TD-Gammon (Gerald Tesauro, 1990s) is roughly equal with top human players (e.g. Malcolm Davis).

• Performance is near optimal
• Program learns by playing (using neural nets)
• Handles "luck" (probability) surprisingly well