Perceptrons
Minsky & Papert, 1969 — a rigorous mathematical critique of the single-layer perceptron that stalled neural network research funding for most of a decade, the first AI winter.
Perceptrons (Marvin Minsky and Seymour Papert, 1969) is not a research paper but a book — and one of the most consequential publications in the field's history regardless, for what it took away rather than what it added: it's the single publication most credited with triggering the first AI winter.
What problem it solved
By the late 1960s, Frank RosenblattFrank RosenblattPsychologist who built the perceptron in 1958, the first neural network that learned its own weights from data rather than having them set by hand.'s perceptronPerceptronThe Perceptron (1958) was the first learning system built from an artificial neuron, and the direct ancestor of the neural network training loop. had generated significant hype — press coverage suggested machines that could see, speak, and think were close at hand. Minsky and Papert set out to answer a narrower, more rigorous question that the hype had outrun: what, precisely, can a single-layer perceptron actually compute?
The key idea
Prove it mathematically rather than argue it informally. The book's central result shows that a single-layer perceptron can only learn functions that are linearly separable — representable by a single straight-line (or hyperplane) decision boundary — and demonstrates that even a function as simple as XOR falls outside that class. No amount of training data or training time lets a single-layer perceptron learn XOR, because the limitation is structural, not a matter of optimization.
Why it mattered
Funding for connectionist (neural network) research collapsed for most of a decade following the book's publication — the period historians call the first AI winter for neural networks specifically, as research attention shifted toward symbolic AI approaches. The irony the field returned to repeatedly afterward: the book's proof was correct, and the fix (hidden layers, trained with backpropagation) was exactly what its own text gestured at as a theoretical possibility. It's a standing reminder that a rigorous negative result about one specific architecture can be misread as a verdict on an entire research direction — and that Marvin MinskyMarvin MinskyCo-founder of the MIT AI Lab and a founding figure of the field — his 1969 book with Seymour Papert exposed the perceptron's limits and helped trigger the first AI winter., one of the field's founders, is also the person most associated with nearly ending one of its most important sub-fields before it had the tools to succeed.
Authors: Marvin Minsky, Seymour Papert (MIT)
See also: Marvin MinskyMarvin MinskyCo-founder of the MIT AI Lab and a founding figure of the field — his 1969 book with Seymour Papert exposed the perceptron's limits and helped trigger the first AI winter. · Frank RosenblattFrank RosenblattPsychologist who built the perceptron in 1958, the first neural network that learned its own weights from data rather than having them set by hand.
Learn more: PerceptronPerceptronThe Perceptron (1958) was the first learning system built from an artificial neuron, and the direct ancestor of the neural network training loop. · History & Landscape · Wikipedia: Perceptrons (book)
Mentioned in
Lessons where this comes up in context.
Introduction
Plain-language explainers for the research papers this course cites most — what problem each one solved, the key idea, and why it mattered.
Efficient Estimation of Word Representations in Vector Space
Mikolov et al., 2013 — showed that a shallow, cheaply-trained neural network could turn words into vectors that captured meaning, kicking off the embedding era that everything from search to LLMs now depends on.