On “Average Is Over”
I’ve long been a fan of Marginal Revolution, Tyler Cowan’s economics blog, although I’ve read it less diligently since the recent financial crisis has eased. Unfortunately, though, I found Cowan’s book “Average Is Over” to be disappointing. Partly, it’s too long. Partly, it’s unhelpful. But mostly it’s just too conjectural and ungrounded to be convincing prediction.
The book presents a wonderfully dismal account of the future, with the rewards of economic growth in the stagnating economy. Frequently, the book seems like a meditation on the state of computation in competitive chess, using contemporary chess as a metaphor for economic trends. Cowan marvels at the accomplishments of computer chess programs, and projects AI success at playing games to human interactions: reading such subconscious metabolic clues as skin temperature to better discern lying or romantic interest than can the misguided intuitions of humans. He imagines, for example, an AI that can prod a man to kiss his date an hour before he might based on his intuition alone.
Of course, chess classically zero-sum) is a very different kind of game than dating and other human relationships, which surely should be positive-sum. Making the right moves in chess should lead to winning, but putting the moves on a girl is only winning if both people are. An adult who is dating and trying to “score” is probably doing it wrong. If strong AI allows us to more easily manipulate each other into extracting short-term need fulfillment rather than building relationships that are rewarding in the long term, that would be an enormous failure of the technology for overall human welfare.
More critically, while anyone can tell you that you should do something, nobody can make it something that is, for you. Trusting people and loving people is not something that people rationally undertake to maximize their self-interest, but something that they feel. A computer, like a person, can offer insight into whether a person is lying or aroused, but it makes as little sense to take the advice of an AI as a person that you should trust someone or you should love someone.
Choosing a good partner is probably the area where it is most important to base decisions upon our feelings and impulses. Cowan supports his argument by citing evidence dating sites, where peoples’ stated preferences (say, a man between 21 and 26 years old) is slightly at odds with their eventual choices (a man 28 years old). But it doesn’t take an AI of superhuman intelligence to realize that people are often overselective on paper about potential mates: a merely human intelligence can see that, and it would be a fairly stupid matchmaker to take such criteria at face value. Of course, most AI is pretty stupid.
The flaws of this attitude can be summed up in the lead to chapter 6: “Love and romance is one area where human intuition is highly imperfect and sometimes even dangerous. Apart from extreme crimes of passion, we take a lot of wrong turns in our pursuit of a good partner, thanks to following our feelings and impulses.”
With love, our feelings are not just our means; they are our ends.
This isn’t necessarily a fatal problem with Cowan’s book, he’s much stronger talking about problems more analogous to chess, like GPS tracking. But the book is almost entirely conjectural, a litany of seemingly unsubstantiated “wills” and “mights”; Cowan has an appendix (rather than proper footnotes) giving a mix of newspaper articles, blog posts, and academic articles to support his arguments. But once you hit some argument that seem less plausible, the plausibility of the book as a whole seems less grounded. Cowan discounts the idea of a Technological Singularity (rightly I think), but mostly on the grounds of what he intuitively finds implausible. It seems that the author’s intuition is the main basis for most of his projections, which must surely be problematic for a book predicated on the fallibility of intuition.