Alchemy and Artificial Intelligence, RAND Corporation Memorandum by Hubert Dreyfus
In December 1965, Hubert Dreyfus, a philosopher at the Massachusetts Institute of Technology consulting for the RAND Corporation, published RAND Memorandum P-3244, 'Alchemy and Artificial Intelligence', arguing that the cognitive assumptions underlying contemporary AI research were philosophically untenable and that the field faced fundamental, not merely technical, limits.
Background
By the mid-1960s, the people working on artificial intelligence had reason to feel optimistic. Programs like the Logic Theorist and the General Problem Solver, built by Allen Newell and Herbert Simon at Carnegie Mellon, had demonstrated that a computer could prove mathematical theorems and solve puzzles by manipulating symbols according to formal rules. The underlying idea was that thinking, at its core, was just this kind of symbol manipulation, and that a sufficiently clever program could therefore replicate it.
The researchers who built these systems believed the remaining problems were technical rather than fundamental. Write better heuristics, meaning rules of thumb that guided a program toward a solution without trying every possibility, and intelligence would follow. The question, in their view, was not whether machines could think but how quickly the engineering could catch up with the theory.
Almost nobody outside the field was pushing back on the theory itself.
What happened
Hubert Dreyfus was a philosopher at the Massachusetts Institute of Technology, working as a consultant for the RAND Corporation. He was not an AI researcher, and he was not persuaded. In December 1965, RAND published his memorandum P-3244, titled Alchemy and Artificial Intelligence. It ran to more than a hundred pages.
His argument was not that the programs of the day were too slow or too limited in memory, though they were both. The argument was that the entire approach rested on assumptions about human cognition that could not be made to work. People do not solve problems by searching through trees of formal rules. They draw on context, on habit, on a kind of bodily familiarity with the world that nobody had ever written down and that Dreyfus doubted could be written down. He drew on the phenomenological philosophy of Maurice Merleau-Ponty and Martin Heidegger to make the case: what computers lacked was not processing power but the kind of situated, embodied understanding that gives human thought its grip on the real world.
The memorandum went through the main claims of AI research one by one and found each resting on what Dreyfus called unjustified assumptions. He grouped these into biological assumptions about how the brain worked, psychological assumptions about how people solved problems, and epistemological assumptions about what knowledge was and how it could be represented. Each layer, he argued, was shakier than the last.
RAND published it. The AI community, by most accounts, was furious. The memorandum was widely circulated partly because prominent researchers, including Marvin Minsky at MIT and Seymour Papert, attacked it publicly. That reaction brought Dreyfus’s critique to a much wider audience than a quiet RAND paper would normally reach.
Why it mattered
Dreyfus's memorandum was one of the earliest sustained philosophical critiques of the foundational assumptions of symbolic AI, drawing on phenomenological philosophy, particularly the work of Merleau-Ponty and Heidegger, to argue that human cognition depends on embodied, context-sensitive understanding that cannot be captured by formal symbol manipulation. The report provoked fierce hostility from leading AI researchers, including Marvin Minsky and Seymour Papert, which gave the critique wide visibility and forced the field to articulate and defend assumptions it had previously left implicit. Although dismissed at the time, many of its concerns (about the frame problem, the brittleness of rule-based systems, and the difficulty of encoding common-sense knowledge) were later recognised as substantive research problems.
People
Organisations
RAND Corporation, Massachusetts Institute of Technology
Sources
- Alchemy and Artificial Intelligence (RAND Memorandum P-3244).RAND Corporation (copy hosted by MIT CSAIL).Primary source
- Alchemy and Artificial Intelligence, RAND Paper P-3244.RAND Corporation.Primary source
Cite this page
AI Achievements. (1965). Alchemy and Artificial Intelligence, RAND Corporation Memorandum by Hubert Dreyfus. Retrieved 2026-08-22, from https://achievements.ai/milestone/alchemy-and-ai-by-hubert
@misc{achievements_alchemy_and_ai_by_hubert,
title = {Alchemy and Artificial Intelligence, RAND Corporation Memorandum by Hubert Dreyfus},
author = {{AI Achievements}},
year = {1965},
url = {https://achievements.ai/milestone/alchemy-and-ai-by-hubert}
}