January 31, 2009

Books to Read While the Algae Grow in Your Fur, January 2009

Charles Tilly, Trust and Rule
Tilly's principal book on "trust networks", how they sustain themselves or fail to do so, and how they relate to states and other forms of political power. Trust networks "consists of ramified interpersonal connections, consisting mainly of strong ties, within which people set valued, consequential, long-term resources and enterprises at risk to the malfeasance, mistakes, or failures of others" (p. 12). This defines trust as a relationship, one of exposing oneself to serious risk through another's malice or mistakes, a definition which is pointedly silent about the feelings accompanying the relationship. I think this nails it.
Tilly goes on (p. 13, italics his):
Most networks support little or no trust. We will sometimes recognize segments of networks that qualify as trust-connected cliques. But the networks of drug use, blood distribution, and sexual contact through which HIV spreads, the networks through which routine political information flows, and the networks established by shared membership in voluntary associations mostly do not qualify. More generally, single-stranded networks containing few triads and sustaining little intimacy among their nodes rarely or never become trust networks.

Characteristic enterprises in which trust networks figure importantly include cohabitation, procreation, provision for children, transmission of property, communication with supernatural forces, joint control of agricultural resources, long-distance trade, protection from [human] predators, maintenance of health, and collective response to disaster. With marked variation from setting to setting, trust networks often take the forms of religious sects and solidarities, lineages, trade diasporas, patron-client chains, credit networks, mutual aid societies, age grades, and local communities.

This is all very good, and I also like that Tilly does not romanticize trust networks, being explicit that terrorist cells, pirates, Russian mobsters, etc., all qualify, and that (as these examples suggest) the risky undertaking enabled by a trust network can be preying on others. Even without that: "Powerful figures within trust networks sometimes tyrannize their members: instill strange beliefs in them, put them through painful initiations, force youngsters into distasteful careers, require shows of respect for unworthy elders, murder young women who challenge their sexual or marital prescriptions. By no means does membership in a trust network guarantee happiness, much less freedom." Still, he convinces the reader, or at least me, that trust networks are enduring and important parts of society. He also offers some intriguing generalizations, like the one about the importance of triads in the network graph. (I think, but I don't recall him ever quite saying this explicitly, this is because triads make monitoring and reputation possible.) There are, as usual, many excellently-presented historical cases, spanning the globe and the centuries.
Nonetheless, I find myself less than fully satisfied. (1) Nobody except Tilly talks about "trust networks", at least not in this sense, and we rarely have historical cases where we can identify them with any precision. So there is a lot of guesswork here. (2) Tilly's stories about the kinds of mechanisms at work sound plausible, as usual, but I despair of ever being able to test them. (3) He offers no guidance about when we should expect different mechanisms to be engaged. Perhaps, to be fair, no guidance can be offered at this level — perhaps it always depends on high-precision historical details. (More minorly: [4], Tilly sometimes, as on p. 81, insists that the ties linking members of a trust network must be a relationship for which the participants have a name: why? I don't even think all his examples meet this criterion. [5] This already-brief book would have been ever shorter if he didn't repeat his definitions of terms over and over.) I can't help but think that Tilly's dislike of game theory may have been a liability.
It's interesting to think about science in terms of trust networks. Scientific collaboration is placing "valued, consequential, long-term resources and enterprises" — viz., the scientist's reputation and career — "at risk to the malfeasance, mistakes, or failures" of the scientist's collaborators. Might this go some way towards explaining features of scientific collaboration networks, like the very high density of triads, and persistent collaborative cliques? As for "strange beliefs", "painful initiations", "distasteful careers", and "deference to unworthy elders" (but fortunately not murder), the jokes draw themselves.
Dave Lapham, Silverfish
Mind candy. Noir crime-fiction in comic-book form, with teenage protagonists. Well-told and well-drawn.
Lois McMaster Bujold, The Sharing Knife, vol. 3, Passage
Mathukumalli Vidyasagar, Learning and Generalization: With Applications to Neural Networks
A very nice textbook on statistical learning theory (a la Vapnik) which, properly, treats it as a branch or extension of empirical process theory, and emphasizes function learning (instead of just classifier learning). Among the many nice features here is the recognition that data in the real world are dependent, and a discussion of conditions under which learning procedures designed for independent data will still work with dependent data, albeit with an efficiency penalty reflecting how quickly correlations decay. (Beta-mixing, for example, is sufficient but not necessary; an interesting open question is what the necessary and sufficient conditions on a mixing process are for probably-approximately-correct learning to remain possible.) Vidyasagar is also good at building connections to computational learning theory, which introduces considerations of time- and sample- complexity.
No prior knowledge of learning theory or even of measure-theoretic probability is required; all the necessary mathematical material is built here. Basic mathematical maturity, of the kind one would expect from graduate students in statistics, computer science, electrical engineering, physics, economics, etc., is essential.
The last two chapters consider, respectively, neural networks and problems in control theory. (Despite the back-cover blurb, support vector machines are discussed on only one page.) The neural network chapter is fairly self-contained, but the control chapter will be largely incomprehensible to those without previous exposure to the subject. This is a shame, since it contains about randomized algorithms for probably-approximately-correct solutions to intractable problems, and about systems identification, which would be of interest to readers whose eyes will glaze over (to say the least) at the sight of transfers functions. This is, however, at most a minor flaw.
If I were teaching a class on statistical learning theory, I would definitely consider using this book.
(Thanks to Dr. Vidyasagar for some interesting correspondence, which prompted me to read his book.)
James K. Galbraith, The Predator State: How Conservatives Abandoned the Free Market and Why Liberals Should Too
I really have nothing to important to add to Aaron Swartz's summary; other than to say read this book. (I am buying copies for friends and relatives.)
(On an un-important note, I think it's inevitable it is unfair but inevitable to compare this J. K. Galbraith writing about economics and public purposes to the other one, who happens to have been his father. [Or it's at least inevitable that I'd make the comparison, since the elder Galbraith is one of my heroes.] This book is in some ways an act of filial piety, losing few opportunities to point out places where the senior Galbraith has been vindicated by events. More broadly, its great theme is the collapse of what JKG I called "countervailing power", the thing which made American capitalism tolerable and even progressive — or, more exactly, the deliberate destruction of such countervailing power. For the most part this Galbraith avoids his father's style, — smooth as silk, and as hard to produce — in favor of more workmanlike prose; there are a few places, but only a few, where he is positively infelicitous, in ways his father would never have allowed into print.)
Dan Simmons, The Terror
Historical horror fiction, based on the Franklin expedition in search of the northwest passage of the 1840s. Some recurring Simmons themes (e.g., the characters with the unlikely fondness for classical Greek (who he should not have had discuss natural selection; I can't decide whether this is a greater offense against historical plausibility or against sheer narrative flow), and the contrast between adapted indigenous cultures (here, the "Esqimaux") and blundering, greedy, self-defeating westerners, though he doesn't hit the reader over the head with that last quite as bluntly as in his master-work Hyperion. (Oddly, non-western high civilizations come off very poorly in Simmons's fiction, as in the brilliant and terrifying yet borderline-racist Song of Kali.) Creepy and intensely compelling.
Steven Johnson, The Invention of Air: A History of Science, Faith, Revolution, and the Birth of America
Popular biography of Joseph Priestley. Johnson tries hard to keep in view both Priestley's individual biography and the larger networks and movements he participated in, so in part this is a bit of a ramble through the 18th century English-speaking house of intellect, which is not a bad thing.
There is, as the subtitle indicates, special emphasis — more than perhaps is truly warranted — on his American connections. This is because the book is very much an attempt by Johnson to claim Priestly as part of a usable American past of Enlightenment progressivism, in which there is no tension between rational religion and scientific advance. There is nothing wrong with this — quite the reverse; this is a part of our national traditions, and we should emphasize it — but at the same time it leads Johnson into some choices in his writing which feel like they make this book more transient than it needed to be.
Annoyances: (1) the formulaic opening scene. (2) the idea that the physiological effects of coffee sparked the Age of Reason was something I tossed off as a joke, complete with counter-examples, several years ago. I am displeased to see this same conceit here (pp. 54ff), being taken seriously not just by Johnson but evidently by others — not, to be clear, because I think I deserve credit (I'm sure it's not original), but because it is stupid.
Errata: p. 20, for "1850s" read "1750s"; p. 22, for "mid-seventeenth" read "mid-eighteenth".
H. P. Lovecraft, The Tomb, and Other Tales
Someone has already expressed my sentiments in lolcat form. (To say the least, Red Hook's changed.) But despite that there is a certain power to the stories.
Jeffrey Alford and Naomi Duguid, Mangoes and Curry Leaves: Culinary Travels through the Great Subcontinent
Very good recipes and nice travel writing, plus lickable-looking photographs.
Hiroshi Kondo, The Book of Saké (a.k.a. Saké: A Drinker's Guide)
Social history, lore and etiquette, a lovingly geeky description of the fermentation process (complete with graphs!), and specific recommendations. (Many thanks to K. for the gift, and for the Yuki no Bosha junmai ginjo.)

Books to Read While the Algae Grow in Your Fur; Scientifiction and Fantastica; The Progressive Forces; Networks; The Dismal Science; Enigmas of Chance; Writing for Antiquity; The Great Transformation; Food; The Running Dogs of Reaction; The Beloved Republic

Posted by crshalizi at January 31, 2009 23:59 | permanent link

Three-Toed Sloth:   Hosted, but not endorsed, by the Center for the Study of Complex Systems