The Bactra Review: Occasional and eclectic book reviews by Cosma Shalizi   26

Turtles, Termites and Traffic Jams

Explorations in Massively Parallel Microworlds

by Mitchel Resnick

with a foreword by Seymour Papert

Complex Adaptive Systems series
MIT Press, 1994

Turtles Up the *

Mitchel Resnick is a reformed physicist and the principal creator of a new form of Logo, StarLogo, which works in parallel and is the pretext for this book. (The name comes from the Unix wildcard, *.) The point of StarLogo is the creation of "microworlds," highly stylized simulations. There are turtles in these microworlds, of course , as many turtles as you like, but they aren't under your direct, command-line control any more. Instead they've become autonomous little critters like Braitenberg's vehicles, their behavior governed by procedures called demons (another Unix hand-me-down, likely to keep the language out of some schools in this country), interacting with each other, and with fixed "patches" of the screen, which are themselves run by demons. (The patches are sophisticated enough to run cellular automata, so the turtles aren't strictly necessary, but they make it a lot easier to describe some processes.) A measure of global control is provided by procedures for an "observer", who basically sets up the turtles and patches and then sits back to watch and contemplate self-organization in action.

StarLogo is very neat, and, at least on the Mac, well-implemented. (It was originally designed for the Connection Machine, a piece of hardware curiously rare in classrooms.) I hope and almost expect that it will colonize the schools. Its description, together with Resnick's experiences in teaching it to high school students, forms the core of the book. The explanation of StarLogo is clear, concise, and accompanied by good sample code. The material on the students is well-integrated with that on StarLogo, and does a remarkable job of bringing out the way the students thought, especially how hard it was from them to twig that central controllers weren't needed. All this material, roughly pp. 31---118, is wholly admirable.

Such a sentence is, of course, a clear sign that I find other parts of the book less than admirable. The solid core is coated by two squishy sections about "the centralized mindset", "the era of decentralization", constructionism, etc., with which I take issue, perhaps with special force because I like the heart of the book so much.

Logo does an admirable job of teaching children how to draw pictures on the screen, and even gives them a certain grasp of plane geometry, and to children, and none-too-sharp parents and teachers, it may seem that that's all. To think this is as naive as thinking that De Bello Gallico teaches European geography and Latin declensions, or Hsiao Ching the most common Chinese characters. Just as those ancient educational staples inculcate Rome, Empire and militarism, and Confucian family and political values, respectively, Logo insinuates the habit of abstract, explicit, operational and analytical thought. It surpasses Euclid, because children actually enjoy it (keeping the name "turtle" for the cursor but abandoning the mechanical device was, in retrospect, a stroke of genius), and the entire process is implicit, even insidious. I approve whole-heartedly, having become one of the turtle-people myself at an early age, and think Logo will go down as one of this century's great achievements in teaching.

Resnick is quite aware of this, and wants to use StarLogo to insinuate a different body of knowledge and a different sort of thinking. Instead of plane geometry he wants to teach parallel processing and self-organization (how do termites build their nests? how do slime-molds aggregate? how does traffic jam?); instead of analysis and anatomy, he wants the "centralized mindset" to give way, at least in large part, to a "decentralized" alternative. The question which bothers him is how something can act coherently and unitedly without a central controller, how, one might say, to build a Leviathan without a sovereign. This is not just a political desideratum (though Resnick makes it that, too) but a real intellectual problem: as one of my neuroscience professors likes to say, "The closer you look at the brain, the less it seems like there's anybody home." In Resnick's mind, the shift to market economies and devolution of power from central governments, the increasing acceptance of "small is beautiful" and related ideas in the general culture, the desire to use parallel and decentralized computing technologies, "society of mind" theories, and the attempts in science to handle large, complex systems are all part of a single trend. He even suggests this trend may change the nature of knowledge, or some such, and hints at links to deconstruction (but Deleuze and Guattari would be closer to the mark). The level of analysis here is superficial, at times (e.g., when he talks about the fall of the Soviet Union, or, may Wollstonecraft defend us, "feminist epistemology") painfully so; but I don't think he has too much invested in this.

More seriously, he underestimates, I think, the virtues of what he calls the "centralized mindset", the guiding assumption that order is due to some central controller. He suggests (pp. 129--130) that it has a triple root, in the fact that many phenomena do have a central organizer; that social systems often have centralized power and authority; and that we experience ourselves as unified selves, even though (if the cognitive scientists are to be believed) we are "composed of thousands of interacting entities." This is probably all true, as far as it goes, but he doesn't give enough weight to it, and he misses two special virtues of the centralized approach: central control is a strong hypothesis which, if true, gives you a lot of information and predictive, manipulative and explanatory power. Rather than having to track the states of all the thousands, if not billions and billons of parts and understanding their interactions, you can just track the state of Control Central, and abstract away from all the details of how it exerts control. (It is a bit surprising that someone who has studied, not just The Sciences of the Artificial, but Abelson and Sussman's Structure and Interpretation of Computer Programs would not see this.) Looking for Control Central is similar to looking for complete determinism: neither is guaranateed, and in fact both are often demonstrably false, but both are good first bets and guiding assumptions. (They are not, of course, the same assumption.) Even when one can show that, at some level of detail, central control does not exist, it can still be a vastly useful, if not essential, simplification: but, again, one shouldn't have to explain the virtues of the design and intentional stances to someone who cites Dennett's Consciousness Explained.

Resnick also overestimates the novelty of the "decentralized mindset." Even if we ignore associationism in psychology, economics, whether classical, Marxist or neo-classical, has been predicated on the absence of Control Central for two hundred years now, and indeed has made such a fetish of it that it struggles in dealing with real economies, where most transactions are within, not between, centralized organizations; Darwin made population thinking the core of biology 140 years ago; and statistical mechanics, the most successful science of emergent properties, is only a bit younger than that. If these are accomplishments of the decentralized mindset, then that mindset is hardly new. (Including Epicureanism could push this much further back, but arguably it was too marginal to the culture to count, and not substantively connected to what came later.) If, on the other hand, Smith, Marx, Darwin, Wallace, Maxwell, Boltzmann, Gibbs & Co., Emergent Behavior Agents, were really centralizers, one wonders what we need the decentralized approach for, other than keeping up with fashions set by management witch-doctors.

So much for decentralization and its rival. Resnick also has a theory of pedagogy, derived from Seymour Papert, which they call constructionism. This asserts that "learning is an active process, in which people actively construct knowledge from their experience in the world", and that "people construct new knowledge with particular effectiveness when they are engaged in constructing products that are personally meaningful." The first idea is supposed to come from Piaget, but, without some specification of what it means to "actively construct knowledge", I can't see how it differs in spirit from the Scholastics' Nihil in mente quod non prius in sensu, or even how it's incompatible with Plato's theory of Recollection. As to the second, it suggests, I think, that students will work more willingly at things they enjoy and which seem worthwhile to them. Clearly Resnick and Papert mean something more than this elementary observation, but it's not clear what, or that the extra really is entailed by their formulæ. The high school students Resnick worked with chose their own projects, which is supposed to have made them "personally meaningful" (as opposed to impersonal, Fregean meaning?) --- but Resnick suggested projects and provided Scientific American-level articles on various sorts of self-organization, as well as "questioning assumptions, helping with programming, and encouraging students to reflect on their experiences" (p. 151). They seem to have learned a lot, but this tells us squat about the effectiveness and accuracy of constructionism. Even if they were not pre-selected for extreme cleverness (and they all volunteered to participate in an experiment at MIT), it's a very poor pedagogical theory indeed which predicts that reasonably bright and healthy young people will not learn a lot when exposed to neat, powerful and not-overwhelming ideas in a focused setting, under the guidance of a knowledgeable, enthusiastic, and at least minimally personable teacher who shapes their interests and directs their thinking. Many of the basic points of statistical and experimental hygiene were neglected (even the minimal control of taking a group of students and assigning them to particular projects wasn't attempted), but this is so common in educational studies as to be scarcely worth remarking.

Constructionism is too vague to be useful, and the decentralized mindset is over-sold; but these are, literally, peripheral issues. The heart of the book, on how to teach people about decentralized processes and emergent behaviors, and the tricky process of designing and controlling them, is excellent. Since I happen to think that self-organization is both quite important and quite neat, this is to me a thing of joy; and no doubt people with different scientific and philosophical biases will find Resnick's woollier ideas congenial. Having gone into my reservations at, doubtless, excessive length, I can in good conscience recommend Turtles, Termites and Traffic Jams to all and sundry, but especially to those interested in parallelism, emergent behavior and education.

163 pages, bibliographic references, no index (which is vexing).
Artificial Life / Cellular Automata / Computers and Computing / Education, and Kindred Ways of Moulding the Young / Popular Science / Self-Organization, Complexity, etc.
Currently in print as a paperback, ISBN 0-262-68093-9, US$12.95 [buy from Powell's], and as a hardback, ISBN 0-262-18162-2, US$24.95 [buy from Powell's]. LoC QA76.5 R47 1991.
10 November 1997; typos fixed 31 May 2001 (with thanks to Sebbo)