Perceptrons – An Introduction to Computational Geometry: The MIT Press

Reissue of the 1988 Expanded Edition with a new foreword by L on Bottou

In 1969, ten years after the discovery of the perceptron--which showed that a machine could be taught to perform certain tasks using examples--Marvin Minsky and Seymour Papert published Perceptrons, their analysis of the computational capabilities of perceptrons for specific tasks. As L on Bottou writes in his foreword to this edition, "Their rigorous work and brilliant technique does not make the perceptron look very good." Perhaps as a result, research turned away from the perceptron. Then the pendulum swung back, and machine learning became the fastest-growing field in computer science. Minsky and Papert's insistence on its theoretical foundations is newly relevant.

Perceptrons--the first systematic study of parallelism in computation--marked a historic turn in artificial intelligence, returning to the idea that intelligence might emerge from the activity of networks of neuron-like entities. Minsky and Papert provided mathematical analysis that showed the limitations of a class of computing machines that could be considered as models of the brain. Minsky and Papert added a new chapter in 1987 in which they discuss the state of parallel computers, and note a central theoretical challenge: reaching a deeper understanding of how "objects" or "agents" with individuality can emerge in a network. Progress in this area would link connectionism with what the authors have called "society theories of mind."