Games

n Games: multi-agent environment

Adversarial Search and Game q What do other agents do and how do they affect our
success?
Playing q Cooperative vs. competitive multi-agent environments.
q Competitive multi-agent environments give rise to
adversarial search a.k.a. games
n Why study games?
Russell and Norvig, Chapter 5
q Fun!
q They are hard
q Easy to represent and agents restricted to small
number of actions sometimes!

2
http://xkcd.com/601/

Relation of Games to Search Types of Games

n Search no adversary
q Solution is (heuristic) method for finding goal Deterministic Chance
q Heuristics and CSP techniques can find optimal solution
q Evaluation function: estimate of cost from start to goal through Perfect chess, go, backgammon
given node information checkers, othello
q Examples: path planning, scheduling activities
Imperfect Bridge, hearts Poker, canasta,
n Games adversary
information scrabble
q Solution is strategy (strategy specifies move for every possible
opponent reply). Our focus: deterministic, turn-taking, two-player,
q Time limits force approximate solutions zero-sum games of perfect information
q Examples: chess, checkers, Othello, backgammon zero-sum game: a participant's gain (or loss) is exactly balanced
by the losses (or gains) of the other participant.
perfect information: fully observable

3 4

Game setup Partial Game Tree for Tic-Tac-Toe

n Two players: MAX and MIN
n MAX moves first and they take turns until the game is over.
n Games as search:
q Initial state: e.g. starting board configuration
q Player(s): which player has the move in a state
q Action(s): set of legal moves in a state
q Result(s, a): the states resulting from a given move.
q Terminal-test(s): game over? (terminal states)
q Utility(s,p): value of terminal states, e.g., win (+1), lose (-1) and draw
(0) in chess.
n Players use search tree to determine next move.

5 6

1
 The Tic-Tac-Toe Search Space
n Is this search space a tree or graph?

n What is the minimum search depth?

n What is the maximum search depth?

n What is the branching factor?

http://xkcd.com/832/
7

Optimal strategies Two-Ply Game Tree

n Find the best strategy for MAX assuming an infallible MIN
opponent.
n Assumption: Both players play optimally.
n Given a game tree, the optimal strategy can be determined
by using the minimax value of each node:

MINIMAX(s)=
UTILITY(s) If s is a terminal
maxa Actions(s) MINIMAX(RESULT(s,a)) If PLAYER(s)=MAX
mina Actions(s) MINIMAX(RESULT(s,a)) If PLAYER(s)=MIN
Definition: ply = turn of a two-player game

9 10

Two-Ply Game Tree Two-Ply Game Tree

The minimax decision!

The minimax value at a min node is the minimum Minimax maximizes the worst-case outcome for max.
of backed-up values, because your opponent will
do whats best for them (and worst for you).
11 12

2
Minimax Algorithm Properties of Minimax
n Minimax explores tree using DFS.
function MINIMAX-DECISION(state) returns an action
return arg maxa Actions(s) MIN-VALUE(RESULT(state,a)) n Therefore:
function MAX-VALUE(state) returns a utility value q Time complexity: O(bm) L
if TERMINAL-TEST(state) then return UTILITY(state) q Space complexity: O(bm)
v J
for each a in ACTIONS(state) do
v MAX(v,MIN-VALUE(RESULT(state,a)))
return v
function MIN-VALUE(state) returns a utility value
if TERMINAL-TEST(state) then return UTILITY(state)
v
for a in ACTIONS(state) do
v MIN(v,MAX-VALUE(RESULT(state,a)))
return v
13 14

Problem of minimax search Alpha-Beta Pruning

n Number of game states is exponential in the n : the highest (i.e. best for Max) value
number of moves. possible
q Solution: Do not examine every node n : the lowest (i.e. best for Min) value possible
q ==> Alpha-beta pruning
n initially and are (-, ).
n Remove branches that do not influence final decision
n General idea: you can bracket the highest/lowest value
at a node, even before all its successors have been
evaluated

15

Alpha-Beta Example Alpha-Beta Example (continued)

Range of possible values!

[-,+]
[-,+]

[-,3]
[-, +]

17 18

3
Alpha-Beta Example (continued) Alpha-Beta Example (continued)

[3,+]
[-,+]

[3,3]
[-,3]

19 20

Alpha-Beta Example (continued) Alpha-Beta Example (continued)

[3,+]
This node is worse ! [3,14] ,
for MAX

[3,3] [-,2]
[3,3] [-,2] [-,14]

21 22

Alpha-Beta Example (continued) Alpha-Beta Example (continued)

[3,5] , [3,3]

[3,3] [-,2] [2,2]

[3,3] [-,2] [-,5]

23 24

4
Alpha-Beta Example (continued)
Alpha-Beta Algorithm

function ALPHA-BETA-SEARCH(state) returns an action

vMAX-VALUE(state, - , +)
[3,3] return the action in ACTIONS(state) with value v

[3,3] [-,2] [2,2] function MAX-VALUE(state, , ) returns a utility value

if TERMINAL-TEST(state) then return UTILITY(state)
v-
for each a in ACTIONS(state) do
v MAX(v, MIN-VALUE(RESULT(state,a), , ))
if v then return v
MAX( ,v)
return v

25 26

Alpha-Beta Algorithm Alpha-beta pruning

n When enough is known about
a node n, it can be pruned.

function MIN-VALUE(state, , ) returns a utility value

if TERMINAL-TEST(state) then return UTILITY(state)
v+
for each a in ACTIONS(state) do
v MIN(v, MAX-VALUE(RESULT(state,a), , ))
if v then return v
MIN( ,v)
return v

27 28

Final Comments about Alpha-Beta Pruning Is this practical?

n Pruning does not affect final results n Minimax and alpha-beta pruning still have exponential
n Entire subtrees can be pruned, not just leaves. complexity.
n Good move ordering improves effectiveness of pruning n May be impractical within a reasonable amount of time.
n With perfect ordering, time complexity is O(bm/2) n SHANNON (1950):
Terminate search at a lower depth
q Effective branching factor of sqrt(b)
q

q Apply heuristic evaluation function EVAL instead of the UTILITY

q Consequence: alpha-beta pruning can look twice as
function
deep as minimax in the same amount of time

29 30

5
Cutting off search Heuristic EVAL
n Change: n Idea: produce an estimate of the expected utility of the game
q if TERMINAL-TEST(state) then return UTILITY(state) from a given position.
into n Performance depends on quality of EVAL.
q if CUTOFF-TEST(state,depth) then return EVAL(state) n Requirements:
q EVAL should order terminal-nodes in the same way as UTILITY.

n Introduces a fixed-depth limit depth q Fast to Compute.

q Selected so that the amount of time will not exceed what the q For non-terminal states the EVAL should be strongly correlated
rules of the game allow. with the actual chance of winning.
n When cuttoff occurs, the evaluation is performed.

31 32

Heuristic EVAL example How good are computers

n Lets look at the state of the art computer
programs that play games such as chess,
checkers, othello, go

Addition assumes
independence

Eval(s) = w1 f1(s) + w2 f2(s) + + wn fn(s)

In chess:

w1 material + w2 mobility + w3 king safety + w4 center control +

33 34

Checkers Chinook
n Components of Chinook:
q Search (variant of alpha-beta). Search space has 1020
states.
q Evaluation function
q Endgame database (for all states with 4 vs. 4 pieces;
roughly 40 billion positions).
q Opening book - a database of opening moves
n Chinook can determine the final result of the game
n Chinook: the first program to win the world within the first 10 moves.
champion title in a competition against a n Author has recently shown that several openings
lead to a draw.
human (1994) Jonathan Schaeffer, Neil Burch, Yngvi Bjornsson, Akihiro Kishimoto, Martin Muller, Rob
Lake, Paul Lu and Steve Sutphen. "Checkers is Solved," Science, 2007.
http://www.cs.ualberta.ca/~chinook/publications/solving_checkers.html

35 36

6
Chess Othello
n 1997: Deep Blue wins a 6-
game match against Garry
Kasparov
n Searches using iterative deepening
alpha-beta; evaluation function has
over 8000 features; opening book
of 4000 positions; end game
database.
n FRITZ plays world champion,
Vladimir Kramnik; wins 6-gam n The best Othello computer programs can
match. easily defeat the best humans (e.g. Logistello,
1997).
37 38

Go Games that include chance

n Possible moves (5-10,5-11), (5-11,19-24),(5-10,10-16) and

n Go: humans still much better! (5-11,11-16)

39 40

Games that include chance Expected minimax value

chance nodes! EXPECTIMINIMAX(s)=
UTILITY(s) If s is a terminal
maxa EXPECTIMINIMAX(RESULT(s,a)) If PLAYER(S)=MAX
mina EXPECTIMINIMAX(RESULT(s,a)) If PLAYER(S)=MIN
r P(r) EXPECTIMINIMAX(RESULT(s,r)) If PLAYER(S)=CHANCE
r is a chance event (e.g., a roll of the dice).
These equations can be propagated recursively in a similar way
to the MINIMAX algorithm.

n Possible moves (5-10,5-11), (5-11,19-24),(5-10,10-16) and

(5-11,11-16)
n [1,1],,[6,6] probability 1/36, all others - 1/18
n Can not calculate definite minimax value, only expected value
41 42

7
TD-Gammon (Tesauro, 1994) Summary
n Games are fun
n They illustrate several important points about AI
Whites turn, with q Perfection is (usually) unattainable -> approximation
a roll of 4-4
q Uncertainty constrains the assignment of values to
states

World class program based on a combination of reinforcement

Learning, neural networks and alpha-beta pruning to 3 plies.

Move analyses by TD-Gammon have lead to some changes in

accepted strategies.

http://www.research.ibm.com/massive/tdl.html
43 44