General Game Playing (GGP)
Marco Adelfio
CMSC 828N – Spring 2009
Classic Game Playing AI
Deep Blue
TD-Gammon
Poki
General Game Playing AI
GGP Agent
General Game Playing
 GGP Goals:
 Create systems to play arbitrary games (given formal game
definitions)
 Eliminate game-specific strategies
 Emphasize generic strategy formulation
 Competition created by Stanford Logic Group
 Hosted during AAAI conference since 2005
 $10,000 Grand Prize
General Game Playing
 Questions:
 What additional challenges arise for GGP agents?
 How should a GGP agent evaluate game states?
 Can a GGP agent transfer knowledge between games?
General Game Playing
 Finitely many players, states
 Game play controlled by Game Manager over network
 Players act synchronously (noops allowed)
 Time limits enforced
 Basic agent must:
 Understand rule specification
 Respond to game states with legal actions
 Recognize a terminal state and its payoffs
Game Definition Language
 A game definition must logically define:
 Set of states in the game
 Legal actions for each player from a given game state
 Transition function
 Initial state
 Terminal states and their payoffs
Game Definition Language - Example
(role p1)
(<= (row ?m ?x)
(role p2)
(true (cell ?m 1 ?x))
(init (cell 1 1 b))
(true (cell ?m 2 ?x))
(init (cell 1 2 b))
(true (cell ?m 3 ?x)))
…
…
(init (control p1)
(<= (line ?x) (row ?m ?x))
…
(<= (line ?x) (column ?m ?x))
(<= (legal ?w (mark ?x ?y))
(<= (line ?x) (diagonal ?x))
(true (cell ?x ?y b))
…
(true (control ?w)))
(<= (goal p1 100)
…
(line x))
(<= (next (cell ?m ?n x))
(<= (goal p1 0)
(does p1 (mark ?m ?n))
(true (cell ?m ?n b)))
…
(line o)
…
(<= terminal
(line x))
Game Communication
Game Manager Message
Game Player Response
(START MATCH.435 WHITE description 90 30)
READY
(PLAY MATCH.435 (NIL NIL))
(MARK 2 2)
(PLAY MATCH.435 ((MARK 2 2) NOOP)))
NOOP
(PLAY MATCH.435 (NOOP (MARK 1 3))
(MARK 1 2)
(PLAY MATCH.435 ((MARK 1 2) NOOP))
NOOP
...
...
(STOP MATCH.435 ((MARK 3 3) NOOP)
DONE
General Game Playing
 Design Challenges:
 Indeterminacy
 Size
 Multi-game Commonalities
 Opponent Recognition
AAAI Competition – Past Winners
 2005 - ClunePlayer (UCLA)
 2006 - FluxPlayer (Technical University of Dresden)
 2007 - CADIA (Reykjavik University)
 2008 - CADIA (Reykjavik University)
Agent 1: ClunePlayer
 Approach: Minimax
 Problem: Needs to assign values to intermediate game states
in arbitrary games.
 Solution:
Calculate a vector of generic features at each node
2. Simulate games to determine which features are “stable” and
correlated with either payoff or control
3. When running minimax, use a combination of those scores as
the evaluation heuristic
1.
Agent 2: CADIA-Player
 Approach: UCT (Variant of Monte Carlo simulation)
 Monte Carlo:
 Pick random actions for each player to descend the tree
 After reaching a terminal state, update expected payoff Q(s,a)
for each visited state s and action a
 Introduces explore/exploit tradeoff
Agent 2: CADIA-Player
 UCT (Upper Confidence bound
for Trees)
 Balance exploration and
exploitation
 Give “bonus” to less travelled
paths
Agent 3: UTexas LARG
 Approach: Knowledge Transfer
 Uses lessons from past games to improve play in new games
 War Games!
 Determines whether a new game is isomorphic or similar to
a previous game. If so, transfer estimated rewards
Summary
 General Game Playing introduces a different set of challenges
than designing game-specific AI
 Biggest challenge is evaluating states in a novel game
 Better understanding of general strategy formation has many
applications
References
 GGP Website: http://games.stanford.edu/
 Hilmar Finnson. CADIA-Player: A General Game Playing
Agent. MSc Thesis, School of Computer Science, Reykjavik
University. 2007.
 Kuhlmann, Gregory and Peter Stone. Graph-Based Domain
Mapping for Transfer Learning in General Games. Lecture
Notes in Computer Science, Volume 4701/2007.