Month: October 2002

Both of them scour the Recently Changed list at weblogs.com and pick up links to books.
Weblog Bookwatch has lists of recently reviewed books, shows you which weblogs have reviewed the books, and which other books those blogs reviewed. Bookwatch collects links to Amazon, Powells, and Barnes and Noble.
AllConsuming.Net does a similar search, but prioritizes its lists based on recently mentioned books, so it’s more of a zeitgeist-tracking tool for those who want to keep up with blog fashion. Also lets you publish a pretty list of books to read, with pictures.
For hours of surfing delight. I’ve been addicted for years to Amazon surfing; start with a subject, look for related books based on recommendations, reviews and lists; and build a list of books to read. Then again, I’ve been addicted for many more years to libraries and bookstores; the vice hasn’t changed, only the medium.

Physicist Steven Weinberg writes about Steven Wolfram’s A New Kind of Science in the New York Review of Books.
Mostly he writes about why particle physics is better than other kinds of science: “although these free-floating theories are interesting and important, they are not truly fundamental, because they may or may not apply to a given system; to justify applying one of these theories in a given context you have to be able to deduce the axioms of the theory in that context from the really fundamental laws of nature.”
Weinberg disclaims this opinion, but he repeats it often enough that it’s clear which side of the flamewar he’s on. Weinberg thinks science should offer one fundamental theory of the world. He is not interested in the idea that there might be different levels of organization in the universe, so that the algorithm that modeled plant growth, say, was different than the algorithm that modeled competition among species in an ecosystem.
In fact Weinberg doesn’t seem convinced by the idea of modeling. “Take snowflakes. Wolfram has found cellular automata in which each step corresponds to the gain or loss of water molecules on the circumference of a growing snowflake. After adding a few hundred molecules some of these automata produce patterns that do look like real snowflakes. The trouble is that real snowflakes don’t contain a few hundred water molecules, but more than a thousand billion billion molecules. If Wolfram knows what pattern his cellular automaton would produce if it ran long enough to add that many water molecules, he does not say so.”
The whole trouble with complex systems is that they are programs that you need to run fully, with identical initial conditions, to get the exact result. If a model can be used regularly to make predictions about a real-world system — even if the model doesn’t duplicate the system — it seems to me that model is worth something.
The most interesting thing Weinberg says that Wolfram should do but doesn’t, is to offer a definition and measure for complexity. A very clever, erudite and witty person named Cosma Shalizi claims to have done this in his doctoral dissertation. Which I have not read yet, and my undergrad-level math may not be sufficient to understand.

A few days ago, I wrote about Tom Ray’s neat dispatch of Ray Kurzweil’s contention that computers will soon be smarter than we are. To give Mr. Kurzweil his due, here’s a link to a lovely essay critiquing Stephen Wolfram’s A New Kind of Science.
Wolfram’s book became a controversial best-seller based on the author’s claim that computational methods enable a revolutionary approach to science. Many people have criticized the book because Wolfram is an egomaniac who claims to be smarter than everyone else on the planet; because he doesn’t go through the traditional scientific peer review process; and because the sprawling, self-published 1192-page tome really could have used an editor.
Kurzweil ignores the gossip and the copy-editing, and deals with the ideas. Kurzweil’s essay analyzes two of Wolfram’s revolutionary claims: that computational approaches based on cellular automata can explain life and intelligence, and that they define physics.
A quick definition: cellular automata are a type of logical system composed of simple objects whose state is determined by following simple rules about the state of fellow objects; like junior high school girls who will wear tomorrow what the popular girls wore today. The results of many cellular automata are quite boring. Either they fall into a steady state, where nothing changes (class 1), or a simple pattern repeats tediously (class 2), or they twitch forever without any detectable pattern (class 3) But some cellular automata (class 4) are much more interesting. A class 4 automaton generates a complicated pattern that, in Kurzweil’s words, “is neither regular nor completely random. It appears to have some order, but is never predictable.” A class 4 automaton can be used to convey information, and hence can be used as a “universal computer.”
Do cellular automata explain life?
Wolfram argues that because cellular automata can generate behavior of arbitrary complexity, they therefore explain living systems and intelligence. Kurzweil neatly explains that just because cellular automata can generate complex patterns, doesn’t mean that life and intelligence will automatically follow.
In Kurzweil’s words, “One could run these automata for trillions or even trillions of trillions of iterations, and the image would remain at the same limited level of complexity. They do not evolve into, say, insects, or humans, or Chopin preludes, or anything else that we might consider of a higher order of complexity than the streaks and intermingling triangles that we see in these images.”
As discussed in this essay on artificial life, the software for life is based on a layered architecture with many components and layers: evolution, growth, metabolism, ecosystems. Just cause we can program computers — using CAs or any other method — doesn’t mean that we know how to build every kind of software in the universe.
Do cellular automata explain physics?
Wolfram claims that cellular automata provide a better model for physics than traditional equations, and more than that, the universe itself is one big cellular automaton.
Kurzweil puts Wolfram’s claims about physics into context, as part of a school of thought whose advocates, including Norbert Weiner and Ed Fredkin, argue that the universe is fundamentally composed of information. Particles and waves, matter and energy, are manifestations of patterns of information.
The way to go about demonstrating this hypothesis is to use cellular automata to emulate the laws of physics, to see if this generates equivalent or better results than the existing sets of equations. The mapping is apparently easy for Newtonian physics; workable but not particularly elegant for Einstein’s special relativity, and potentially an elegant and even superior way to represent quantum physics, because CAs generate patterns that are recognizably regular, whose details are impossible to predict.
In summary, Kurtzweil thinks that Wolfram’s thesis regarding physics is plausible, but it has yet to be proven, and Wolfram hasn’t proved it.
One thing I don’t understand about this hypothesis is why it proves that the universe IS a computer. If you prove that computation is a better model for physical phenomena, how have you proven that the model is reality itself? A equation can predict where a ball will land, based on the speed and direction of its flight, but the ball itself isn’t an equation. Some day, I’ll take a look at Kurzweil’s book The Age of Intelligent Machines, which covers this topic, and see what I think.
With Kurzweil’s synopsis as a guide, I’ll take a stab at reading Wolfram’s tome. Not because I think it will contain the answer to every question, but because I expect an interesting exploration of cellular automata, and an interesting take on the information hypothesis to physics.

Tom Ray wrote a very nice essay critiquing Ray Kurzweil’s argument that machines will soon be smarter than we are.
The first point is plain logic. Kurzweil observes that following Moore’s law, computers will have more processing power than the human brain within a couple of decades. Ray points out that the power of software is not improving at anywhere near the same rate. There’s plenty of evidence that complicated software is outstripping our ability to design and maintain it effectively.
The second point is more subtle. Kurzweil argues that it will be possible to implement human intelligence in silicon, simply by reverse engineering the brain and mapping its neural connections into software. Ray notes that there are many aspects of human intelligence that depend on subtle properties of chemistry, for example, the delicate balance of hormones that influences temperament and mood, shaping our decisions, communication, and art.
It may be possible to create AI. Ray, who created the “Tierra” artificial life ecosystem, believes that the most promising method is to create digital a-life systems and let them evolve on their own. If such intelligence evolved, it would be different than human intelligence, depending on the very different properties of its technology and environment.
At any rate, the mechanisms to create artificial intelligence aren’t obvious, and there isn’t any reason to believe that it will happen any time soon.