One Book, and One Winter
The book Perceptrons used mathematical proof to point out that Frank Rosenblatt’s single-layer perceptron could not learn simple nonlinear logical operations such as "XOR," a conclusion the academic world of the time widely read as "the neural-network path is a dead end," directly causing this field to be all but cut off from research funding for over a decade, the first "AI winter." Minsky later said this widespread pessimistic reading in a sense misread his intent — the book actually also mentioned that multi-layer networks could in theory overcome this limit, but at the time there was neither enough compute nor an effective way to train multi-layer networks, a gap not filled until the backpropagation algorithm was rediscovered over a decade later.