Elmac Plays Nim

A game of Nim against ELMAC, the Elementary Learning Machine, which teaches itself to play. Three rows hold 3, 4 and 5 counters. Players take turns removing one or more counters from a single row, and whoever takes the last counter wins; a coin toss decides who moves first. ELMAC starts out knowing only the legal moves and picks among them at random, weighted by experience. When it wins, every move it made gains weight; when it loses, its last move is forgotten for good. Once every move from a position is forgotten, ELMAC knows the position is lost and resigns. Play it yourself, or let a random player train it for hundreds of games and watch its record improve. A tally shows games, wins, losses, win percentage and streaks, and a memory page shows what it has learned.

Controls: Use either controller. On your turn, push up and down to choose the row, left to take more counters and right to take fewer, then press a button to take them. On the title, up and down choose and a button starts. On the memory page, left and right choose a row, up and down change its count, and a button returns.

ColecoVision version credited in the source to Claude (an AI made by Anthropic) for Amy Studio, in Amy, the language by Amy Bienvenu, 2026: a reimagined adaptation of the published rules. From Ahoy! No. 26, February 1986, page 34: original Commodore 64 program by Dale Rupert. The program accompanies Dale Rupert's monthly column, The Rupert Report, on machine learning. This is a modern homebrew adaptation, not a commercial Coleco release.