AI: Complex things made simple

What the talk is about

Artem Durnev and a colleague explain how their team keeps game AI simple to build and maintain. They first survey the approaches: hard-coded rule sets, behavior trees, finite state machines, goal-oriented planning, utility-based AI, and machine learning with its black-box drawback. They then show two games: a team robot shooter where two AIs run in parallel (movement and shooting), and a turn-based collectible RPG that moved from rigid selectors to utility scoring. The core idea is that tools shape what designers can create, so programmers should hide complexity behind simple code that game designers can read and edit as rules. They also stress small methods, convenient logging of AI decisions, and testing game situations with SpecFlow, which testers can write too. Useful for programmers and game designers building AI.