Around 1950, working at Bell Labs, Claude Shannon built a demonstration he called Theseus, after the Greek hero who found his way out of the labyrinth. The “mouse” was a small magnet-tipped figure on wheels that ran through a maze of movable partitions laid out on a board of 25 squares. The intelligence was not in the mouse but underneath the board: a large bank of telephone relays recorded each turn the mouse made, and an electromagnet on a moving carriage beneath the maze pulled the mouse along.
On its first run the mouse explored by trial and error, bumping into walls and trying directions until it reached a brass target. The relays stored the successful path. Placed back at the start, Theseus then ran the whole maze without a single wrong turn, retracing the route it had learned. If the maze was changed, the mouse would detect that its remembered path no longer worked and learn the new route. The original machine is preserved at the MIT Museum, and a Bell Labs film of Shannon demonstrating it survives.
Theseus is one of the earliest public demonstrations of a machine that learned from experience and stored what it learned. Built from ordinary telephone relays just a few years after the first electronic computers, it made the abstract idea of machine learning tangible: a device that tries, remembers, and then succeeds. It anticipated, in hardware, the maze-running learning experiments of the 1950s and the reinforcement-learning systems that followed.