The 1969 Book that Killed AI Research

The neural nets at the core of AI were invented in 1943, and popularized in 1957-60 as the Perceptron. It had the basic idea that was later developed into today's AI LLMs.

But not everyone recognized it. Prominent MIT professors wrote a 1969 book, Perceptrons: An Introduction to Computational Geometry, trashing the concept.

Reports by the New York Times and statements by Rosenblatt claimed that neural nets would soon be able to see images, beat humans at chess, and reproduce. ...

This book is the center of a long-standing controversy in the study of artificial intelligence. It is claimed that pessimistic predictions made by the authors were responsible for a change in the direction of research in AI, concentrating efforts on so-called "symbolic" systems, a line of research that petered out and contributed to the so-called AI winter of the 1980s, when AI's promise was not realized.

The book claimed to prove that perceptrons were hopelessly limited, as they could not even approximate a simple xor function.

The authors doubled down in 1988 updated edition, saying "little of significance [has] changed since 1969". At the time there was a revival of interest in neural nets, but the authors said they would fail to scale up.

We now know that those neural nets of the 1940s, 50s, and 80s do scale up, and perform spectacularly well. The bold predictions of the perceptron inventor must have been seen as wildly optimistic at the time, but they turned out to be correct.

I post this as an example of how smart people can be wrong about what is possible, and how brilliant research could be sidetracked.

Rosenblatt made a hardware computer perceptron in 1960, and that is now in the Smithsonian Museum. What no one knew until about 1912 was that video game chips would be the key to scaling up neural nets.

Here is a recent video that tries to relate these old ideas about neural nets to current research.