The Journey of AI in Games: A Technical and Philosophical overview
summary
Argues that modern games still rely on rule-based systems and behavior trees rather than large neural models, demonstrates a live Unreal Engine 5.1 demo of two unscripted bots whose behavior tree branches (patrol, investigate, attack, retreat) work reliably, cites A and Minimax as proven tools and OpenAI Dota as an experimental win, warns that true LLM-driven NPCs hit limits — 400 GB VRAM vs 16 GB consoles*, network latency, nonsensical quests — and finds generative AI useful for concept art, story and prototyping but not yet for core gameplay.
AI-written summary · about
chapters (13)
- 0:00AI in games introduction
- 3:17Misconceptions about AI
- 9:01Why advanced AI isn't used
- 10:08Rule-based systems in games
- 14:24Cheating AI in retro games
- 15:27Finite state machines in bots
- 20:07Behavior trees demo
- 29:56Why behavior trees dominate
- 35:00Advanced AI algorithms
- 39:24Neural networks for training
- 41:13Future of AI-driven NPCs
- 47:18AI tools for production
- 50:28Summary
related talks
29mThe Present & Future of AI and Games
01h 01mIntroduction to Practical AI
51mWhat AI Can, Can’t, and Shouldn’t Do for Games
01h 01mThe Forgotten AI of Video Games
46mMind your NPCs and Q’s: what’s next for Game AI?
45mPanel: Can Generative AI be infused in Gaming?
02h 02mAI & Video Games, from Origins to Today: a (neural) link to the past
32mPanel: The State Of AI: Sustainable Or A Bubble Waiting To Burst?