those are not the kind of attitudes that make for a revolutionary future." (A craving for strongmen like Mussolini is not revolutionary?)

So what does that leave us? A weak diagnostic followup to TGS. One of the longest and most interesting writeups of Advanced Chess around. Some vague speculation about specifics of software/AI improvements to other sectors of the economy, badly handicapped by being written in 2013 (hopefully Cowen could do a much better job now). Some weak solutions or bandaids like MOOCs. And a reasonable but pessimistic extrapolation. Overall, not particularly worth reading unless you are interested in chess.