Adversarial search and minimax
**Adversarial search** plans moves when another agent tries to beat you—classically board games. **Minimax** assumes both sides play optimally: you maximize your score; the opponent minimizes it.
What it is
Adversarial search plans moves when another agent tries to beat you—classically board games. Minimax assumes both sides play optimally: you maximize your score; the opponent minimizes it.
Why it matters
It is the cleanest intro to decision-making under competition—and a bridge to modern game AIs that mix search with learned evaluations.
How it works (plain)
Build a tree of moves. At your turn pick the best for you; at their turn pick the worst for you. Depth limits + heuristics score unfinished games. Alpha-beta pruning skips branches that cannot change the decision.
Everyday example
Choosing a negotiation offer while anticipating the other party’s counter—simplified and turn-based.
Try it
On a tic-tac-toe empty board, list your first move options and the opponent’s replies for one line of play.
Myths
- ⚠️ Myth: Minimax needs perfect whole-game search to be useful.
- ✓ Reality: Bounded depth + evaluation functions built decades of strong play.
- ⚠️ Myth: Game search is unrelated to modern AI.
- ✓ Reality: Search + learned value/policy nets still powers many game systems.
Sources
- Russell & Norvig AIMA (games chapter; cite edition you use)
- CS50 AI search/game materials (verify): https://cs50.harvard.edu/ai/ ↗
- Course 06 search-and-planning