Frank Rosenblatt Publishes Principles of Neurodynamics

Ah, the perceptron, one of the most (in)famous words in AI history. As mentioned previously in this timeline, Frank Rosenblatt first described the Perceptron in 1958.

In 1962 Spartan Books in Washington published Frank Rosenblatt's Principles of Neurodynamics: Perceptrons and the Theory of Brain Mechanisms. This 616-page tome was an account of the perceptron research he had begun at the Cornell Aeronautical Laboratory in 1957.

The book set out the theory and experiments behind networks of simple neuron-like units that learn to classify patterns by adjusting their connections after each error, and it proved that such learning succeeds whenever a solution exists. Rosenblatt also discussed networks with more than one layer of adjustable units, which he could not yet train reliably.

Due to this unreliability in multi-layer perceptrons especially, seven years later in 1969 Minsky and Papert's Perceptrons supposedly disproved the effectiveness of perceptrons and thus neural networks. The impact of this cannot be overstated, as it essentially split the field into connectionism, favoring neural nets and Rosenblatt's perceptron, and symbolism, favoring hard-coded expert systems.

It was the learning versus rules AI split that wouldn't really resolve until around 2010. For decades, neural networks were disavowed and denied funding, largely because "Oh, hadn't they already been disproven by Minsky?"

They weren't and, of course, as you know. The connectionist/perceptron/neural net/Rosenblatt "side" would end up "winning" and is known today as deep learning.