In 1969 Marvin Minsky and Seymour Papert published the book “Perceptrons: An Introduction to Computational Geometry” with MIT Press. Only the year is documented for the original edition; the archived copy cited here is the later expanded MIT Press edition of the same work. The book gave a rigorous mathematical analysis of what single-layer perceptrons could and could not do.
Their most famous result showed that a simple perceptron cannot learn the exclusive-or (XOR) function, a basic logical operation. More broadly, they demonstrated that single-layer networks were fundamentally limited in the kinds of patterns they could distinguish. While multi-layer networks could in principle do more, no good method was then known to train them.
The book is widely credited with cooling the enthusiasm around Rosenblatt’s perceptron and helping bring on the first “neural network winter,” a long stretch of reduced funding and interest. Neural networks would not fully recover until the backpropagation algorithm made multi-layer networks trainable in the 1980s.