“Alchemy and Artificial Intelligence” is a December 1965 memorandum (RAND paper P-3244, 94 pages) by the philosopher Hubert Dreyfus, written while he was a consultant to the RAND Corporation. A scan of the original is hosted by the Computer History Museum, courtesy of Dreyfus himself. It was the first sustained philosophical attack on the assumptions of artificial intelligence, and its tone made it notorious within the field.
Dreyfus identified what he called a recurrent pattern in AI research: early, dramatic success on a simplified version of a problem, followed by sudden and unexpected difficulty when researchers try to scale up to the real thing. He surveyed the leading efforts of the day - machine translation, game playing, problem solving, and pattern recognition - and argued that each had hit exactly this wall. The alchemy metaphor was deliberate: like alchemists who turned base metals slightly more gold-colored and concluded they were on the path to transmutation, AI researchers were mistaking local tricks for progress toward general intelligence.
His underlying claim, developed at length in his later book “What Computers Can’t Do,” was that human intelligence depends on embodiment, context, and tacit background knowledge that resists being written down as explicit symbolic rules - the very thing the dominant approach assumed it could do. The memo was received badly; RAND reportedly hesitated over publishing it, and parts of the AI community treated it as an attack rather than an argument.
The paper matters historically because the difficulties Dreyfus pointed to - brittleness, the explosion of common-sense knowledge required, the gap between toy problems and the open world - were among the forces that produced the AI winters and that the field is, in different form, still working through.