Most of the discussion of artificial intelligence that has filtered through the mass media concerns computers. Freedman begins there, explaining how scientists have used computers to emulate human and animal activity. He offers a reasonably complete short description of high and low levels of thinking (reasoning versus developing heuristics), expert systems that apply rules provided by human experts, and mobile machines that learn how to navigate. Mobile machines, he notes, generally performed better when their programming was reactive rather than reasoning. The reasoning machines were slow and quickly became lost once they veered slightly from their planned course; reactive machines, on the other hand, learned to recognize patterns and developed ways of overcoming obstacles to their motion. Low-level reasoning thus appeared more appropriate for this aspect of dealing with a real environment, as opposed to a model presented to the computer.

Freedman proceeds to a discussion of the cutting edge of artificial intelligence research, offering speculations on where the next major discoveries will be made. One main line of study is neural networks, in which responses in one part of a computer trigger responses in another, much as neurons work in the human brain. One Japanese researcher has found a way to photograph how neurons fire in a rat’s brain and hopes to be able to wire a computer to react the same way. Another project programmed enormous amounts of data into a computer and provided some links among various facts. The computer was programmed to make its own connections and was able to write analogies relatively quickly.

Other research has turned to biology. Bioengineers have discovered how various cells work in nature and are trying to use those functions in concert with mechanical devices, thus developing partly organic computers. Programmers have written computer code that can “evolve,” testing programs for their ability to solve certain problems and then “mating” the best programs, exchanging lines of code in “offspring” programs with occasional mutations thrown in. Such evolved programs in some cases were better at their tasks than were programs written by human programmers. The most promising avenues for research, Freedman suggests, lie in discovering how machines can emulate nature rather than resorting to pure logic.