Perceptrons

Minsky and Papert set out to examine parallel computation in its most basic form, offering the first careful mathematical treatment of the perceptron, a simple network of threshold units. For years it has stayed a foundational text on these networks.

During a period when artificial intelligence was largely built around programs running on conventional serial computers, interest was drifting back toward the notion that intelligence could arise from large webs of neuron-like elements. This book was among the first to prove, through rigorous mathematics, precisely what a class of machines that might plausibly model the brain could and could not do.

Since then, new mathematical methods, physicists' work on disordered matter, fresh psychological models of the brain, and fast computers capable of simulating networks have all raised the book's relevance. The authors' new chapter surveys today's parallel machines, traces developments since the 1972 edition, and points to research directions in connectionism. They identify a central open problem: how individual "objects" or "agents" can emerge within a network, a question that connects to their "society theories of mind."

Marvin L. Minsky is Donner Professor of Science in the Electrical Engineering and Computer Science Department at MIT. Seymour A. Papert is a professor of media technology at MIT.