The PerceptronThe Perceptron

In 1958, Cornell psychologist Frank Rosenblatt demonstrated the perceptron, a system that learned to classify patterns by adjusting numeric weights when it made mistakes. Instead of being programmed with rules, it was trained on examples, an idea that seemed almost heretical at the time.

The perceptron was modeled loosely on a neuron. Inputs were multiplied by weights, summed, and passed through a threshold to produce a yes-or-no output. Rosenblatt proved that if a clean boundary between two classes existed, his simple learning rule was guaranteed to find it.

Funded by the US Navy, Rosenblatt eventually built the Mark I Perceptron, a room-sized machine with a 20 by 20 camera for an eye and weights implemented as motor-driven potentiometers. At a 1958 press conference, the New York Times reported the Navy's expectation of machines that would walk, talk, see, write, and be conscious of their existence.

The hype set up a fall. In 1969, Marvin Minsky and Seymour Papert published 'Perceptrons', a rigorous analysis showing that single-layer perceptrons could not compute functions as simple as XOR. Funding for neural network research collapsed, and the field entered a long winter.

Rosenblatt had actually anticipated that multi-layer networks were the answer, but no one knew how to train them, and he did not live to see the problem solved. He died in a sailing accident in 1971, on his 43rd birthday.

When backpropagation revived multi-layer networks in the 1980s, it vindicated his core bet: that learning from data, not hand-written rules, was the road to machine intelligence. Every modern neural network is a descendant of the machine the Navy oversold in 1958.