<s>
In	O
artificial	B-Application
intelligence	I-Application
,	O
an	O
intelligent	B-General_Concept
agent	I-General_Concept
(	O
IA	O
)	O
is	O
an	O
agent	O
acting	O
in	O
an	O
intelligent	O
manner	O
;	O
It	O
perceives	B-General_Concept
its	O
environment	O
,	O
takes	O
actions	O
autonomously	O
in	O
order	O
to	O
achieve	O
goals	O
,	O
and	O
may	O
improve	O
its	O
performance	O
with	O
learning	O
or	O
acquiring	O
knowledge	O
.	O
</s>
<s>
An	O
intelligent	B-General_Concept
agent	I-General_Concept
may	O
be	O
simple	O
or	O
complex	O
:	O
A	O
thermostat	O
or	O
other	O
control	O
system	O
is	O
considered	O
an	O
example	O
of	O
an	O
intelligent	B-General_Concept
agent	I-General_Concept
,	O
as	O
is	O
a	O
human	O
being	O
,	O
as	O
is	O
any	O
system	O
that	O
meets	O
the	O
definition	O
,	O
such	O
as	O
a	O
firm	O
,	O
a	O
state	O
,	O
or	O
a	O
biome	O
.	O
</s>
<s>
Leading	O
AI	B-Application
textbooks	O
define	O
"	O
artificial	B-Application
intelligence	I-Application
"	O
as	O
the	O
"	O
study	O
and	O
design	O
of	O
intelligent	B-General_Concept
agents	I-General_Concept
"	O
,	O
a	O
definition	O
that	O
considers	O
goal-directed	O
behavior	O
to	O
be	O
the	O
essence	O
of	O
intelligence	O
.	O
</s>
<s>
For	O
example	O
,	O
a	O
reinforcement	O
learning	O
agent	O
has	O
a	O
"	O
reward	O
function	O
"	O
that	O
allows	O
the	O
programmers	O
to	O
shape	O
the	O
IA	O
's	O
desired	O
behavior	O
,	O
and	O
an	O
evolutionary	B-Algorithm
algorithm	I-Algorithm
's	O
behavior	O
is	O
shaped	O
by	O
a	O
"	O
fitness	O
function	O
"	O
.	O
</s>
<s>
Intelligent	B-General_Concept
agents	I-General_Concept
in	O
artificial	B-Application
intelligence	I-Application
are	O
closely	O
related	O
to	O
agents	O
in	O
economics	O
,	O
and	O
versions	O
of	O
the	O
intelligent	B-General_Concept
agent	I-General_Concept
paradigm	O
are	O
studied	O
in	O
cognitive	O
science	O
,	O
ethics	O
,	O
the	O
philosophy	O
of	O
practical	O
reason	O
,	O
as	O
well	O
as	O
in	O
many	O
interdisciplinary	O
socio-cognitive	O
modeling	O
and	O
computer	O
social	O
simulations	O
.	O
</s>
<s>
Intelligent	B-General_Concept
agents	I-General_Concept
are	O
often	O
described	O
schematically	O
as	O
an	O
abstract	O
functional	O
system	O
similar	O
to	O
a	O
computer	O
program	O
.	O
</s>
<s>
Abstract	O
descriptions	O
of	O
intelligent	B-General_Concept
agents	I-General_Concept
are	O
called	O
abstract	B-General_Concept
intelligent	I-General_Concept
agents	I-General_Concept
(	O
AIA	O
)	O
to	O
distinguish	O
them	O
from	O
their	O
real-world	O
implementations	O
.	O
</s>
<s>
An	O
autonomous	B-General_Concept
intelligent	I-General_Concept
agent	I-General_Concept
is	O
designed	O
to	O
function	O
in	O
the	O
absence	O
of	O
human	O
intervention	O
.	O
</s>
<s>
Intelligent	B-General_Concept
agents	I-General_Concept
are	O
also	O
closely	O
related	O
to	O
software	B-General_Concept
agents	I-General_Concept
(	O
an	O
autonomous	O
computer	O
program	O
that	O
carries	O
out	O
tasks	O
on	O
behalf	O
of	O
users	O
)	O
.	O
</s>
<s>
and	O
defines	O
the	O
field	O
of	O
"	O
artificial	B-Application
intelligence	I-Application
"	O
research	O
as	O
:	O
</s>
<s>
Padgham	O
&	O
Winikoff	O
(	O
2005	O
)	O
agree	O
that	O
an	O
intelligent	B-General_Concept
agent	I-General_Concept
is	O
situated	O
in	O
an	O
environment	O
and	O
responds	O
in	O
a	O
timely	O
(	O
though	O
not	O
necessarily	O
real-time	O
)	O
manner	O
to	O
environment	O
changes	O
.	O
</s>
<s>
However	O
,	O
intelligent	B-General_Concept
agents	I-General_Concept
must	O
also	O
proactively	O
pursue	O
goals	O
in	O
a	O
flexible	O
and	O
robust	O
way	O
.	O
</s>
<s>
Kaplan	O
and	O
Haenlein	O
define	O
artificial	B-Application
intelligence	I-Application
as	O
"	O
A	O
system	O
's	O
ability	O
to	O
correctly	O
interpret	O
external	O
data	B-General_Concept
,	O
to	O
learn	O
from	O
such	O
data	B-General_Concept
,	O
and	O
to	O
use	O
those	O
learnings	O
to	O
achieve	O
specific	O
goals	O
and	O
tasks	O
through	O
flexible	O
adaptation.	O
"	O
</s>
<s>
This	O
definition	O
is	O
closely	O
related	O
to	O
that	O
of	O
an	O
intelligent	B-General_Concept
agent	I-General_Concept
.	O
</s>
<s>
Philosophically	O
,	O
this	O
definition	O
of	O
artificial	B-Application
intelligence	I-Application
avoids	O
several	O
lines	O
of	O
criticism	O
.	O
</s>
<s>
Thus	O
,	O
there	O
is	O
no	O
need	O
to	O
discuss	O
if	O
it	O
is	O
"	O
real	O
"	O
vs	O
"	O
simulated	O
"	O
intelligence	O
(	O
i.e.	O
,	O
"	O
synthetic	O
"	O
vs	O
"	O
artificial	O
"	O
intelligence	O
)	O
and	O
does	O
not	O
indicate	O
that	O
such	O
a	O
machine	O
has	O
a	O
mind	O
,	O
consciousness	O
or	O
true	O
understanding	O
(	O
i.e.	O
,	O
it	O
does	O
not	O
imply	O
John	O
Searle	O
's	O
"	O
strong	O
AI	B-Application
hypothesis	O
"	O
)	O
.	O
</s>
<s>
More	O
importantly	O
,	O
it	O
has	O
a	O
number	O
of	O
practical	O
advantages	O
that	O
have	O
helped	O
move	O
AI	B-Application
research	I-Application
forward	O
.	O
</s>
<s>
If	O
the	O
AI	B-Application
is	O
programmed	O
for	O
"	O
reinforcement	O
learning	O
"	O
,	O
it	O
has	O
a	O
"	O
reward	O
function	O
"	O
that	O
encourages	O
some	O
types	O
of	O
behavior	O
and	O
punishes	O
others	O
.	O
</s>
<s>
Alternatively	O
,	O
an	O
evolutionary	O
system	O
can	O
induce	O
goals	O
by	O
using	O
a	O
"	O
fitness	O
function	O
"	O
to	O
mutate	O
and	O
preferentially	O
replicate	O
high-scoring	O
AI	B-Application
systems	O
,	O
similar	O
to	O
how	O
animals	O
evolved	O
to	O
innately	O
desire	O
certain	O
goals	O
such	O
as	O
finding	O
food	O
.	O
</s>
<s>
Some	O
AI	B-Application
systems	O
,	O
such	O
as	O
nearest-neighbor	O
,	O
instead	O
of	O
reason	O
by	O
analogy	O
,	O
these	O
systems	O
are	O
not	O
generally	O
given	O
goals	O
,	O
except	O
to	O
the	O
degree	O
that	O
goals	O
are	O
implicit	O
in	O
their	O
training	O
data	B-General_Concept
.	O
</s>
<s>
In	O
the	O
generative	B-Algorithm
adversarial	I-Algorithm
networks	I-Algorithm
of	O
the	O
2010s	O
,	O
an	O
"	O
encoder	O
"	O
/	O
"	O
generator	O
"	O
component	O
attempts	O
to	O
mimic	O
and	O
improvise	O
human	O
text	O
composition	O
.	O
</s>
<s>
While	O
GOFAI	B-General_Concept
systems	O
often	O
accept	O
an	O
explicit	O
goal	O
function	O
,	O
the	O
paradigm	O
can	O
also	O
be	O
applied	O
to	O
neural	B-Architecture
networks	I-Architecture
and	O
to	O
evolutionary	O
computing	O
.	O
</s>
<s>
Reinforcement	O
learning	O
can	O
generate	O
intelligent	B-General_Concept
agents	I-General_Concept
that	O
appear	O
to	O
act	O
in	O
ways	O
intended	O
to	O
maximize	O
a	O
"	O
reward	O
function	O
"	O
.	O
</s>
<s>
AlphaZero	B-Application
chess	O
had	O
a	O
simple	O
objective	O
function	O
;	O
each	O
win	O
counted	O
as	O
+1	O
point	O
,	O
and	O
each	O
loss	O
counted	O
as	O
-1	O
point	O
.	O
</s>
<s>
Evolutionary	O
computing	O
can	O
evolve	O
intelligent	B-General_Concept
agents	I-General_Concept
that	O
appear	O
to	O
act	O
in	O
ways	O
intended	O
to	O
maximize	O
a	O
"	O
fitness	O
function	O
"	O
that	O
influences	O
how	O
many	O
descendants	O
each	O
agent	O
is	O
allowed	O
to	O
leave	O
.	O
</s>
<s>
The	O
theoretical	O
and	O
uncomputable	O
AIXI	O
design	O
is	O
a	O
maximally	O
intelligent	B-General_Concept
agent	I-General_Concept
in	O
this	O
paradigm	O
;	O
however	O
,	O
in	O
the	O
real	O
world	O
,	O
the	O
IA	O
is	O
constrained	O
by	O
finite	O
time	O
and	O
hardware	O
resources	O
,	O
and	O
scientists	O
compete	O
to	O
produce	O
algorithms	O
that	O
can	O
achieve	O
progressively	O
higher	O
scores	O
on	O
benchmark	O
tests	O
with	O
real-world	O
hardware	O
.	O
</s>
<s>
Simple	O
reflex	O
agents	O
act	O
only	O
on	O
the	O
basis	O
of	O
the	O
current	O
percept	B-General_Concept
,	O
ignoring	O
the	O
rest	O
of	O
the	O
percept	B-General_Concept
history	O
.	O
</s>
<s>
A	O
model-based	O
reflex	O
agent	O
should	O
maintain	O
some	O
sort	O
of	O
internal	O
model	O
that	O
depends	O
on	O
the	O
percept	B-General_Concept
history	O
and	O
thereby	O
reflects	O
at	O
least	O
some	O
of	O
the	O
unobserved	O
aspects	O
of	O
the	O
current	O
state	O
.	O
</s>
<s>
Percept	B-General_Concept
history	O
and	O
impact	O
of	O
action	O
on	O
the	O
environment	O
can	O
be	O
determined	O
by	O
using	O
the	O
internal	O
model	O
.	O
</s>
<s>
Search	O
and	O
planning	B-Application
are	O
the	O
subfields	O
of	O
artificial	B-Application
intelligence	I-Application
devoted	O
to	O
finding	O
action	O
sequences	O
that	O
achieve	O
the	O
agent	O
's	O
goals	O
.	O
</s>
<s>
The	O
performance	O
element	O
is	O
what	O
we	O
have	O
previously	O
considered	O
to	O
be	O
the	O
entire	O
agent	O
:	O
it	O
takes	O
in	O
percepts	B-General_Concept
and	O
decides	O
on	O
actions	O
.	O
</s>
<s>
Belief-desire-intention	O
agents	O
–	O
in	O
which	O
decision	O
making	O
depends	O
upon	O
the	O
manipulation	O
of	O
data	B-General_Concept
structures	O
representing	O
the	O
beliefs	O
,	O
desires	O
,	O
and	O
intentions	O
of	O
the	O
agent	O
;	O
and	O
finally	O
,	O
</s>
<s>
In	O
2013	O
,	O
Alexander	O
Wissner-Gross	O
published	O
a	O
theory	O
pertaining	O
to	O
Freedom	O
and	O
Intelligence	O
for	O
intelligent	B-General_Concept
agents	I-General_Concept
.	O
</s>
<s>
To	O
actively	O
perform	O
their	O
functions	O
,	O
Intelligent	B-General_Concept
Agents	I-General_Concept
today	O
are	O
normally	O
gathered	O
in	O
a	O
hierarchical	O
structure	O
containing	O
many	O
“	O
sub-agents	O
”	O
.	O
</s>
<s>
Taken	O
together	O
,	O
the	O
intelligent	B-General_Concept
agent	I-General_Concept
and	O
sub-agents	O
create	O
a	O
complete	O
system	O
that	O
can	O
accomplish	O
difficult	O
tasks	O
or	O
goals	O
with	O
behaviors	O
and	O
responses	O
that	O
display	O
a	O
form	O
of	O
intelligence	O
.	O
</s>
<s>
A	O
simple	O
agent	O
program	O
can	O
be	O
defined	O
mathematically	O
as	O
a	O
function	O
f	O
(	O
called	O
the	O
"	O
agent	O
function	O
"	O
)	O
which	O
maps	O
every	O
possible	O
percepts	B-General_Concept
sequence	O
to	O
a	O
possible	O
action	O
the	O
agent	O
can	O
perform	O
or	O
to	O
a	O
coefficient	O
,	O
feedback	O
element	O
,	O
function	O
or	O
constant	O
that	O
affects	O
eventual	O
actions	O
:	O
</s>
<s>
The	O
program	O
agent	O
,	O
instead	O
,	O
maps	O
every	O
possible	O
percept	B-General_Concept
to	O
an	O
action	O
.	O
</s>
<s>
We	O
use	O
the	O
term	O
percept	B-General_Concept
to	O
refer	O
to	O
the	O
agent	O
's	O
perceptional	O
inputs	O
at	O
any	O
given	O
instant	O
.	O
</s>
<s>
Intelligent	B-General_Concept
agents	I-General_Concept
are	O
applied	O
as	O
automated	B-Protocol
online	I-Protocol
assistants	I-Protocol
,	O
where	O
they	O
function	O
to	O
perceive	O
the	O
needs	O
of	O
customers	O
in	O
order	O
to	O
perform	O
individualized	O
customer	O
service	O
.	O
</s>
<s>
Such	O
an	O
agent	O
may	O
basically	O
consist	O
of	O
a	O
dialog	B-General_Concept
system	I-General_Concept
,	O
an	O
avatar	B-Application
,	O
as	O
well	O
an	O
expert	B-Application
system	I-Application
to	O
provide	O
specific	O
expertise	O
to	O
the	O
user	O
.	O
</s>
<s>
They	O
can	O
also	O
be	O
used	O
to	O
optimize	O
coordination	O
of	O
human	O
groups	O
online	B-Algorithm
.	O
</s>
<s>
People	O
's	O
behavior	O
is	O
imitated	O
by	O
artificial	B-General_Concept
agents	I-General_Concept
based	O
on	O
data	B-General_Concept
of	O
real	O
human	O
behavior	O
.	O
</s>
<s>
"	O
Intelligent	B-General_Concept
agent	I-General_Concept
"	O
is	O
also	O
often	O
used	O
as	O
a	O
vague	O
marketing	O
term	O
,	O
sometimes	O
synonymous	O
with	O
"	O
virtual	B-Protocol
personal	I-Protocol
assistant	I-Protocol
"	O
.	O
</s>
<s>
These	O
examples	O
are	O
known	O
as	O
software	B-General_Concept
agents	I-General_Concept
,	O
and	O
sometimes	O
an	O
"	O
intelligent	B-Protocol
software	I-Protocol
agent	I-Protocol
"	O
(	O
that	O
is	O
,	O
a	O
software	B-General_Concept
agent	I-General_Concept
with	O
intelligence	O
)	O
is	O
referred	O
to	O
as	O
an	O
"	O
intelligent	B-General_Concept
agent	I-General_Concept
"	O
.	O
</s>
