Pattern recognition and modern computers
Selfridge posed recognition as a problem of extracting features, and showed that features can be computed in parallel and independently.
Why it matters
Instead of one deciding rule came the idea of many simple detectors whose votes are combined.
The talk precedes both the perceptron and the same author's Pandemonium. Selfridge insisted that a machine must first turn an image into a set of features, and that choosing the features is the real problem. The framing returned in convolutional networks, where the features are no longer hand-specified.