The question or thesis
Artificial intelligence in games means more than enemy behaviour: it includes NPC control, search and planning, procedural generation, player modelling, and machine-learning-driven adaptation. For most students, the best learning order is authored agent AI first, then search and planning methods, then broader AI-and-games topics such as generation, modelling, and learning.
What the evidence suggests
Three broad layers recur across the books:
-
Authored agent behaviour Includes ai-state-machine-pattern, behaviour-trees, utility-ai, goal-oriented-action-planning, and steering-behaviours. This is the practical production core of classic game AI. (Millington, Artificial Intelligence for Games, see source-artificial-intelligence-for-games; Buckland, Programming Game AI by Example, see source-programming-game-ai-by-example)
-
Search, optimisation, and planning Includes pathfinding-algorithms, GOAP-style planners, minimax/MCTS, genetic-algorithms, particle-swarm-optimisation, and ant-colony-optimisation. These methods treat play or design as search through a space of possible actions or solutions. (Bhargava, Grokking Algorithms, see source-grokking-algorithms; Brabazon et al., Natural Computing Algorithms, see source-natural-computing-algorithms)
-
Learning, generation, and modelling Includes machine-learning-games, neuroevolution, procedural-generation, and player-modelling. This is the broader field view emphasised by Yannakakis and Togelius, where AI in games also means adapting content, understanding players, and using games as AI benchmarks. (Yannakakis and Togelius, Artificial Intelligence and Games, see source-ai-and-games)
Disagreements or tensions
- Optimality vs fun: Stronger AI is not automatically better game AI. Several sources stress visibility, readability, and player experience over raw competence.
- Authored control vs learned behaviour: Authored systems are easier to debug and ship; learned systems can generalise better but are harder to control.
- Research breadth vs production practicality: Academic surveys treat PCG and player modelling as central; production textbooks spend more time on movement, pathfinding, and decision logic.
What to investigate next
- A dedicated synthesis linking player-modelling to game-analytics and self-determination-theory
- Worked Unity examples for behaviour-trees, goal-oriented-action-planning, and fuzzy-logic-for-games
- A clearer lecturer-facing bridge between classical game AI and modern generative/ML workflows
- A course-facing reading route for CRE341 that distinguishes core agent-control architectures from adjacent AI-and-games topics
Related
game-ai-agent-design · overview-cre341-agent-ai-route · machine-learning-games · procedural-generation · player-modelling · source-artificial-intelligence-for-games · source-ai-and-games