IT `THINKS,' THEREFORE IT IS?
IN THE IMAGE OF THE BRAIN
Little, Brown -- 348pp -- $21.95
Pantheon -- 390pp -- $25
A foot-long robotic cokroach clambers over a telephone book as it crawls across a room. Named Genghis, it was born at the Massachusetts Institute of Technology. The metallic bug has six motorized legs, six sensors for a head, and a torso loaded with computer chips. Plus a new pal, Attila. Packed with 150 sensors, Attila is much smarter: When dropped on its back, Attila can rotate its head and legs 180 degrees so it can continue walking or--eek!--climbing.
Two new books profile Genghis, Attila, and other projects that move beyond artificial intelligence as we have come to know it. AI's disciples have long been obsessed with replicating the mind's abstract thought patterns. The traditional approach has been to use computers to crunch through long lists of the mind's rules about the world--for example, that rooms have four walls. By contrast, these books explore efforts to mimic thought and behavior more precisely through neurobiology and genetics. Genghis and Attila, for instance, have no concept of the room they are in. Instead, they rely on base neurological instincts to avoid objects, wander around, and learn about their environment.
Journalist Jim Jubak and author Steven Levy both profile the scientists and entrepreneurs on this frontier. Jubak's approach is scientific, while Levy's is philosophical. Jubak's In the Image of the Brain grapples with the technology of duplicating the human brain and nervous system. It asks: How close can we get? Levy's Artificial Life ponders the implications of creating organisms out of software and silicon. It asks: How will the existence of artificial life change our view of natural life and ourselves?
Jubak shows how Genghis and Attila shape up against lab rats in navigating an environment. The robo-bugs seem to learn simple actions one step at a time--is there something in the way, or not? Eventually, they find their way around and perform complex routines. Rats store a mental map of landmarks. But change the layout, and they resort to the same devices as Genghis and Attila.
Such comparisons between computers and animals are central to the field of so-called neural networks, which Jubak explores in depth. A neural network simulates a nervous system using chips or software instead of neurons. The theory is that the brain, or any neural network, is really just the sum of its physical parts rather than some mysterious reasoning machine. "The biology of the brain is the cause of the mind," Jubak writes. Taking that as a premise, one group of researchers at Carnegie Mellon University has succeeded in building a neural net that drives a Chevy van by learning the patterns of the road. Other neural nets can select a familiar face from a photo or weigh mortgage risks.
Levy takes a much more Asimovian view, challenging the accepted definition of life itself. Levy argues that Genghis and Attila may evolve into something that could be considered--EEK!--alive. What if you built an insect that behaved so much like an insect that for all practical purposes it was an insect? Such an image is especially jarring in light of the MIT scientists' vision of using a colony of Attilas to explore distant planets.
For the most part, though, Levy focuses not on robots but on computer simulations of life. He sees enormous implications in alternative life forms that move, eat, reproduce, and kill silently on computer screens and disk drives. He interviews some very quirky people at some very respectable research labs who are hard at work creating Darwinian universes inhabited by software critters that can pass along complex genetic code to their offspring and evolve through thousands of generations in a few hours. This work may someday lead to a better understanding of computer viruses or maybe even human evolution.
Levy takes various passes at the basic question: What is life, anyway? In what may be the ultimate in political correctness, he quotes researchers who argue that the human definition of what constitutes life is biased because it is unquestioningly human-centered. Who says living organisms have to be carbon-based? Why can't they exist within a computer and subsist on data? "These other forms of life, artificial ones, want to come into existence," says Chris Langton, an impassioned pioneer of the field. "And they are using me as their vehicle."
Man gives birth to mutant ninja software? Well, it's fun to contemplate whether humans can bestow their traits on the sterile 1s and 0s of computer programs. Such thoughts force us to appreciate our own complexities. In this respect, it's not hard to see how a writer can get carried away. Jubak doesn't. As a result, In the Image of the Brain provides a sober snapshot of what's going on now in the fascinating field of neural networks. In Artificial Life, Levy clearly overreaches, but he provides an eloquent discussion of the consequences if the people he profiles happen to be right. Both books portend intriguing things to come.EVAN I. SCHWARTZ