“Steps Toward Artificial Intelligence” is a 1961 survey paper by Marvin Minsky, published in the Proceedings of the IRE (volume 49, pages 8 to 30) and hosted in full on an MIT-archived copy of his papers. Written just a few years after the 1956 Dartmouth workshop that named the field, it was an early attempt to organize the scattered work into a coherent agenda.
Minsky divided the problem of building intelligent machines into five areas. Search covered how a system explores the space of possible solutions without examining every one, and the heuristics that prune that space. Pattern recognition covered how a system classifies a situation so it can pick an appropriate method. Learning covered how a system improves from experience, including the credit-assignment problem of deciding which earlier decisions deserve reward when success finally arrives. Planning covered breaking a hard problem into more tractable subproblems. Induction covered the deepest and least understood area, how a system might form genuinely new general concepts.
The paper is notable for how much of the later field it anticipated in outline, and for its extensive bibliography, which gathered the early literature in one place. Minsky framed AI as fundamentally about overcoming combinatorial explosion - the way the number of possibilities grows beyond any brute-force reach - through cleverness about which possibilities to consider.
It stands as one of the founding survey documents of artificial intelligence, a snapshot of the questions the discipline set itself at the start, several of which - especially learning and the formation of new concepts - remain central to the field today.