Chapter 3 Adversarial Search

COMP 472 Artificial Intelligence

Russell & Norvig – Section 5.1 & 5.4

Some slides from: robotics.stanford.edu/~latombe/
2 Adversarial Search

´ MiniMax Search
´ Alpha – beta pruning
3 Alpha-Beta Pruning

´ Optimization over minimax, that:

´ ignores (cuts off, prunes) branches of the tree that
cannot possibly lead to a better solution
´ reduces branching factor
´ allows deeper search with same effort
4 Alpha-Beta Pruning Example 1

´ With minimax, we look at all possible nodes at the n-ply depth

´ With α-β pruning, we ignore branches that could not possibly
contribute to the final decision
B will be >= 5
So we can ignore B’s right
branch, because A must be 3
D will be <= 0
But C will be >= 3
So we can ignore D’s right
branch
E will be <= 2.
So we can ignore E’s right
branch
Because C will be 3.
5 A Closer Look

£ -1 ³ 3 A max level

min level
3 B £ -1 C D E

¬ Pruning
node C will not contribute its
-1
value to node A… so this part
of the tree can’t have any
effect on the value that will
be backed up to node A
6 Alpha-Beta Pruning Algorithm
´ α : lower bound on the final backed-up value.
´ β : upper bound on the final backed-up value.
´ Alpha pruning:
´ eg. if MAX node's α = 6, then the search can prune branches
from a MIN descendant that has a β <= 6.
´ if child β <= ancestor α à prune

a=6 b=+∞ MAX

value ³ 6
incompatible…
so stop searching the right branch;
the value cannot come from there! a=-∞ b=5 MIN
value £ 5
7 Alpha-Beta Pruning Algorithm
´ α : lower bound on the final backed-up value.
´ β : upper bound on the final backed-up value.
´ Beta pruning:
´ eg. if a MIN node's β = 6, then the search can prune branches
from a MAX descendant that has an α >= 6.
´ if ancestor β <= child α à prune

a=-∞ b=6 MIN

value £ 6
incompatible…
so stop searching the right branch;
the value cannot come from there! a=7 b=+∞ MAX
value ³ 7
8
Alpha-Beta Pruning Algorithm
01 function alphabeta(node, depth, α, β, maximizingPlayer)
02 if depth = 0 or node is a terminal node
03 return the heuristic value of node
04 if maximizingPlayer Initial call:
05 v := -∞ alphabeta(origin, depth, -∞, +∞, TRUE)
06 for each child of node
07 v := max(v, alphabeta(child, depth - 1, α, β, FALSE))
08 α := max(α, v)
09 if β ≤ α
10 break (* β cut-off *)
11 return v
12 else
13 v := ∞
14 for each child of node
15 v := min(v, alphabeta(child, depth - 1, α, β, TRUE))
16 β := min(β, v)
17 if β ≤ α
18 break (* α cut-off *)
19 return v source: http://en.wikipedia.org/wiki/Alpha%E2%80%93beta_pruning
9 Example: tic-tac-toe

max level

min level
10 Example: tic-tac-toe
max level

a=-∞ b=2
min level
value £ 2

e(n) = 2
11 Example: tic-tac-toe

max level

a=-∞ b=2 1
min level
value £ 2 1

e(n) = 2 e(n) = 1
12 Example: tic-tac-toe
a=1 b=+∞
max level
value ³ 1

min level value = 1

e(n) = 2 e(n) = 1
13 Example: tic-tac-toe
a=1 b=+∞
max level value ³ 1

a=-∞ b=-1

min level value = 1 value £ -1

e(n) = 2 e(n) = 1 e(n) = -1

14 Example: tic-tac-toe
incompatible… a=1 b=+∞
so stop searching the right branch;
max level
the value cannot come from there!
max level value ³ 1

a=-∞ b=-1

min level value = 1

child β <= ancestor α à stop search value £ -1

Search can be discontinued below

any MIN node whose β is <= α
of one of its MAX ancestors

e(n) = 2 e(n) = 1 e(n) = -1

15 Alpha – Beta Pruning Example 2

Max ≥3
≥6 =6
-------------------------------------------------------------------------------------------------------
Min ≤5 =3 ≤6ü =6 ≤5x
-------------------------------------------------------------------------------------------------------
Max ≥5 =5 =3 ≥6ü=6 =7 =5
-------------------------------------------------------------------------------------------------------
Min ≤5 =5 ≤7ü ≤4 =3 =6 ≤6x ≤6 =7 =5
≤4 x
--------------------------------------------------------------------------------------------------------
=5 =6 =7 =4 =3 =6 =6 =7 =5
16 Alpha – Beta Pruning Example 2
17 Alpha – Beta Pruning Example 3
18 Alpha – Beta Pruning Example 3
19 Alpha – Beta Pruning Example 3

<=4
a=-∞
β=4
20 Alpha – Beta Pruning Example 3

<=3
a=-∞
β=3
21 Alpha – Beta Pruning Example 3

<=3
a=-∞
β=3
22 Alpha – Beta Pruning Example 3

=3
a=-∞
β=3
23 Alpha – Beta Pruning Example 3

>=3
a=3
β=+∞
=3
a=-∞
β=3
24 Alpha – Beta Pruning Example 3

>=3
a=3
β=+∞
=3
25 Alpha – Beta Pruning Example 3

>=3
α=3

=3 <=2
β=3 β=2
26 Alpha – Beta Pruning Example 3
child β <= ancestor α à prune

>=3
a=3
β=+∞
=3 <=2
a=-∞
β=2
27 Alpha – Beta Pruning Example 3

=3

=3
28 Alpha – Beta Pruning Example 3

<=3

=3

=3
29 Alpha – Beta Pruning Example 3

<=3

=3

=3 <=4
30 Alpha – Beta Pruning Example 3

<=3

=3

=3 <=2
31 Alpha – Beta Pruning Example 3

<=3

=3

=3 =2
32 Alpha – Beta Pruning Example 3

<=3

=3 =2

=3 =2
33 Alpha – Beta Pruning Example 3

<=2

=3 =2

=3 =2
34 Alpha – Beta Pruning Example 3

<=2

=3 =2

=3 =2 <=5
35 Alpha – Beta Pruning Example 3

<=2

=3 =2

=3 =2 =4
36 Alpha – Beta Pruning Example 3

<=2

=3 =2 >=2

=3 =2 =4
37 Alpha – Beta Pruning Example 3
ancestor β <= child α à prune

<=2

=3 =2 >=4

=3 =2 =4
38 Alpha – Beta Pruning Example 3

=2

=3 =2

=3 =2 =4
39 Alpha – Beta Pruning Example 3
>= 2

=2

=3 =2

=3 =2 =4
40 Alpha – Beta Pruning Example 3
>= 2

=2

=3 =2

=3 =2 =4
41 Alpha – Beta Pruning Example 3
child β <= ancestor α à prune
>= 2
deep cut!

=2

=3 =2

=3 =2 =4 <=1
42 Alpha – Beta Pruning Example 3
child β <= ancestor α à prune
>= 2

<=1
=2

=3 =2

=3 =2 =4 <=1
43 Alpha – Beta Pruning Example 3
child β <= ancestor α à prune
>= 2

<=1
=2

=3 =2

=3 =2 =4 <=1

11 nodes explored out of 27

44 Efficiency

´ Depends on the order the siblings

´ In worst case:
´ alpha-beta provides no pruning
´ In best case:
´ branching factor is reduced to its square root
´ so can search can go twice as deep with the same
amount of computation
45 Alpha-Beta: Best ordering

Original (arbitrary) game tree

Best ordering for alpha-beta

source: G. Luger (2005)
46 Alpha-Beta: Best ordering

´ best ordering:
1. children of MIN : smallest node first
2. children of MAX: largest node first
47 Alpha-Beta: Best ordering

2 0
48 Alpha-Beta: Best ordering

2 0

2 3 7 0 3 3
49 Alpha-Beta: Best ordering

2 0

2 3 7 0 3 3

2 3 1 7 4 2 0 3 0 3 2 1

8 nodes explored out of 27

50 Applications

´ 1992 – 1994 Checkers: Tinsley VS Chinook

´ Marion Tinsley – World Champion for over
40 years
´ Chinook – Developed by Jonathan
Schaeffer, professor at University of Alberta
´ In 1992, Tinsley beat Chinook in 4 games to
2, with 33 draws.
´ In 1996: 6 draws https://webdocs.cs.ualber
ta.ca/~chinook/play/
´ In 2007, Schaeffer announced that
checkers was solved, and anyone playing
against Chinook would only be able to
draw, never win.
51 Applications

´ 1997 Othello: Murakami VS Logistello

´ Takeshi Murakami – World Othello
Champion
´ Logistello – Developed by Michael Buro
runs on a standard PC
´ Logistello beat Murakami by 6 games to 0

https://skatgame.net/mburo/log.html
(including source code)
52 Applications

´ 1997 Chess: Kasparov VS Deep Blue

´ Garry Kasparov – 50 billion neurons 2
positions/sec
´ Deep Blue – 32 RISC (Reduced Instruction-
Set Computer) processors + 256 VLSI (Very
Large-Scale Integration) chess engines
200,000,000 pos/sec
´ Deep Blue wins by 3 wins, 1 loss, and 2
draws
53 Applications

´ 1997 Chess: Kasparov VS Deep Junior

´ Garry Kasparov – still 50 billion neurons 2
positions/sec
´ Deep Junior – 8 CPU, 8 GB RAM, Win 2000
2,000,000 pos/sec Available at $100
´ Match ends in a 3/3 tie!
54 Applications

´ 2016 Go: AlphaGo VS Lee Se-dol

´ GO was always considered a much harder
game to automate than chess because of
its very high branching factor (35 for chess
vs 250 for Go!)
´ In 2016, AlphaGo beat Lee Sedol in a five-
game match of GO.
´ In 2017 AlphaGo beat Ke Jie, the world
No.1 ranked player at the time

´ uses a Monte Carlo tree search algorithm

to find its moves based on knowledge
previously "learned" by deep learning
55 The End