High society on the brain.

High society in the brain

Computer scientist Marvin Minsky of the Massachusetts Institute of Technology readily acknowledges that many researchers working on neural networks -- computer models designed to simulate the behavior of small groups of neurons involved in functions such as seeing, smelling and even speaking -- consider him "the devil." In 1969, Minsky and a coleague demonstrated that elementary neural networks popular at the time could not identify certain simple patterns. Shortly thereafter, work in this field slowed to a trickle.

Thanks to more sophisticated computer approaches, neural network research has revived in the last few years. And despite his reputation, Minsky says biological neural networks, or assemblies of brain cells responsible for various simple activities, are an integral part of his theory of mind.

"The mind uses all sorts of neural networks together, and each one is good at certain things," he maintains. In Minsky's view, millions of distinct networks work together, enabling people to think and behave. A special class of cell groups records what other networks do, activates memories and allows higher-level networks to call on the services of lower-order cell assemblies. This teeming "society" of networks is responsible for producing goal-directed behavior, Minsky argues.

But computer models of brain function still do not provide a clear picture of Minsky's "society of mind." Artificial neural networks can learn to recognize visual patterns or understand everyday speech, he says, "but sometimes you have a devil of a time figuring out how they did it. We still don't have a good way of characterizing the proper questions for neural networks."