November 23, 2003

More Assorted Reading

Umberto Eco talks in Alexandria about the future of books. For something completely different, Dan Sperber on the future of writing.

Peter Bergen tears apart the theories of Laurie Mylroie, who believes Saddam Hussein was responsible for both attacks on the World Trade Center, the anthrax attacks in 2001, just about every terrorist act attributed to al Qaeda, and the Oklahoma City bombing.

William Dalrymple's Murder in Karachi is a wonderful review of two books on the murder of Daniel Pearl: what seems like a very good one, by his widow, and a very bad one by Bernard-Henri Levy. The Daniel Pearl Foundation is a highly worthy cause. Melancholy "small world" factoid: Daniel Pearl's father is Judea Pearl, who is (deservedly) famous in artificial intelligence and statistics for his work on graphical models of causality.

At LanguageLog, a group blog on linguistics, John McWhorter has a nice post on an irritating way of making other languages seem more exotic than they really are: assuming that a word which do not translate cleanly into a single English word refers to a single very strange and profound concept, rather than simply having multiple senses, a point he illustrates nicely by considering the English word "stand". The inverse fallacy of translation is of course the Great Eskimo Vocabulary Hoax, claissically debunked by another poster at Language Log, Geoffrey Pullum. Mark Liberman (who seems to be the ringleader of the blog) suggests a very nice analogy for how it could simultaneously be the case that any human thought is expressible in any human language, and that differences in language profoundly shape human thought. Liberman also has a nice thought on the pragmatics of cold-reading. My exceedingly ill-informed guess is that the questions he raises could be dealt with in Sperber and Wilson's relevance theory, or at least it should not be hard to get relevance theory to make predictions about what will happen in cold reading situations.

John Holbo has fun (1, 2) with conservatives trying to think above their station. There is a strain of conservative thought (Burke comes to mind, and Oakeshott, and even, in a way, the late Wittgenstein) which counsels, or at least hints, that conservatives should refrain from trying to give articulate, rational, principled justifications for their positions. This takes several forms: a denial that tradition (or intimations, or forms of life, or what have you) needs rational justification. Or: it makes no more sense to justify a form of life than to justify an orange. Or: it could be rationally justified, but it is so much more subtle than the average capacity that attempts to justify it will only lead to misunderstanding, doubt and sophistry. (I think this was Burke's view, but I speak under correction.) Or: the very attempt to justify it replaces a living, unreflective tradition with a self-conscious traditionalism, which is just another ism with none of the ineffable qualities that make a genuine tradition superior to merely thinking about what one is doing. Or, perhaps wisest of all: Conservative positions cannot be rationally justified. --- Speaking of rhetoric (we were, weren't we?), notice the tactic that John uses on poor Frum in the first post I linked to: saying F advocates X, but an obvious and immediate consequence of actually getting X would be Y, so F really advocates Y. Bertrand Russell used to do this a lot too, which puts John in good company. It's not psychologically accurate --- people don't want all the consequences of their desires, if only because they can never think of all the consequences --- but it can be effective. I wonder whether it isn't somehow related to training in constructing reductio ad absurdam proofs. (Like me, John posts more when his wife is away, so we can expect more of John and Belle in the next few months, though I hope not less of Belle. Oh, and get a glimpse of the latest future member of Adult Children of Academic Bloggers.)

The Decembrist (a.k.a. Mark Schmitt) has nice posts about the multiplication of risks and the possibilities of a new American social contract. Think about health insurance, which Americans are currently supposed to get through their employers. There is the well-known overhead cost of firms administering these plans, plus the additional overhead costs of the HMOs, but there are subtler problems. It's easy to compare two different wage offers, but, as anyone knows who's had to do it, very hard to compare two different health-care plans --- they're too intricate, too high-dimensional, sometimes simply incomparable. But this means it is very costly for workers to evaluate different employment offers, which in itself makes the labor market inefficient. Additional inefficiencies come from the relcutance to leave a job with good benefits, even for a position where one might be much more productive, because of the uncertainty it entails. In particular, workers are going to be very reluctant to start their own enterprises. But, collectively, we want people to take that kind of risk, because it's very valuable when it pays off. Universal coverage removes that wholly artificial uncertainty from the decision problem, so that people are left with merely the ordinary risk that the enterprise will fail. So a well-designed welfare state, a social insurance state, would actually promote risk-taking, entrepreneurship and innovation. I think this whole topic of the relationship between public policy and risk is very important, and I hope Schmitt returns to it soon.

Robert Reich makes the basic point that manufacturing jobs are vanishing not because of international trade, but because manufacturing productivity is rising faster than manufacturing demand. Same number of widgets divided by more widgets per worker equals fewer workers; full-stop, 75--80 percent of story. It's an elementary point, but it's astonishing how much cant gets cleared away by keeping it firmly in mind. (I know I found this through a blog, but cannot now reconstruct which.)

Abstract Dynamics's Sex Slave to the Data Set wins the "phrase I most want to use in a referee report" award, and points out a very clear example of the mistake of thinking that the variables you happen to have access to must be the important ones.

Christopher Genovese has a long but fascinating post on the biology of mating responses. Another, very neat post explains how collaborative filtering is like boosting. In the future, however, I would appreciate it if Prof. Genovese would check whether any of these clever little posting notions of his actually are actually thoughts for papers snatched directly from my brain because I've not, in over a year, been able to find the time to work on them. (Yes, I'd have more time to write papers if I wasn't posting. Be quiet.) Actually the work I have in mind is at once more general, comparing different strategies for combining predictors with different forms of social organization for collective decision-making, and more technical, involving some issues of the effective VC dimension of ensemble classifiers. Still, reading that post gave me a bit of a turn... And on that note, to work.

Minds, Brains, and Neurons; Linkage; Enigmas of Chance; The Continuing Crisis; The Progressive Forces

Posted at November 23, 2003 12:37 | permanent link

Three-Toed Sloth