BOOKS OF THE TIMES; Making Electronics Mimic Biology
Brainmakers How Scientists Are Moving Beyond Computers to Create a Rival to the Human Brain By David Freedman 214 pages. Simon & Schuster. $22.
Last month when Dante II, the 1,700-pound robot built by engineers at Carnegie Mellon University, misstepped, tumbled and couldn't right itself while returning from exploring the crater of an Alaskan volcano, it provided a fitting metaphor for both the progress and the limits of the three-decade-old science of artificial intelligence.
Indeed, since its inception in the late 1950's, the field has actually made remarkable progress. There are now machines that see, expert systems that advise us on tough decisions, and it is possible to pick up a telephone and chat with a computer that is standing in for a long-distance operator. There are even hand-held gadgets such as Apple Computer's Newton personal digital assistant that do a laborious, and imperfect, job of recognizing handwritten notes.
But while progress has been made in mimicking human sensory and reasoning skills, scientists' perception of the size of the job has grown much larger even as computers have gained remarkable new powers of performance.
In "Brainmakers: How Scientists Are Moving Beyond Computers to Create a Rival to the Human Brain," a Boston-based science writer, David Freedman, sets out to survey the current state of the art of artificial intelligence.
His premise is that the original artificial intelligence researchers, traditionalists as he refers to them, failed because their attempts to codify a logic representing human thought proved bankrupt. Elaborate models written in computer languages such as Lisp and Prolog were incapable of emulating the simplest kinds of human behavior.
Mr. Freedman then charts the outlines of a new movement that has emerged to carry the search for artificial intelligence forward by drawing on the biological underpinnings of human intelligence.
Scientists are now at work in what is called artificial life research, tinkering with electronic circuits that mimic biological neural networks, with software programs known as genetic algorithms that improve themselves through simulated evolution, with chemical brews that attempt to synthesize life itself, and with weird amalgams composed of biological cells grown on semiconductors.
The new studies have already yielded a great deal of useful technology: neural circuits enhance vision systems as well as speech and handwriting recognition; genetic algorithms create programs that are more efficient than the finest human-designed programs.
It is a seductive idea: a new artificial intelligence revolution can be achieved by copying the simplest biological processes and enhancing those processes with immensely powerful supercomputers and arrays of neural circuits made in the laboratory.
Mr. Freedman takes us from the heartland of the traditionalists at the Massachusetts Institute of Technology to the home of a researcher turned venture capitalist in the hills above Silicon Valley, to the University of Tsukuba in Japan, and in the process shows us a scientific world that still lacks a useful road map.
Indeed, even while Mr. Freedman draws disparaging portraits of such pioneers in artificial intelligence as Marvin Minsky and Seymour Papert, computer scientists at M.I.T., and John McCarthy of Stanford (who is described as looking and acting like a homeless person), his more affectionate portraits of the younger artificial life wizards contain enough ominous signs to warn the reader that this new generation is no closer to the holy grail of a self-conscious machine.
At one point an optimistic neural network researcher is quoted as saying: "We're now about 25 years away from a silicon brain."
Those words strike a particularly eerie note because 25 years is about the same time span that the traditionalists originally thought would be necessary to create a thinking machine.
But the real problem with "Brainmakers" is that Mr. Freedman's study lacks both journalistic depth and passion. In fact, at many points the author seems to want to distance himself from the premise of his subtitle: "How scientists are moving beyond computers to create a rival to the human brain." It would appear that revolutionary breakthroughs are not yet on the horizon.
Particularly devastating to those who are optimistic about the possibility of creating self-conscious machines is the work of Stuart Hameroff, a neurological researcher at the University of Arizona. For two decades he has studied structures inside neurons known as microtubules. These protein cylinders, which lend structure to neurons, are computing systems unto themselves, he argues, adding unimaginable complexity to the network of three billion neurons that make up the human brain.
But in a brief conclusion Mr. Freedman writes, "The nature-based A.I. movement seems exceptionally well positioned to succeed." Yet in the following paragraph he admits, "On the other hand, conventional A.I. also got off to a breathtaking start . . . It wasn't for nearly two decades that the endeavor began to lose momentum."
Mr. Freedman has provided a quick travelogue that offers a glimpse of the work of some leading researchers in artificial intelligence. Ultimately, however, the reader is left wondering why he made the journey.