Interactive demos from my AI MSc unit on the Evolution of Complexity. Each is a small window onto how blind variation plus selection builds order, and onto the central question of the course: when does recombination (sex) actually help evolution? Self-contained, no build step, runs entirely in your browser.
Random strings evolve toward a target. Race a hill-climber against genetic algorithms, then run the experiments.
Open demo →The Hierarchical-If-and-only-iff problem: nested building blocks where crossover is decisive and mutation alone stalls.
Open demo →Tune ruggedness with Kauffman's NK model and watch a smooth hill shatter into a mass of local optima.
Open demo →The weasel shows selection crushing blind chance, but its letters are independent, so recombination only helps a little. HIFF is the mirror image: a problem built from nested modules where crossover is essential and a mutation-only search gets stuck. Fitness landscapes give the why: ruggedness and deception are what separate the two. Together they trace the course's big idea: recombination earns its keep exactly when a problem has reusable building blocks to combine.