AI and GAMES

1
AI and GAMES
CSC 8520, Villanova University Spring, 2004 Paula
Matuszek Robin McEntire
2
Some games to explore

• Yahoo AI Games directory. Mostly chatbots 20
  questions is good.
• Yahoo Artificial Life . I like Insaniquarium.
• Chinook. The world man-machine champion checkers
  player.
• A Mastermind variant.
• Othello (Reversi). One of the classic neural net
  testbeds.
• Battleship (Armada).
• The entire Yahoo web games directory can be fun
  to explore. Which of these seem to have an AI
  component?

3
GAMES and AI

• Most of these games are using AI. Here are some
  discussion points
• Agents which of these games have autonomous
  agents? Whats the PEAS description?
• What AI techniques are being used?
• A search (eg, for path finding)
• Alpha-Beta search (eg, for choosing a move)
• Knowledge representation
• Natural language processing
• Artificial life and flocking/swarming
• Reasoning
• Learning

4
Insaniquarium

• Agents with PEAS?
• Performance measure
• Environment
• Actuators
• Sensors
• Search?
• Knowledge Representation?
• Natural Language processing?
• Logic and inference?
• Other?

5
Chinook

• Agents with PEAS?
• Performance measure
• Environment
• Actuators
• Sensors
• Search?
• Knowledge Representation?
• Natural Language processing?
• Logic and inference?
• Other?

6
Battleship

• Agents with PEAS?
• Performance measure
• Environment
• Actuators
• Sensors
• Search?
• Knowledge Representation?
• Natural Language processing?
• Logic and inference?
• Other?

7
Mastermind

• Agents with PEAS?
• Performance measure
• Environment
• Actuators
• Sensors
• Search?
• Knowledge Representation?
• Natural Language processing?
• Logic and inference?
• Other?

8
Role Playing Games Non-Player Characters
(Helpers, Opponents)

• Agents with PEAS?
• Performance measure
• Environment
• Actuators
• Sensors
• Search?
• Knowledge Representation?
• Natural Language processing?
• Logic and inference?
• Other?

9
Summary

• Even very basic AI can contribute to games in a
  variety of ways.
• Most important single AI technique for gaming?
  Probably A.
• AI can provide opponents, partners, support
  characters, story directors, commentators.
• AI and artificial life are fun. ALife doesnt
  have to be Turing-test level of intelligence for
  good gaming.

10
From a GAME perspective

• There are two fields that are relevant here
• AI
• AI has techniques to contribute to games.
• Games are a fertile field for AI research.
• Games
• Gaming has enormous market importance.
• Games are growing increasingly sophisticated in
  all dimensions
• A game has to be fun use whatever AI techniques
  aid that.

11
Game Design, Game AI
Based on Doug Church, October, 2001
ai.eecs.umich.edu/people/laird/game-seminar/Doug-C
hurch.ppt
12
Background Game Genres

• Racing other cars
• Action enemies, allies
• Real Time Strategy (RTS) units, commanders (for
  both sides)
• Role Playing Games (RPG) monsters, party
  members, Non Player Characters
• Sports players, coaches, commentators
• Shooter targets, more targets

Based on Doug Church, October, 2001
ai.eecs.umich.edu/people/laird/game-seminar/Doug-C
hurch.ppt
13
Obvious Examples of AI

• Car Game write a virtual driver
• Shooter write a virtual player
• Sports Games write a virtual coach
• RTS write a virtual general

Based on Doug Church, October, 2001
ai.eecs.umich.edu/people/laird/game-seminar/Doug-C
hurch.ppt
14
Mini Case Study Racing Game History

• No AI cars, only real players are drivers
• AI cars which follow scripted path
• Follow path, adjust speed
• Feedback system to follow path
• Rubber band near player
• Attempt to have driver personality

Based on Doug Church, October, 2001
ai.eecs.umich.edu/people/laird/game-seminar/Doug-C
hurch.ppt
15
Somewhat Real Examples

• Car Game write AI to keep races close
• Shooter enemies die lots, win little
• Sports commentators, help player
• RTS generals who work on pacing
• It is A Question of Design Purpose

Based on Doug Church, October, 2001
ai.eecs.umich.edu/people/laird/game-seminar/Doug-C
hurch.ppt
16
Game Development

• What is the goal for Player Experience?
• How is the AI going to further that goal?
• Role of the designer
• Role of the programmer
• How design evolves
• Constraints

Based on Doug Church, October, 2001
ai.eecs.umich.edu/people/laird/game-seminar/Doug-C
hurch.ppt
17
Roles of a Game AI

• Tool for the designer
• Foil for the player, creates opportunity
• Dynamic challenge
• Events Emulation v. Simulation
• Assists in Driving the action

Based on Doug Church, October, 2001
ai.eecs.umich.edu/people/laird/game-seminar/Doug-C
hurch.ppt
18
Shipping Games vs. Test Code

• Hard constraint on CPU usage
• Reproducibility is vital, for test and design
• Must be fun, not correct
• Must succeed, finish, do something, always

Based on Doug Church, October, 2001
ai.eecs.umich.edu/people/laird/game-seminar/Doug-C
hurch.ppt
19
Mini Case Study

• The Thief AI

Based on Doug Church, October, 2001
ai.eecs.umich.edu/people/laird/game-seminar/Doug-C
hurch.ppt
20
Design Goals

• Player is going to be a Thief
• i.e. Sneak Around, Ambush, Hide, Steal
• AI must allow players to make plans
• And react to player actions, provide challenge
• Game will feature a loose overall story
• Ability to script/override behavior
• In-game actions fed back out to story control

Based on Doug Church, October, 2001
ai.eecs.umich.edu/people/laird/game-seminar/Doug-C
hurch.ppt
21
Watch-able by the player

• Has to go about its business with intent
• Actions must make sense to player
• interestingly predictable
• present play opportunities for player
• Overemphasize thoughts
• Telegraph all actions
• Goals must be very explicit

Based on Doug Church, October, 2001
ai.eecs.umich.edu/people/laird/game-seminar/Doug-C
hurch.ppt
22
Implementation

• Model senses, detection, awareness
• Simple event based reaction scripts
• Tagging of world objects which notify AI
• Patrol paths, dynamic go-to-object
• Rule based match database for speech
• Heavily designer driven, toolset approach

Based on Doug Church, October, 2001
ai.eecs.umich.edu/people/laird/game-seminar/Doug-C
hurch.ppt
23
Requirements

• Focus on the Player Experience
• Allow player to understand AI actions
• Needs to achieve design aim (and fun)
• Configurable, Override-able, Testable
• Satisfies data and speed constraints

Based on Doug Church, October, 2001
ai.eecs.umich.edu/people/laird/game-seminar/Doug-C
hurch.ppt
24
Themes

• Player Player Player Player Player Player
• How can AI enhance player experience
• AI is facilitator of the fun
• Enable creative expression for player
• Allow player to impact the world
• Put player in interesting situations

Based on Doug Church, October, 2001
ai.eecs.umich.edu/people/laird/game-seminar/Doug-C
hurch.ppt