Exploring Decision-Making in Artificial Intelligence
How do machines make decisions in adversarial environments? This project explores the algorithms behind game-playing AI — from minimax trees to constraint satisfaction — through interactive visualizations, empirical benchmarks, and comparative analysis.
Exhaustive game-tree search with intelligent branch elimination. Explores optimal play in two-player zero-sum games.
Try Demo →Systematic variable assignment with constraint checking. Guaranteed to find a solution through exhaustive search.
Try Demo →Backtracking enhanced with domain propagation. Eliminates 58% more nodes but at 3× the overhead cost.
Try Demo →Heuristic local search with random restarts. Achieves O(n) average-case on N-Queens — the fastest solver.
Try Demo →Information-theoretic approach: each guess eliminates ~82% of candidates. Solves any code in ≤6 guesses.
Try Demo →A mathematical algorithm that always identifies a chosen card in exactly 3 rounds through ternary subdivision — determinism replacing what looks like random guessing.
Try Demo →Algorithm efficiency depends on the pruning cost-to-benefit ratio, not just branch reduction.
Near-zero overhead → 94.2% node reduction at depth 10. The pruning cost is negligible relative to the exponential search savings.
58% fewer nodes explored, but 3× slower than basic backtracking. Domain copying overhead exceeds the search savings.
82% candidate elimination per guess. Expensive per-step computation is justified by the small, finite search space (360 codes).
Step through backtracking, forward checking, and local search
Launch Demo →Play against alpha-beta AI with live evaluation scoring
Launch Demo →Break codes or watch the AI eliminate 82% of candidates per guess
Launch Demo →Deterministic convergence through ternary subdivision in exactly 3 rounds
Launch Demo →Every game encodes a decision problem. N-Queens asks: can you eliminate contradictions early enough? Connect 4 asks: can you prune the irrelevant? Mastermind asks: can you extract maximum information from minimum data?
This project doesn't just implement algorithms — it asks when each approach breaks down, and why. The central finding: an algorithm's real-world efficiency depends not on theoretical complexity alone, but on the ratio between the cost of its optimization strategy and the savings that strategy produces.
Alpha-beta pruning costs almost nothing and saves exponentially. Forward checking costs a lot per step and saves only linearly. The math looks different. The truth is in the benchmarks.
Algorithm implementations with empirical benchmarks
Interactive visualizations, no frameworks
Real-time algorithm animation
Stochastic averaging, controlled experiments