- Slides: 46
Now you see it - Now you don’t Intro to Game AI
Book Learnin’ l University of Chicago – – l BA in General Studies Masters work in AI & Information Systems Northwestern University – Ph. D work in AI: Robotics & Video Games
Street Smarts l l l Game Design and Tuning Workshop Indie Game Jam/Experimental Gameplay Workshop IGDA Education Committee . . . and kickin’ it with developers, in general
What is AI? Some terms
AI is. . . l l l a philosophy an abstraction a tool
Philosophy l People do things based on rough goals and sketchy plans – modifying their behavior when problems or opportunities arise.
Tool: Finite State Machine start end
Tool: Planner state goal
What is Game AI? Some Context
Environment l l l Big projects, getting bigger Dynamic worlds/systems Time is still an issue
Problems l l Pathfinding Sensing Reacting Expressing
People l l l Programmers Designers Players
Goals: Programmers l l Intelligent, coherent, reactive behavior Consistent, accurate sensing
Goals: Designers l l Extensible, flexible code Intuitive interface
Goals: Players l l l Coherent Believable Entertaining! Fortress of Solitude: “If they worked properly, they’d be tools - not toys”
Currently l Slow progress on – – l sensing reacting Problems crop up in – – code interface
State of the Art Game AI Today
State of Game AI? l “State machines and hierarchical AIs. The simple rules-based finite- and fuzzy-state machines (FSMs and Fu. SMs) continue to be the tools of choice for developers, overshadowing more "academic" technologies such as neural networks and genetic algorithms. ” – Woodcock 2001
Finite State Machines l Systems, Sub-systems and Sub-Sub systems – – Proliferation of nodes = slower code Changes in Agent design = chaos Schizophrenic behavior = common Difficult to debug
Rule-Based Systems l Special Cases made easy – – – Scripting can be tedious Debugging can be difficult Flexibility? Extensibility?
Hierarchical AI l Commanders, Middlemen, Drones – Lots of reasoning Lots of inter-level communication What does the user see? – Best in strategy contexts? – –
Other approaches? l Black and White, Creatures, The Sims – – Neural Net Learning/Reinforcement Transparency? (why? ) Transitions? (how? )
Other approaches? l Thief I & II – – Complex Sensing Brooksian, Distributed AI Scalability? (how many brains? ) Intuitiveness? (which brain? )
More Recently l Optimizations – l Blackboard Systems (NOLF 2) Broadening scope – – Diagnostics (DX 2) Expressive Characters (HL 2) l Paul Eckman
Recent Issues l New systems and genres introduce new problems. . . – – – Dynamic Path Planning Strategic Team-based AI Massively-Multiplayer AI
Next-Generation AI? l l Will new technology engender new and improved techniques? Will new problems have clear, simple solutions?
What about Middleware? Game AI In a Box?
Middleware l Animation & Locomotion Control
Middleware l Behavior Control Systems
Middleware l Structural Organization/Interface
Middleware l l Hard to get developers to use it Not clear what value it provides – – l are they the right abstractions? will the tools be flexible enough? AI seems specific to game genre/content
Looking Forward Juicy AI Problems
Dynamic Game Adjustment l l Difficulty Pacing Story Characters – The future looks. . . complicated!
Natural Language Generation l l l Context-aware conversations Common-sense responses Dynamic phrase generation
Expressive Characters l l l Memory Feelings Independent goals
Persistence l Character Changes, World Changes – Chess Examples
Persistence “There are no character classes in Project Ego, only personality styles that evolve according to the decisions you make. Your character will age as time progresses, and the actions you take put an idiosyncratic spin on his or her appearance. Early wounds will show up later as scars. Lethargy will manifest as a little extra blubber around your middle, and anaerobic exercise will show up in the form of Olympian muscles. Even more important than your looks, your reputation will depend on the kind of life you lead. The incredible freedom of gameplay allows you to pursue every occupation from dashing hero to indolent bum to mischievous cutpurse. ”
How would you build “Fable”? l l l Persistent beliefs about the main character Clear presentation of change/progress Characters that remember events Gossip? Grudges?
Wrapping up What we covered. . .
Game AI Then l l Big talk up to about 2001 Plenty of proposed ideas tanked/ignored – – Republic Black & White
Game AI Now l l Gamasutra/Game Developer AI Programming Wisdom
Game AI Tomorrow l l l Look to next round of games for ideas? AI ‘ 04 – workshop in the works GDC ’ 04 – hopefully some tidbits
Contact [email protected] northwestern. edu