Truck and Barter: Krugman, Darwin, Keynes, Popper, Kolmogorov, and Buchanan

April 12, 2005

Krugman, Darwin, Keynes, Popper, Kolmogorov, and Buchanan

By William

Recently I saw Paul Krugman's analogy between disrespecting Keynes and disrespecting Darwin -- quoted with no registration required here. Is Keynes really comparable to Darwin? Instead of, perhaps, someone with fuzzier and more controversial ideas (punctuated equilibrium, sociobiology...) laid over Darwin?

Is there a good layman's summary somewhere of nice sharp phenomena that Keynes' ideas cleanly and reliably predict that rival ideas didn't? For Darwin, off the top of my head, I think of (1) sex (of course!:-), (2) the way that molecular biological genetic relatedness so closely matches relatedness worked out by other methods before 1950, and (3) the way that so many of the most common human genetic illnesses (most famously sickle cell anemia) are simple mutations away from wild type which provide crude protection against some important threat in some historical environment. As far as I have heard (as a layman but probably more informed than the target NYT audience that Krugman was writing for), the footprint of Keynes' theories in the real world might not include anything particularly sharp at all, and probably includes nothing comparable to the endless megabytes of regularities in the flood of genetic data which started coming in only long after Darwin's death. I was under the impression that the footprint of Keynes' theories is more like the footprint of theories about the effectiveness of capital punishment in deterring murder: one can come to conclusion "yes" or "no" (esp. if one is predisposed to prefer that conclusion!) but either way, one's strongest conclusion might be that the effect is not so huge compared to the other confounding effects that it's easy to pick it out without clever statistical analysis.

On the flip side of predictive power, Darwin's ideas perennially take flak for allegedly not being "falsifiable" in the sense of Karl Popper; one example I picked up from a little Googling is here. But in at least one sense, natural selection seems cleanly falsifiable: I can easily imagine a history of the last 130 years in an alternate universe where natural selection was wrong, and where we would have realized its falsity by now. If, e.g., our universe had no nuclear fusion, so that Lord Kelvin's famous analysis showed correctly that the Sun could not be nearly old enough to have kept things running for a suitably long period of evolution, then we would at least have very serious doubts about natural selection origins of modern life forms. And if instead of all one genetic code, it had turned out that apparently-related organisms generally had gratuitously incompatible genetic mechanisms, with no reasonable way to evolve incrementally from common ancestors, the theory of evolution might have dropped dead right then. Is there an alternate history from 1930 to 2005, in some alternate universe where Keynes was wrong, that would have convinced the Keynesians in that universe that they were mistaken?

Of course, even if it turns out that Keynes' ideas have relatively weak predictive footprint in the observable world compared to Darwin's ideas, that doesn't mean that they're necessarily unimportant or wrong. Human nutrition doesn't explain most disease, and picking its effects out of messy real world datasets can be tricky, and some formulations like "nutrition affects health" might even be hard to falsify in general. Still, it would be foolish to deny that nutrition is vital in understanding disease and health, and if in the early 20th century Senyek had exploded the conventional wisdom by revealing that nutrition affects health, perhaps he might be compared with Darwin. On another hand, though, the footprint of nutrition isn't always hard to see: it would be easy to point nutrition skeptics to extreme cases like scurvy on early ocean voyages where the effect of nutrition on health is enormous. And while theories with hard-to-see footprints can be important, tolerance for them is a necessary condition for people believing for centuries that human health is usefully modelled with humors.:-| So I sort of hope that the Keynesians have a standard evidentiary smackdown for ignorant but interested skeptics.

Incidentally, sometimes I see flareups of an ongoing low-intensity conflict about how much mathematics is appropriate in economics, and whether current math emphasis tends to displace other useful methods of analysis. The intellectual history of the falsifiability-related ideas that I'm slinging around here might be an interesting case study to think about. Popper and Lakatos did a lot of qualitative philosophical work on the qualitative idea of falsifiability, but at about the same time, people like Shannon and Kolmogorov were working out the math of information theory and computational complexity, which can be used to make quantitative versions of similar arguments. And to me, it looks as though the math results are more useful. In particular, my superficial study of the philosophers (mostly in one semester of undergrad philosophy of science) left me with the impression that the philosophers got tied in knots by situations like Newton's theory being mostly right, but "falsified" a little bit: subtly but clearly in the real world by the precession of Mercury, or grossly and clearly in a hypothetical world where, e.g., a moon of Neptune reversed the direction of its orbit every two years. For theories with numerical values, like computational complexity (number of bits) and information theory (entropy), this seems to be less of a scary problem, just a messy one. If you're working with qualitative definitions, it's easy to get stuck on, more or less, "a little bit falsified is like a little bit pregnant." But if your definition of a theory's goodness is quantitative, so e.g. "the messages in this stream are written in English" means "with an English-specific computer program I can compress the messages effectively" then when it turns out that some of the messages have Russian quotes in them, you can still say useful things about how useful the English-specific computer program is in compressing the mostly-English stream, instead of just saying "the old hypothesis has been falsified and must be discarded!" It's like Occam's Razor on performance drugs: once you can actually measure ad-hoc-ness, you can in principle say something like "yes, I grant this part is ad hoc, but still my theory is still 14.8 surprisons more predictive than yours over the entire dataset." In this one case, it seems to me that the math folk were ahead of the (nonetheless justifiably highly regarded) qualitative folk.

Another possible point of comparison is that Darwin's ideas have turned out to be fruitful in ways which would have been very hard to anticipate in the 19th century. In particular, now that we have powerful computers, people can use computerized "genetic algorithms," simulating natural selection in a computer program by allowing millions of partial solutions to "mutate" randomly and "breed" and "reproduce" with no designer input other than granting more offspring to partially successful solutions. Not only do such genetic algorithms turn out to be a surprisingly effective way of solving some kinds of problems, but the detailed analogy to natural selection in biology provides clear guidance for algorithmic improvements. In particular, features borrowed from biological sexual reproduction are extremely effective for helping genetic algorithms find good solutions faster. I would be interested to hear of Keynes' insights cross-pollinating other fields; I haven't so far.

Incidentally, I have also heard people say that some of Hayek's ideas have also been fruitful outside of the field of economics. I gather that Krugman would be hostile to such claims, and I don't know enough to address his hostility on that point. Perhaps fair is fair: I have heard people say that Marx's insights have been fruitful outside economics, and *I'm* hostile to that.:-|

Also incidentally, perhaps genetic algorithms aren't really a case of Darwin's insights cross-pollinating other fields, but a case of Darwin's controversial analysis being validated by later independent discoveries in other fields. If Darwin had been eaten by a Galapagos Giant Finch before writing up, and somehow no one else in biology had come up with his ideas by 1980, I'd guess that workers in algorithms would have reinvented these search tricks independently. Then, if they had, the analogy to sex is so strong that I'd rather expect that algorithms researchers would have been able to follow the analogy in the opposite direction and conclude that aha! Nature seems to be refining its designs using a selection process analogous to what we use in our computers!)

Finally, by (happy?) coincidence, if I were to nominate a theme in twentieth century economics comparable to natural selection, and if I were to nominate such a theme in twentieth-century economics to twit Krugman by highlighting left-wing closedmindedness and willingness to ignore the Nobel-prize-level well-known when it is politically inconvenient, it would be the same theme: not Keynesianism, but public choice theory. With or without the idea of natural selection, messy biological arms race situations (like the ecology of parasites and infectious disease, and the immune system, omigod) are still bewilderingly complicated. But once you appreciate natural selection, you find regularities all over the place which reliably help to simplify things... Posted at April 12, 2005 05:52 PM

Comments