The computer opponent is a neural network trained using reinforcement learning. It was exported to ONNX and now runs right here in your browser. See Connect-Zero and the follow-up posts for details.
This isn’t a particularly strong model yet: it still has many tactical and strategic blind spots. We’ll train much stronger models over the course of the series.