## July 31, 2008

### Books to Read While the Algae Grow in Your Fur, July 2008

Nicolò Cesa-Bianchi and Gábor Lugosi, Prediction, Learning, and Games
A wonderful synthesis of the literature on competitive, individual-sequence forecasting with expert advice. That is, the problems considered are all variants on a situation where you need to make a prediction about the future (or more generally take an action whose consequences will only be revealed in the future), have access to a range of "experts" or forecasting algorithms, and want to ensure that, no matter what actually happens, your performance will be close to that of the best expert. This is thus a study of sequential decision-making under uncertainty without probability. Often, but not always, the solution lies in taking weighted averages of the experts, giving more weight to those which have done well in the past. This works not because past performance provides any kind of inductive evidence of future success, but merely because it keeps your predictions from drifting too far from what is, in fact, working. (Perversely, many of the proofs rely on probabilistic arguments, but they don't make probabilistic assumptions.) Of course, it may be that even the best expert is very bad, but the possibility of improving on the experts is not really considered, though it's certainly possible (at least with convex loss functions).
Anyone at all interested in machine learning, forecasting, information, game theory, or decision-making under uncertainty needs to read this. It may also be useful to epistemologists (cf.).
Max Brooks, World War Z: An Oral History of the Zombie War
An oral history (a la Studs Terkel) of the early 21st-century global struggle against the zombie apocalypse (a la George Romero). This is a happy choice of form, because it lets him tell the story of a global disaster from many viewpoints, without taking the space which would be required in a conventional cast-of-thousands novel. Also, he gets to tell lots of variously creepy, horrifying, thrilling, and/or moving stories this way.
Query: does this qualify as a "modern epic", sensu Moretti?
Marc Van De Mieroop, The Ancient Mesopotamian City
As every school-child knows, "history begins at Sumer", with the first cities and the first writing. This book is the only accessible synoptic view of the cities of ancient Mesopotamia as such.
After opening by quoting some jaw-droppingly ignorant (and recent) remarks by classicists on how there were no real cities before the Greeks, Van De Mieroop describes the geographic scene, and lays out some of the limitations on our evidence — peculiarities in what scribes thought worth recording, and other peculiarities in what archaeologists have thought worth excavating. Next he considers theories of the origins of cities in Mesopotamia, a peculiarly difficult problem since there were no other cities to learn from or be influenced by. He favors the idea that they originated around the temples, which acted as institutions for redistributing the products of multiple ecological regions, but he is fair to other ideas. (He is even fair to Jane Jacobs's wacky idea that cities preceded, and caused, agriculture, which is to say he does some simple calculations to show it makes no sense whatsoever.) He then goes on to consider social organization, leading institutions like the palace and the temple, the hints of self-government among city-dwellers and their growth over time, the relations between cities and their agricultural hinterlands, how food moved into the cities, long-distance trade, credit and finance, and cities as centers of religion and learning, including divination and astronomy. (He says scribes were taught "calculus", presumably meaning "calculation".) He quotes frequently from Mesopotamian documents, without any philological apparatus, and despite a ritual rejection of strict "positivism", he is very cautious in advancing hypotheses, and very good about marking conjectures as such, and emphasizing that we simply have little or no evidence about many matters.
Mesopotamian history is usually considered to last from the first writing around -3100 to the Macedonian conquests around -300. As Van De Mieroop says, this period of 2800 years is longer than the interval separating us from Homer. It is an astonishing act of hubris, or at least of abstraction, to try to summarize the features of all cities over such a period, even in a restricted region — one can only presume that there must have been extensive variation. Nonetheless, Van De Mieroop does a really remarkable job.
You remember "Installing Linux on a Dead Badger", don't you? Well, imagine a full hundred page book of such stories. C'mon, you know you want it.
Jack Campbell, Valiant
Continuing science-fictional anabasis; see here for previous installments.
Charles E. Lindblom and David K. Cohen Usable Knowledge: Social Science and Social Problem Solving
This is a short (100 pp.) book from 1979, largely given over to sketches of arguments and directions for further inquiry (mostly not undertaken in the ensuing years) about why the social sciences, and "professional social inquiry" more generally, have not been very directly useful for social problem solving. They suggest that this rests on a number of basic widespread mistakes about how social problems are solved. In particular, they allege, social scientists vastly under-rate the importance and competence of ordinary-life social knowledge, and, yet more consequentially, fail to see that social problems can be solved either by analyzing them in some discursive/analytic form, or by setting up patterns of social interaction where the participants' acts collectively solve the problem, though none of them need to grasp the solution or even realize that is what they are doing. Markets are of course one example of such "interactive problem solving", but they also, and quite correctly, emphasize others: democratic politics, bargaining processes, and the "republic of science". They emphasize that interactive problem-solving should not be seen as a poor substitute for formal problem solving, to be displaced in due time by scientifically-informed social engineering, but rather as inevitable, and indeed often superior.
The alternative, of analytically finding solutions to social problems, is basically impossible, because the problems are too complex, and even systematic investigation into them is not just prohibitively expensive, but so slow that the world has moved on before research findings can become very accurate or precise. (Obviously these obstacles can all be bigger or smaller in various cases, and I don't think they'd quibble if someone wanted to assert that very small, stable social problems could be successfully analyzed if enough resources were thrown at them.) Worse, the very definitions of "social problems" are themselves contested, and properly so. The authors' view is that while the natural sciences can (often) legitimately claim independent authority, for social scientists to aim at such authority is to set themselves a target they cannot possibly hit. Since they do aim at that target, however, social scientists and other "practitioners of professional social inquiry" systematically waste their efforts.
Given all this, fruitful roles for social analysts become things like advising individual participants in the interactions, or looking at the over-all performance of an interactive mechanism and searching for ways in which it might be improved. (They suggest that economists are better about this than other social scientists. Given the recent vogue among economists for replacing all kinds of institutions with arbitrary intellectual constructions, planned by analogy with the idealized markets of their Micro 1 textbooks, I suspect the authors might wish to revise and extend these remarks.) A further, if more diffusive, constructive role would be in hoping to shape the general framework within which participants in interactive problem-solving think about things; and of course the kind of detailed reportage which statistical bureaus engage in.
The name "Hayek" does not appear anywhere in this book.
Douglas W. Tallamy, Bringing Nature Home: How Native Plants Sustain Wildlife in Our Gardens
This is a rather straight-forward argument for North American suburbanites (and urbanites, too) to plant more native plants, roughly as follows. (1) We like having wildlife such as birds and (some) mammals around. (2) These animals sit high in the food-web; below them are many insects, especially insect larvae. (3) Most insects are specialized to eat only a small range of plants with which they have co-evolved, in no small part because of the chemical defenses evolved by plants. (4) Thus, native plants support a much larger and much more diverse population of insects than do introduced ones. (5) The process of evolutionary adaptation is very slow, and even plants introduced almost 500 years ago are still substantially less good as insect hosts than natives. (6) Suburbia occupies such a huge part of the American landscape that if native plants are to thrive anywhere, it has to be there. Therefore, (7) suburbanites should shift what they plant towards natives.
Obviously, the key empirical parts of this argument are (4) and (5); here the evidence that Tallamy presents is good, but — he is admirably up-front about this — not conclusive, and he is happy to admit that there are some species of introduced plants which are so closely related to natives that bugs like them just fine. Unlike many books on native plants, this is empirical, consequentialist, modest, and un-mystical.
David Pollard, Empirical Processes: Theory and Applications (full text free online)
The simplest sort of empirical process arises when trying to estimate a probability distribution from sample data. The difference between the empirical distribution function $F_n(x)$ and the true distribution function $F(x)$ converges to zero everywhere (by the law of large numbers), and — this is non-trivial — the maximum difference between the empirical and true distribution functions converges to zero, too (by the Glivenko-Cantelli theorem, a uniform law of large numbers). The "empirical process" $E_n(x)$ is the re-scaled difference, $\sqrt{n} \left[ F_n(x) - F(x) \right]$, and it converges to a Gaussian stochastic process that only depends on the true distribution (by the functional central limit theorem). Empirical process theory is concerned with generalizing this sort of material to other stochastic processes determined by random samples, and indexed by infinite classes (like the real line, or the class of all Borel sets on the line, or some space parameterizing a regression model). The typical objects of concern are proving uniform limit theorems, and with establishing distributional limits. (For instance, one might one want to prove that the errors of all possible regression models in some class will come close to their expected errors, so that maximum-likelihood or least-squares estimation is consistent. [For more on that line of thought, see Sara van de Geer's book on Empirical Processes in M-Estimation.]) This endeavor is closely linked to Vapnik-Chervonenkis-style learning theory, and in fact one can see VC theory as an application of empirical process theory. (I'd guess Vapnik himself would disagree with that, however.)
This short book by Pollard is an introduction to empirical process theory by a statistician for statisticians. As such it succeeds admirably; as always, Pollard does a really good job of explaining what the technical apparatus is doing and why it takes the form it does. People coming to it from other backgrounds (I am particularly thinking of computer scientists) will probably find it harder going, not least because the implied reader has an extremely sure grasp on measure theory. (Such as one might acquire from, oh, Pollard's User's Guide to Measure-Theoretic Probability.) If you can handle Pollard's 1989 survey paper, then you will probably enjoy this book; and if not, not. The applications he describes are all interesting and challenging, though I was a bit disappointd that none of them involve dependent data.
Clubbing
Fall of Cthulhu: The Fugue
Scalped: Indian Country
Proof: Goatsucker
Various flavors of comic-book mind-candy. Sequels to Scalped: 2--4, 5
Jenny Davidson, The Explosionist
It is impossible to describe this better than the author, so I'll steal her words:
the story of a 15-year-old girl growing up in an alternate version of 1930s Edinburgh, one where the legacy of Napoleon's victory a century earlier at Waterloo is a standoff between a totalitarian Federation of European States and a group of independent northern countries called the New Hanseatic League. This world is preoccupied with technology (everything from electric cookers to high explosives) but also with spiritualism, a movement our world largely abandoned in the early twentieth century; Sigmund Freud is a radio talk-show crank, cars run on hydrogen and the most prominent scientists experiment with new ways of contacting the dead.
My biggest complaint with this book is that it ends in the middle of the story, and nothing warns the reader about this. Grrrr.
Charles Tilly, Democracy
An attempt to explain the mechanisms by which states come to engage in "broad, equal, protected, mutually-binding consultation" with their citizens, singling out three especially important processes: (1) integrating "trust networks" into public politics, (2) screening off public politics from categorical forms of inequality, and (3) suppressing non-state centers of coercive power. As usual with Tilly, he draws on a huge range of historical sources, in an impressive display of erudition and clear thinking. Also as usual with Tilly, one does not get a comprehensive theory, but perhaps this is the sort of material where such a theory isn't really possible, and the best one can hope for is a catalog of recurring mechanisms.
Charles Tilly, Big Structures, Large Processes, Huge Comparisons
A brisk debunking of pernicious ideas about how societies work and change that we have inherited from the 19th century, together with a smart and enthusiastic brief for comparative, historical social science. (The description at Powell's is definitely for another book!)
Thanks to Doug White for lending me his copy.

Posted at July 31, 2008 23:59 | permanent link