<s>
Computer	B-General_Concept
Arimaa	I-General_Concept
refers	O
to	O
the	O
playing	O
of	O
the	O
board	O
game	O
Arimaa	O
by	O
computer	B-Application
programs	I-Application
.	O
</s>
<s>
In	O
2002	O
,	O
Indian-American	O
computer	O
engineer	O
Omar	O
Syed	O
published	O
the	O
rules	O
to	O
Arimaa	O
and	O
announced	O
a	O
$	O
10,000	O
prize	O
,	O
available	O
annually	O
until	O
2020	O
,	O
for	O
the	O
first	O
computer	B-Application
program	I-Application
(	O
running	O
on	O
standard	O
,	O
off-the-shelf	O
hardware	O
)	O
able	O
to	O
defeat	O
each	O
of	O
three	O
top-ranked	O
human	O
players	O
in	O
a	O
three-game	O
series	O
.	O
</s>
<s>
The	O
prize	O
was	O
claimed	O
in	O
2015	O
,	O
when	O
a	O
computer	B-Application
program	I-Application
played	O
7:2	O
against	O
three	O
human	O
players	O
.	O
</s>
<s>
The	O
following	O
techniques	O
are	O
used	O
by	O
some	O
or	O
all	O
of	O
the	B-Application
artificial	I-Application
intelligence	I-Application
programs	O
that	O
play	O
Arimaa	O
:	O
</s>
<s>
Several	O
aspects	O
of	O
Arimaa	O
make	O
it	O
difficult	O
for	O
computer	B-Application
programs	I-Application
to	O
beat	O
good	O
human	O
players	O
.	O
</s>
<s>
Because	O
so	O
much	O
effort	O
has	O
gone	O
into	O
the	O
development	O
of	O
strong	O
chess-playing	B-Application
software	I-Application
,	O
it	O
is	O
particularly	O
relevant	O
to	O
understand	O
why	O
techniques	O
applicable	O
to	O
chess	B-Application
are	O
less	O
effective	O
for	O
Arimaa	O
.	O
</s>
<s>
The	O
simplest	O
chess	B-Application
programs	I-Application
use	O
brute-force	O
searching	O
coupled	O
with	O
static	O
position	O
evaluation	O
dominated	O
by	O
material	O
considerations	O
.	O
</s>
<s>
This	O
is	O
known	O
as	O
the	O
branching	B-Data_Structure
factor	I-Data_Structure
.	O
</s>
<s>
The	O
average	O
branching	B-Data_Structure
factor	I-Data_Structure
in	O
a	O
game	O
of	O
Chess	B-Application
is	O
about	O
35	O
,	O
whereas	O
in	O
Arimaa	O
it	O
is	O
about	O
17,000	O
.	O
</s>
<s>
These	O
differing	O
branching	B-Data_Structure
factors	I-Data_Structure
imply	O
that	O
a	O
computer	O
which	O
can	O
search	O
to	O
a	O
depth	O
of	O
eight	O
turns	O
for	O
each	O
player	O
in	O
chess	B-Application
,	O
can	O
only	O
search	O
about	O
three	O
turns	O
deep	O
for	O
each	O
player	O
in	O
Arimaa	O
:	O
</s>
<s>
Brute	B-Algorithm
force	I-Algorithm
search	I-Algorithm
depth	O
,	O
for	O
chess	B-Application
software	I-Application
,	O
is	O
nearly	O
doubled	O
by	O
alpha-beta	B-Algorithm
pruning	I-Algorithm
,	O
which	O
allows	O
the	O
software	O
to	O
conclude	O
that	O
one	O
move	O
is	O
better	O
than	O
another	O
without	O
examining	O
every	O
possible	O
continuation	O
of	O
the	O
weaker	O
move	O
.	O
</s>
<s>
Furthermore	O
,	O
the	O
usefulness	O
of	O
alpha-beta	B-Algorithm
pruning	I-Algorithm
is	O
heavily	O
dependent	O
on	O
the	O
order	O
in	O
which	O
moves	O
are	O
considered	O
.	O
</s>
<s>
In	O
Arimaa	O
software	O
the	O
speedup	O
provided	O
by	O
alpha-beta	B-Algorithm
pruning	I-Algorithm
is	O
less	O
,	O
because	O
captures	O
are	O
rarer	O
.	O
</s>
<s>
In	O
rated	O
games	O
played	O
on	O
arimaa.com,	O
only	O
3%	O
of	O
steps	O
result	O
in	O
capture	O
,	O
compared	O
to	O
about	O
19%	O
of	O
chess	B-Application
moves	O
that	O
result	O
in	O
capture	O
.	O
</s>
<s>
Compared	O
to	O
chess	B-Application
,	O
Arimaa	O
allows	O
either	O
player	O
to	O
delay	O
captures	O
for	O
longer	O
.	O
</s>
<s>
Indeed	O
,	O
the	O
median	O
move	O
number	O
of	O
the	O
first	O
capture	O
in	O
chess	B-Application
is	O
turn	O
6	O
,	O
whereas	O
in	O
Arimaa	O
it	O
is	O
turn	O
12	O
.	O
</s>
<s>
Thus	O
the	O
strength	O
of	O
computer	B-Application
programs	I-Application
(	O
examining	O
millions	O
of	O
positions	O
)	O
is	O
not	O
as	O
significant	O
as	O
their	O
weakness	O
(	O
judging	O
the	O
position	O
apart	O
from	O
who	O
has	O
more	O
pieces	O
)	O
.	O
</s>
<s>
In	O
chess	B-Application
every	O
game	O
starts	O
from	O
the	O
same	O
position	O
.	O
</s>
<s>
By	O
compiling	O
before	O
the	O
game	O
a	O
list	O
of	O
stock	O
replies	O
to	O
all	O
standard	O
opening	O
moves	O
,	O
chess	B-Application
programs	I-Application
may	O
often	O
make	O
a	O
dozen	O
or	O
more	O
excellent	O
moves	O
before	O
starting	O
to	O
"	O
think	O
"	O
.	O
</s>
<s>
Humans	O
do	O
the	O
same	O
,	O
but	O
have	O
a	O
smaller	O
and	O
less	O
reliable	O
memory	O
of	O
openings	O
,	O
which	O
puts	O
humans	O
at	O
a	O
relative	O
disadvantage	O
in	O
chess	B-Application
.	O
</s>
<s>
Against	O
a	O
conservative	O
opponent	O
it	O
is	O
almost	O
impossible	O
to	O
bust	O
open	O
the	O
position	O
in	O
Arimaa	O
,	O
whereas	O
in	O
chess	B-Application
it	O
is	O
merely	O
difficult	O
.	O
</s>
<s>
One	O
additional	O
technique	O
from	O
computer	B-Application
chess	I-Application
which	O
does	O
not	O
apply	O
to	O
Arimaa	O
is	O
endgame	O
tablebases	O
.	O
</s>
<s>
Master-level	O
chess	B-Application
games	O
sometimes	O
trade	O
down	O
into	O
unclear	O
endgames	O
with	O
only	O
a	O
few	O
pieces	O
,	O
for	O
example	O
king	O
and	O
knight	O
vs.	O
king	O
and	O
rook	O
.	O
</s>
<s>
Equal	O
exchanges	O
of	O
pieces	O
are	O
less	O
common	O
than	O
in	O
chess	B-Application
,	O
so	O
it	O
is	O
rare	O
for	O
a	O
game	O
of	O
Arimaa	O
to	O
"	O
trade	O
down	O
"	O
and	O
still	O
be	O
unclear	O
.	O
</s>
<s>
An	O
average	O
game	O
of	O
Arimaa	O
has	O
only	O
eight	O
captures	O
(	O
compared	O
to	O
seventeen	O
for	O
chess	B-Application
)	O
,	O
and	O
top	O
humans	O
can	O
often	O
defeat	O
top	O
programs	O
in	O
Arimaa	O
without	O
losing	O
a	O
single	O
piece	O
,	O
for	O
example	O
.	O
</s>
<s>
Omar	O
Syed	O
hopes	O
that	O
,	O
because	O
traditional	O
artificial	B-Application
intelligence	I-Application
techniques	O
are	O
only	O
moderately	O
effective	O
for	O
Arimaa	O
,	O
programmers	O
will	O
be	O
forced	O
to	O
use	O
new	O
artificial	B-Application
intelligence	I-Application
techniques	O
to	O
create	O
a	O
strong	O
Arimaa-playing	O
program	O
.	O
</s>
<s>
The	O
successful	O
quest	O
to	O
build	O
a	O
world-championship-caliber	O
chess	B-Application
program	I-Application
has	O
produced	O
many	O
techniques	O
to	O
successfully	O
play	O
games	O
,	O
but	O
has	O
contributed	O
essentially	O
nothing	O
to	O
more	O
general	O
reasoning	O
;	O
in	O
fact	O
,	O
the	O
techniques	O
of	O
chess	B-Application
playing	O
programs	O
have	O
been	O
excluded	O
from	O
some	O
definitions	O
of	O
artificial	B-Application
intelligence	I-Application
;	O
a	O
goal	O
for	O
Arimaa	O
is	O
that	O
the	O
techniques	O
involved	O
in	O
playing	O
it	O
will	O
help	O
the	O
larger	O
goals	O
of	O
artificial	B-Application
intelligence	I-Application
.	I-Application
</s>
<s>
The	O
structure	O
of	O
Syed	O
's	O
man-against-machine	O
challenge	O
is	O
focused	O
on	O
rewarding	O
advances	O
in	O
AI	B-Application
software	I-Application
and	O
not	O
advances	O
in	O
hardware	O
.	O
</s>
<s>
The	O
challenge	O
would	O
not	O
be	O
open	O
to	O
anyone	O
requiring	O
expensive	O
multi-processor	O
machines	O
such	O
as	O
those	O
used	O
to	O
challenge	O
top-level	O
chess	B-Application
players	O
,	O
much	O
less	O
something	O
like	O
the	O
custom-built	O
supercomputer	O
Deep	B-General_Concept
Blue	I-General_Concept
,	O
even	O
though	O
it	O
was	O
the	O
success	O
of	O
this	O
hardware-intensive	O
approach	O
which	O
inspired	O
Arimaa	O
's	O
invention	O
.	O
</s>
<s>
Supercomputers	O
might	O
already	O
have	O
the	O
power	O
to	O
conquer	O
Arimaa	O
by	O
brute	O
force	O
using	O
conventional	O
AI	B-Application
software	I-Application
,	O
and	O
eventually	O
personal	O
computers	O
will	O
too	O
,	O
if	O
hardware	O
continues	O
to	O
advance	O
at	O
the	O
current	O
rate	O
.	O
</s>
