<s>
In	O
video	B-Application
game	I-Application
AI	I-Application
,	O
a	O
utility	B-General_Concept
system	I-General_Concept
,	O
or	O
utility	B-General_Concept
AI	I-General_Concept
,	O
is	O
a	O
simple	O
but	O
effective	O
way	O
to	O
model	B-Algorithm
behaviors	I-Algorithm
for	O
non-player	O
characters	O
.	O
</s>
<s>
One	O
of	O
the	O
benefits	O
of	O
utility	B-General_Concept
AI	I-General_Concept
is	O
that	O
it	O
is	O
less	O
"	O
hand-authored	O
"	O
than	O
many	O
other	O
types	O
of	O
game	B-Application
AI	I-Application
architectures	O
.	O
</s>
<s>
While	O
behaviors	O
in	O
a	O
utility	B-General_Concept
system	I-General_Concept
are	O
often	O
created	O
individually	O
(	O
and	O
by	O
hand	O
)	O
,	O
the	O
interactions	O
and	O
priorities	O
between	O
them	O
are	O
not	O
inherently	O
specified	O
.	O
</s>
<s>
For	O
example	O
,	O
behavior	B-Application
trees	I-Application
(	O
BTs	O
)	O
require	O
the	O
designer	O
to	O
specify	O
priorities	O
in	O
sequence	O
to	O
check	O
if	O
something	O
should	O
be	O
done	O
.	O
</s>
<s>
Only	O
if	O
that	O
behavior	O
(	O
or	O
tree	O
branch	O
)	O
is	O
NOT	O
executed	O
will	O
the	O
behavior	B-Application
tree	I-Application
system	O
fall	O
through	O
to	O
check	O
the	O
next	O
one	O
.	O
</s>
<s>
By	O
comparison	O
,	O
behaviors	O
in	O
many	O
utility	B-General_Concept
systems	I-General_Concept
sort	O
themselves	O
out	O
by	O
priority	O
based	O
on	O
the	O
scores	O
generated	O
by	O
any	O
mathematical	O
modeling	O
that	O
defines	O
every	O
given	O
behavior	O
.	O
</s>
<s>
12%	O
chance	O
to	O
perform	O
Action	O
X	O
)	O
was	O
an	O
early	O
step	O
into	O
utility	B-General_Concept
AI	I-General_Concept
.	O
</s>
<s>
Only	O
in	O
the	O
early	O
21st	O
century	O
,	O
however	O
,	O
has	O
that	O
method	O
started	O
to	O
take	O
on	O
more	O
of	O
a	O
formalized	O
approach	O
now	O
referred	O
to	O
commonly	O
as	O
"	O
utility	B-General_Concept
AI	I-General_Concept
"	O
.	O
</s>
<s>
In	O
The	B-Application
Sims	I-Application
(	O
2000	O
)	O
an	O
NPCs	O
current	O
"	O
need	O
"	O
for	O
something	O
(	O
e.g.	O
</s>
<s>
This	O
was	O
one	O
of	O
the	O
first	O
visible	O
uses	O
of	O
utility	B-General_Concept
AI	I-General_Concept
in	O
a	O
game	O
.	O
</s>
<s>
In	O
The	B-Application
Sims	I-Application
3	I-Application
(	O
2009	O
)	O
,	O
Richard	O
Evans	O
used	O
a	O
modified	O
version	O
of	O
the	O
Boltzmann	O
distribution	O
to	O
choose	O
an	O
action	O
for	O
a	O
Sim	O
,	O
using	O
a	O
temperature	O
that	O
is	O
low	O
when	O
the	O
Sim	O
is	O
happy	O
,	O
and	O
high	O
when	O
the	O
Sim	O
is	O
doing	O
badly	O
to	O
make	O
it	O
more	O
likely	O
that	O
an	O
action	O
with	O
a	O
low	O
utility	O
is	O
chosen	O
.	O
</s>
<s>
He	O
also	O
incorporated	O
"	O
personalities	O
"	O
into	O
the	B-Application
Sims	I-Application
.	O
</s>
<s>
In	O
his	O
book	O
,	O
Behavioral	O
Mathematics	O
for	O
Game	B-Application
AI	I-Application
,	O
Dave	O
Mark	O
detailed	O
how	O
to	O
mentally	O
think	O
of	O
behavior	O
in	O
terms	O
of	O
math	O
including	O
such	O
things	O
as	O
response	O
curves	O
(	O
converting	O
changing	O
input	O
variables	O
to	O
output	O
variables	O
)	O
.	O
</s>
<s>
These	O
lectures	O
served	O
to	O
inject	O
utility	B-General_Concept
AI	I-General_Concept
as	O
a	O
commonly-referred-to	O
architecture	O
alongside	O
finite	O
state	O
machines	O
(	O
FSMs	O
)	O
,	O
behavior	B-Application
trees	I-Application
,	O
and	O
planners	O
.	O
</s>
<s>
While	O
the	O
work	O
of	O
Richard	O
Evans	O
,	O
and	O
subsequent	O
AI	O
programmers	O
on	O
the	B-Application
Sims	I-Application
franchise	O
such	O
as	O
David	O
"	O
Rez	O
"	O
Graham	O
were	O
heavily	O
based	O
on	O
utility	B-General_Concept
AI	I-General_Concept
,	O
Dave	O
Mark	O
and	O
his	O
co-worker	O
from	O
ArenaNet	O
,	O
Mike	O
Lewis	O
,	O
went	O
on	O
to	O
lecture	O
at	O
the	O
AI	O
Summit	O
during	O
the	O
2015	O
GDC	O
about	O
a	O
full	O
stand-alone	O
architecture	O
he	O
had	O
developed	O
,	O
the	O
Infinite	O
Axis	O
Utility	B-General_Concept
System	I-General_Concept
(	O
IAUS	O
)	O
.	O
</s>
<s>
In	O
a	O
way	O
,	O
this	O
made	O
it	O
similar	O
to	O
behavior	B-Application
trees	I-Application
and	O
planners	O
where	O
the	O
reasoner	O
(	O
what	O
makes	O
the	O
decisions	O
)	O
was	O
fully	O
established	O
and	O
it	O
was	O
left	O
to	O
the	O
development	O
team	O
to	O
add	O
behaviors	O
into	O
the	O
mix	O
as	O
they	O
saw	O
fit	O
.	O
</s>
<s>
Bill	O
Merrill	O
wrote	O
a	O
segment	O
in	O
the	O
book	O
,	O
Game	B-Application
AI	I-Application
Pro	O
,	O
entitled	O
"	O
Building	O
Utility	O
Decisions	O
into	O
Your	O
Existing	O
Behavior	B-Application
Tree	I-Application
"	O
with	O
examples	O
of	O
how	O
to	O
re-purpose	O
selectors	O
in	O
BTs	O
to	O
use	O
utility-based	O
math	O
.	O
</s>
<s>
This	O
made	O
for	O
a	O
powerful	O
hybrid	O
that	O
kept	O
much	O
of	O
the	O
popular	O
formal	O
structure	O
of	O
behavior	B-Application
trees	I-Application
but	O
allowed	O
for	O
some	O
of	O
the	O
non-brittle	O
advantages	O
that	O
utility	O
offered	O
.	O
</s>
<s>
Utility	O
decision-making	O
is	O
relatively	O
fast	O
,	O
in	O
terms	O
of	O
real-time	O
performance	O
,	O
compared	O
to	O
more	O
computationally	O
expensive	O
planning	O
approaches	O
such	O
as	O
Monte	B-Application
Carlo	I-Application
tree	I-Application
search	I-Application
.	O
</s>
<s>
This	O
mainly	O
stems	O
from	O
the	O
fact	O
that	O
Utility	B-General_Concept
System	I-General_Concept
is	O
reactive	O
,	O
i.e.	O
,	O
chooses	O
decision	O
based	O
on	O
the	O
present	O
state	O
.	O
</s>
<s>
In	O
a	O
conference	O
paper	O
about	O
AI	O
in	O
Tactical	O
Troops	O
:	O
Anthracite	O
Shift	O
game	O
,	O
Utility	B-General_Concept
System	I-General_Concept
is	O
responsible	O
for	O
high-level	O
strategical	O
decision	O
making	O
,	O
whereas	O
Monte	B-Application
Carlo	I-Application
Tree	I-Application
Search	I-Application
is	O
responsible	O
for	O
deep	O
tactical	O
situations	O
which	O
require	O
more	O
exact	O
planning	O
.	O
</s>
