RAND Corporation

Contribution to AI

The first programme anyone was willing to describe as reasoning ran on JOHNNIAC, the copy of von Neumann's machine design that the corporation finished building for itself in 1953. Logic Theorist, written there by Allen Newell, Cliff Shaw and Herbert Simon across 1955 and 1956, proved theorems from Principia Mathematica by heuristic search rather than exhaustive enumeration, and Newell and Simon carried it to the Dartmouth meeting that summer as the only running system in the room. Making it work meant inventing the tools underneath it: the Information Processing Languages brought list structures, dynamic memory allocation and recursion into practice, McCarthy drew on that idea for Lisp, and the same machinery carried General Problem Solver in 1957 and the NSS chess programme in 1958. The mathematics came from a different corridor. Richard Bellman's work in the early 1950s produced the recursion over value functions that reinforcement learning still computes, along with the phrase curse of dimensionality for why it is hard, and a 1950 experiment by Merrill Flood and Melvin Dresher yielded the payoff matrix Albert Tucker later retold as the prisoner's dilemma. The scepticism was homegrown too: Hubert Dreyfus wrote Alchemy and Artificial Intelligence as a consultant in 1965, and the institution funding symbolic AI's best results also published the first serious argument that they would not generalise.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).