The History of AI

SNARC

Era 2 · AI is born (1943–1979) › Neural networks and learning

Lineage
Neurons and networks
Kind
Implementation

Who

Harvard University, Princeton University

What happened

In the summer of 1951, at Harvard, Marvin Minsky and Dean Edmonds, graduate students at Princeton, build SNARC (Stochastic Neural Analog Reinforcement Calculator), a network of 40 artificial neurons made of vacuum tubes, motors, and clutches.

Why it mattered

One of the first neural network machines able to learn: when it receives a reward, it strengthens the connections that have just been used, and so finds its way out of a maze.

The full story

Minsky, a mathematics student at Princeton, wanted to build a machine that learned the way animals do in conditioning experiments. The Harvard psychologist George Miller secured a few thousand dollars from the Office of Naval Research, and Minsky recruited Dean Edmonds, a physics student with a knack for electronics.

The machine had about forty randomly connected units, each with about six vacuum tubes and a knob that set the probability that a signal would pass through. When a signal crossed a connection, a capacitor remembered it for a few seconds. If the operator pressed the reward button within that interval, a motor taken from the autopilot of a B-24 bomber moved a chain that raised the probability of the connections just used. It was, in copper and vacuum tubes, a version of Hebb's rule combined with reinforcement.

The task was to simulate rats in a maze, represented by lights. With rewards, the rats made fewer and fewer mistakes, and the builders noticed that when one rat found a good path, the others tended to follow it.

SNARC was an isolated experiment. Minsky came to think that networks of this kind were too limited and moved over to symbolic artificial intelligence, becoming one of its founders.

Videos

Builds on

  • Hebb's rule (1949)

    Inherited, with no proven direct link

    SNARC strengthens the connections that have just fired, a mechanical version of Hebb's rule combined with reward

Threads that pass through here

Sources

How to cite

“SNARC.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/snarc

This entry was last reviewed on November 13, 2026.

Open in the interactive timeline