Minimax explores a game tree where MAX (the AI) and MIN (the opponent) alternate turns, each picking the move that is best for them. Alpha-beta pruning skips branches that provably cannot influence the final decision.
MAX: α = max(α, value)
MIN: β = min(β, value)
if α ≥ β → prune remaining siblings
- Search depth — how many plies (half-moves) ahead the AI looks before scoring a position.
- Branching factor — legal moves per position; chess ≈35, Go ≈250 (why MCTS replaced plain minimax there).
- α (alpha) — the best score MAX can already guarantee; β (beta) — the best MIN can already guarantee.
- Pruning — once α ≥ β, the rest of that branch is skipped: no move there could change the parent's decision.
This exhaustive, deterministic search is what powered classic board-game AI (chess engines, checkers). Modern Game AI adds Monte Carlo Tree Search for huge branching factors (Go/AlphaGo), procedural content generation for levels, and behavior trees for NPCs — minimax's alpha-beta pruning is still the backbone of turn-based game engines and AI QA bots that explore game states exhaustively.