Four in a Row

You are blue. The AI is a small neural network that first learned this game from about 12,000 games played by an exact solver, then improved by playing thousands of games against itself. It picks its moves the way AlphaZero does: the network's hunch guides a search, and the search decides. While it thinks, the panel beside the board shows that search tree growing, its raw instinct next to what the search concludes, and the win odds. Turn the strength down to Instinct and it plays on the hunch alone, which sometimes misses a mate in one. The search is what fixes that.

Policy and value network with PUCT search, trained by self-play following the AlphaZero method (Silver et al., 2017, DeepMind). All code here is original and runs in your browser.