← Back to Home
ON THIS DAY 2026-10-18

1946: The Second Macy Conference on Cybernetics Convenes

On 17–18 October 1946, a small and unusually eclectic group of scholars gathered at the Beekman Hotel in New York for the second of what would become ten legendary Macy Conferences. Officially titled "Feedback Mechanisms and Circular Causal Systems in Biological and Social Systems," the meeting brought together mathematicians, neurophysiologists, engineers, anthropologists, and psychiatrists around a single provocative idea: that the same principles of feedback and control might explain machines, organisms, and societies alike.

Among those present were Norbert Wiener, who would coin the word "cybernetics" two years later; John von Neumann, already sketching the architecture of the stored-program computer; Warren McCulloch and Walter Pitts, whose 1943 paper on idealized neural networks had shown that simple logical "neurons" could in principle compute anything a Turing machine could; and anthropologists Margaret Mead and Gregory Bateson, who pushed the engineers to think about communication in human systems as rigorously as in electrical ones.

The Macy meetings were less a single discovery than a crucible. Across ten sessions through 1953, participants hammered out a shared vocabulary — feedback, information, noise, homeostasis — that dissolved old boundaries between the life sciences and the engineering sciences. It was this cross-pollination, concentrated in rooms like the one on 18 October 1946, that gave the postwar study of mind-as-mechanism its first real institutional home, years before the word "artificial intelligence" existed.

The intellectual debts run directly into AI's founding generation. McCulloch and Pitts's formal neuron became the ancestor of Frank Rosenblatt's perceptron and, much later, of the artificial neurons stacked by the millions inside today's deep neural networks. Wiener's feedback loops anticipated reinforcement learning's reward signals. Even the Macy group's insistence on studying information independent of its physical substrate — mind or machine, it need not matter — is the same assumption that lets today's large language models be judged on what they do rather than what they are made of.

When engineers at the Dartmouth workshop coined "artificial intelligence" a decade later, they were, whether they credited it or not, formalizing a research program the Macy conferees had already begun arguing into existence around a hotel conference table.