The Present & Future of AI and Games
summary
Historical game AI relies on behavior trees and navigation meshes for pathing while studios still reinvent bespoke solutions; modern shifts claim deep learning handles pattern recognition, animation controllers and player modeling. Forza’s Drivatar uses Bayesian learning and trains on microsoft azure servers since 2013, MotoGP 2019 ships a neural driver codenamed Anna, and Valve’s VACNet boosts cheat-detection accuracy to ~80% versus ~30% for raw player reports. AlphaStar-scale training costs tens of millions and most projects spend ~90% effort on preparing and training models.
AI-written summary · about
chapters (26)
- 0:00Introduction
- 0:51About AI and Games
- 2:09What is AI
- 4:32AI in games
- 5:56Complex multidisciplinary process
- 7:03Building better bad guys
- 7:42Finite State Machines
- 8:46Behavior Trees
- 9:16GoalOriented Action Planning
- 10:21Hierarchical Task Network Planning
- 11:01The AI Problem
- 13:08Machine Learning
- 13:43How Machine Learning Works
- 15:18What is Deep Learning
- 16:44Deep Learning Tools
- 17:19Training
- 18:51Andrew Webb
- 19:52bayesian learning
- 21:31animation controllers
- 21:58motion matching
- 22:47lip syncing
- 23:15cheat detection
- 24:31candy crush
- 25:10other innovations
- 26:42conclusion
- 28:29outro
related talks
48mMachine Takeover - Improving Your Game With Machine Learning & AI
52mThe Journey of AI in Games: A Technical and Philosophical overview
01h 01mIntroduction to Practical AI
22mDeveloping AI in Racing Games Using Reinforcement Learning
41mThe Current State Of AI In Games
01h 01mThe Forgotten AI of Video Games
52mPhysical AI: Crafting Resilient Systems with Emotional Intelligence
56mCan I do AI?