<s>
An	O
artificial	O
general	O
intelligence	O
(	O
AGI	B-General_Concept
)	O
is	O
a	O
hypothetical	O
intelligent	B-General_Concept
agent	I-General_Concept
which	O
can	O
learn	O
to	O
replicate	O
any	O
intellectual	O
task	O
that	O
human	O
beings	O
or	O
other	O
animals	O
can	O
.	O
</s>
<s>
AGI	B-General_Concept
has	O
also	O
been	O
defined	O
alternatively	O
as	O
an	O
autonomous	O
system	O
that	O
surpasses	O
human	O
capabilities	O
at	O
the	O
majority	O
of	O
economically	O
valuable	O
work	O
.	O
</s>
<s>
Developing	O
AGI	B-General_Concept
is	O
a	O
primary	O
goal	O
of	O
some	O
artificial	B-Application
intelligence	I-Application
research	I-Application
and	O
for	O
several	O
AI	B-Application
companies	O
such	O
as	O
OpenAI	O
,	O
DeepMind	B-Application
,	O
and	O
Anthropic	O
,	O
as	O
well	O
as	O
a	O
common	O
topic	O
in	O
science	O
fiction	O
and	O
futures	O
studies	O
.	O
</s>
<s>
The	O
timeline	O
for	O
AGI	B-General_Concept
development	O
remains	O
a	O
subject	O
of	O
ongoing	O
debate	O
among	O
researchers	O
and	O
experts	O
,	O
with	O
some	O
arguing	O
it	O
may	O
be	O
possible	O
in	O
the	O
coming	O
years	O
or	O
decades	O
,	O
others	O
maintaining	O
it	O
might	O
take	O
up	O
to	O
a	O
century	O
or	O
longer	O
,	O
and	O
a	O
minority	O
believing	O
it	O
may	O
never	O
be	O
achieved	O
at	O
all	O
.	O
</s>
<s>
Additionally	O
,	O
there	O
is	O
debate	O
regarding	O
whether	O
modern	O
deep	O
learning	O
systems	O
,	O
such	O
as	O
GPT-4	O
,	O
are	O
an	O
early	O
yet	O
incomplete	O
form	O
of	O
AGI	B-General_Concept
or	O
if	O
new	O
approaches	O
are	O
required	O
.	O
</s>
<s>
Contention	O
exists	O
over	O
the	O
potential	O
for	O
AGI	B-General_Concept
to	O
pose	O
a	O
threat	O
to	O
humanity	O
;	O
for	O
example	O
,	O
OpenAI	O
treats	O
it	O
as	O
an	O
existential	O
risk	O
,	O
while	O
others	O
find	O
the	O
development	O
of	O
AGI	B-General_Concept
to	O
be	O
too	O
remote	O
to	O
present	O
a	O
risk	O
.	O
</s>
<s>
A	O
2020	O
survey	O
identified	O
72	O
active	O
AGI	B-General_Concept
R&D	O
projects	O
spread	O
across	O
37	O
countries	O
.	O
</s>
<s>
AGI	B-General_Concept
is	O
also	O
called	O
strong	O
AI	B-Application
,	O
full	O
AI	B-Application
,	O
or	O
general	O
intelligent	O
action	O
,	O
although	O
some	O
academic	O
sources	O
reserve	O
the	O
term	O
"	O
strong	O
AI	B-Application
"	O
for	O
computer	O
programs	O
that	O
experience	O
sentience	O
or	O
consciousness	O
.	O
</s>
<s>
Strong	O
AI	B-Application
contrasts	O
with	O
weak	O
AI	B-Application
(	O
or	O
narrow	O
AI	B-Application
)	O
,	O
which	O
is	O
not	O
intended	O
to	O
have	O
general	O
cognitive	O
abilities	O
but	O
is	O
designed	O
to	O
solve	O
exactly	O
one	O
problem	O
.	O
</s>
<s>
(	O
Some	O
academic	O
sources	O
use	O
"	O
weak	O
AI	B-Application
"	O
to	O
refer	O
more	O
broadly	O
to	O
any	O
programs	O
that	O
do	O
not	O
experience	O
consciousness	O
or	O
do	O
not	O
have	O
a	O
mind	O
in	O
the	O
same	O
sense	B-General_Concept
people	O
do	O
.	O
)	O
</s>
<s>
Related	O
concepts	O
include	O
human-level	O
AI	B-Application
,	O
transformative	O
AI	B-Application
,	O
and	O
superintelligence	O
.	O
</s>
<s>
However	O
,	O
there	O
is	O
wide	O
agreement	O
among	O
artificial	B-Application
intelligence	I-Application
researchers	O
that	O
intelligence	O
is	O
required	O
to	O
do	O
the	O
following	O
:	O
</s>
<s>
represent	O
knowledge	O
,	O
including	O
common	O
sense	B-General_Concept
knowledge	O
;	O
</s>
<s>
plan	B-Application
;	O
</s>
<s>
communicate	O
in	O
natural	B-Language
language	I-Language
;	O
</s>
<s>
and	O
,	O
if	O
necessary	O
,	O
integrate	B-Architecture
all	I-Architecture
these	I-Architecture
skills	I-Architecture
in	O
completion	O
of	O
any	O
given	O
goal	O
.	O
</s>
<s>
input	O
as	O
the	O
ability	O
to	O
sense	B-General_Concept
(	O
e.g.	O
</s>
<s>
see	B-Application
,	O
hear	O
,	O
etc	O
.	O
</s>
<s>
This	O
includes	O
the	O
ability	O
to	O
detect	O
and	O
respond	O
to	O
hazard	B-General_Concept
.	O
</s>
<s>
cognitive	O
science	O
,	O
computational	B-General_Concept
intelligence	I-General_Concept
,	O
and	O
decision	O
making	O
)	O
consider	O
additional	O
traits	O
such	O
as	O
imagination	O
(	O
taken	O
as	O
the	O
ability	O
to	O
form	O
mental	O
images	O
and	O
concepts	O
that	O
were	O
not	O
programmed	O
in	O
)	O
and	O
autonomy	O
.	O
</s>
<s>
see	B-Application
computational	O
creativity	O
,	O
automated	O
reasoning	O
,	O
decision	B-Application
support	I-Application
system	I-Application
,	O
robot	B-Algorithm
,	O
evolutionary	O
computation	O
,	O
intelligent	B-General_Concept
agent	I-General_Concept
)	O
;	O
however	O
,	O
there	O
is	O
disagreement	O
as	O
to	O
whether	O
modern	O
AI	B-Application
systems	O
possess	O
all	O
of	O
them	O
to	O
an	O
adequate	O
degree	O
.	O
</s>
<s>
A	O
mathematically	O
precise	O
specification	O
of	O
AGI	B-General_Concept
was	O
proposed	O
by	O
Marcus	O
Hutter	O
in	O
2000	O
.	O
</s>
<s>
Named	O
AIXI	O
,	O
the	O
proposed	O
AGI	B-General_Concept
agent	O
maximises	O
“	O
the	O
ability	O
to	O
satisfy	O
goals	O
in	O
a	O
wide	O
range	O
of	O
environments	O
”	O
.	O
</s>
<s>
This	O
type	O
of	O
AGI	B-General_Concept
,	O
characterized	O
by	O
the	O
ability	O
to	O
maximise	O
a	O
mathematical	O
definition	O
of	O
intelligence	O
rather	O
than	O
exhibit	O
human-like	O
behaviour	O
,	O
is	O
also	O
called	O
universal	O
artificial	B-Application
intelligence	I-Application
.	I-Application
</s>
<s>
If	O
enactivism	O
holds	O
and	O
Mind	O
–	O
body	O
dualism	O
does	O
not	O
,	O
then	O
compression	O
is	O
not	O
necessary	O
or	O
sufficient	O
for	O
intelligence	O
,	O
calling	O
into	O
question	O
widely	O
held	O
views	O
on	O
intelligence	O
(	O
see	B-Application
also	O
Hutter	O
Prize	O
)	O
.	O
</s>
<s>
Whether	O
an	O
AGI	B-General_Concept
that	O
satisfies	O
one	O
of	O
these	O
formalizations	O
exhibits	O
human-like	O
behaviour	O
(	O
such	O
as	O
the	O
use	O
of	O
natural	B-Language
language	I-Language
)	O
would	O
depend	O
on	O
many	O
factors	O
,	O
for	O
example	O
the	O
manner	O
in	O
which	O
the	O
agent	O
is	O
embodied	O
,	O
or	O
whether	O
it	O
has	O
a	O
reward	O
function	O
that	O
closely	O
approximates	O
human	O
primitives	O
of	O
cognition	O
like	O
hunger	O
,	O
pain	O
,	O
and	O
so	O
forth	O
.	O
</s>
<s>
Several	O
tests	O
meant	O
to	O
confirm	O
human-level	O
AGI	B-General_Concept
have	O
been	O
considered	O
,	O
including	O
:	O
</s>
<s>
For	O
example	O
,	O
even	O
specific	O
straightforward	O
tasks	O
,	O
like	O
machine	B-Application
translation	I-Application
,	O
require	O
that	O
a	O
machine	O
read	O
and	O
write	O
in	O
both	O
languages	O
(	O
NLP	B-Language
)	O
,	O
follow	O
the	O
author	O
's	O
argument	O
(	O
reason	O
)	O
,	O
know	O
what	O
is	O
being	O
talked	O
about	O
(	O
knowledge	O
)	O
,	O
and	O
faithfully	O
reproduce	O
the	O
author	O
's	O
original	O
intent	O
(	O
social	O
intelligence	O
)	O
.	O
</s>
<s>
A	O
problem	O
is	O
informally	O
called	O
"	O
AI-complete	O
"	O
or	O
"	O
AI-hard	O
"	O
if	O
it	O
is	O
believed	O
that	O
to	O
solve	O
it	O
one	O
would	O
need	O
to	O
implement	O
strong	O
AI	B-Application
,	O
because	O
the	O
solution	O
is	O
beyond	O
the	O
capabilities	O
of	O
a	O
purpose-specific	O
algorithm	O
.	O
</s>
<s>
AI-complete	O
problems	O
are	O
hypothesised	O
to	O
include	O
general	O
computer	B-Application
vision	I-Application
,	O
natural	B-General_Concept
language	I-General_Concept
understanding	I-General_Concept
,	O
and	O
dealing	O
with	O
unexpected	O
circumstances	O
while	O
solving	O
any	O
real-world	O
problem	O
.	O
</s>
<s>
AI-complete	O
problems	O
cannot	O
be	O
solved	O
with	O
current	O
computer	O
technology	O
alone	O
,	O
and	O
require	O
human	O
computation	O
.	O
</s>
<s>
Modern	O
AI	B-Application
research	I-Application
began	O
in	O
the	O
mid-1950s	O
.	O
</s>
<s>
The	O
first	O
generation	O
of	O
AI	B-Application
researchers	O
were	O
convinced	O
that	O
artificial	O
general	O
intelligence	O
was	O
possible	O
and	O
that	O
it	O
would	O
exist	O
in	O
just	O
a	O
few	O
decades	O
.	O
</s>
<s>
AI	B-Application
pioneer	O
Herbert	O
A	O
.	O
Simon	O
wrote	O
in	O
1965	O
:	O
"	O
machines	O
will	O
be	O
capable	O
,	O
within	O
twenty	O
years	O
,	O
of	O
doing	O
any	O
work	O
a	O
man	O
can	O
do.	O
"	O
</s>
<s>
Their	O
predictions	O
were	O
the	O
inspiration	O
for	O
Stanley	O
Kubrick	O
and	O
Arthur	O
C	O
.	O
Clarke	O
's	O
character	O
HAL	B-General_Concept
9000	I-General_Concept
,	O
who	O
embodied	O
what	O
AI	B-Application
researchers	O
believed	O
they	O
could	O
create	O
by	O
the	O
year	O
2001	O
.	O
</s>
<s>
AI	B-Application
pioneer	O
Marvin	O
Minsky	O
was	O
a	O
consultant	O
on	O
the	O
project	O
of	O
making	O
HAL	B-General_Concept
9000	I-General_Concept
as	O
realistic	O
as	O
possible	O
according	O
to	O
the	O
consensus	O
predictions	O
of	O
the	O
time	O
.	O
</s>
<s>
He	O
said	O
in	O
1967	O
,	O
"	O
Within	O
a	O
generation	O
...	O
the	O
problem	O
of	O
creating	O
'	O
artificial	B-Application
intelligence	I-Application
 '	O
will	O
substantially	O
be	O
solved	O
"	O
.	O
</s>
<s>
Several	O
classical	B-General_Concept
AI	I-General_Concept
projects	I-General_Concept
,	O
such	O
as	O
Doug	O
Lenat	O
's	O
Cyc	B-Language
project	I-Language
(	O
that	O
began	O
in	O
1984	O
)	O
,	O
and	O
Allen	O
Newell	O
's	O
Soar	B-Architecture
project	O
,	O
were	O
directed	O
at	O
AGI	B-General_Concept
.	O
</s>
<s>
Funding	O
agencies	O
became	O
skeptical	O
of	O
AGI	B-General_Concept
and	O
put	O
researchers	O
under	O
increasing	O
pressure	O
to	O
produce	O
useful	O
"	O
applied	O
AI	B-Application
"	O
.	O
</s>
<s>
In	O
the	O
early	O
1980s	O
,	O
Japan	O
's	O
Fifth	B-Device
Generation	I-Device
Computer	I-Device
Project	O
revived	O
interest	O
in	O
AGI	B-General_Concept
,	O
setting	O
out	O
a	O
ten-year	O
timeline	O
that	O
included	O
AGI	B-General_Concept
goals	O
like	O
"	O
carry	O
on	O
a	O
casual	O
conversation	O
"	O
.	O
</s>
<s>
In	O
response	O
to	O
this	O
and	O
the	O
success	O
of	O
expert	B-Application
systems	I-Application
,	O
both	O
industry	O
and	O
government	O
pumped	O
money	O
back	O
into	O
the	O
field	O
.	O
</s>
<s>
However	O
,	O
confidence	O
in	O
AI	B-Application
spectacularly	O
collapsed	O
in	O
the	O
late	O
1980s	O
,	O
and	O
the	O
goals	O
of	O
the	O
Fifth	B-Device
Generation	I-Device
Computer	I-Device
Project	O
were	O
never	O
fulfilled	O
.	O
</s>
<s>
For	O
the	O
second	O
time	O
in	O
20	O
years	O
,	O
AI	B-Application
researchers	O
who	O
predicted	O
the	O
imminent	O
achievement	O
of	O
AGI	B-General_Concept
had	O
been	O
mistaken	O
.	O
</s>
<s>
By	O
the	O
1990s	O
,	O
AI	B-Application
researchers	O
had	O
a	O
reputation	O
for	O
making	O
vain	O
promises	O
.	O
</s>
<s>
They	O
became	O
reluctant	O
to	O
make	O
predictions	O
at	O
all	O
and	O
avoided	O
mention	O
of	O
"	O
human	O
level	O
"	O
artificial	B-Application
intelligence	I-Application
for	O
fear	O
of	O
being	O
labeled	O
"	O
wild-eyed	O
dreamer[s]	O
 "	O
.	O
</s>
<s>
In	O
the	O
1990s	O
and	O
early	O
21st	O
century	O
,	O
mainstream	O
AI	B-Application
achieved	O
commercial	O
success	O
and	O
academic	O
respectability	O
by	O
focusing	O
on	O
specific	O
sub-problems	O
where	O
AI	B-Application
can	O
produce	O
verifiable	O
results	O
and	O
commercial	O
applications	O
,	O
such	O
as	O
artificial	B-Architecture
neural	I-Architecture
networks	I-Architecture
and	O
statistical	O
machine	O
learning	O
.	O
</s>
<s>
These	O
"	O
applied	O
AI	B-Application
"	O
systems	O
are	O
now	O
used	O
extensively	O
throughout	O
the	O
technology	O
industry	O
,	O
and	O
research	O
in	O
this	O
vein	O
is	O
heavily	O
funded	O
in	O
both	O
academia	O
and	O
industry	O
.	O
</s>
<s>
Most	O
mainstream	O
AI	B-Application
researchers	O
hope	O
that	O
strong	O
AI	B-Application
can	O
be	O
developed	O
by	O
combining	O
programs	O
that	O
solve	O
various	O
sub-problems	O
.	O
</s>
<s>
Hans	O
Moravec	O
wrote	O
in	O
1988	O
:	O
I	O
am	O
confident	O
that	O
this	O
bottom-up	O
route	O
to	O
artificial	B-Application
intelligence	I-Application
will	O
one	O
day	O
meet	O
the	O
traditional	O
top-down	O
route	O
more	O
than	O
half	O
way	O
,	O
ready	O
to	O
provide	O
the	O
real	O
world	O
competence	O
and	O
the	O
commonsense	O
knowledge	O
that	O
has	O
been	O
so	O
frustratingly	O
elusive	O
in	O
reasoning	O
programs	O
.	O
</s>
<s>
Fully	O
intelligent	B-Application
machines	I-Application
will	O
result	O
when	O
the	O
metaphorical	O
golden	O
spike	O
is	O
driven	O
uniting	O
the	O
two	O
efforts	O
.	O
</s>
<s>
If	O
the	O
grounding	O
considerations	O
in	O
this	O
paper	O
are	O
valid	O
,	O
then	O
this	O
expectation	O
is	O
hopelessly	O
modular	O
and	O
there	O
is	O
really	O
only	O
one	O
viable	O
route	O
from	O
sense	B-General_Concept
to	O
symbols	O
:	O
from	O
the	O
ground	O
up	O
.	O
</s>
<s>
AGI	B-General_Concept
research	O
activity	O
in	O
2006	O
was	O
described	O
by	O
Pei	O
Wang	O
and	O
Ben	O
Goertzel	O
as	O
"	O
producing	O
publications	O
and	O
preliminary	O
results	O
"	O
.	O
</s>
<s>
The	O
first	O
summer	O
school	O
in	O
AGI	B-General_Concept
was	O
organized	O
in	O
Xiamen	O
,	O
China	O
in	O
2009	O
by	O
the	O
Xiamen	O
university	O
's	O
Artificial	O
Brain	O
Laboratory	O
and	O
OpenCog	O
.	O
</s>
<s>
MIT	O
presented	O
a	O
course	O
in	O
AGI	B-General_Concept
in	O
2018	O
,	O
organized	O
by	O
Lex	O
Fridman	O
and	O
featuring	O
a	O
number	O
of	O
guest	O
lecturers	O
.	O
</s>
<s>
,	O
most	O
AI	B-Application
researchers	O
devote	O
little	O
attention	O
to	O
AGI	B-General_Concept
,	O
with	O
some	O
claiming	O
that	O
intelligence	O
is	O
too	O
complex	O
to	O
be	O
completely	O
replicated	O
in	O
the	O
near	O
term	O
.	O
</s>
<s>
However	O
,	O
a	O
small	O
number	O
of	O
computer	O
scientists	O
are	O
active	O
in	O
AGI	B-General_Concept
research	O
,	O
and	O
many	O
contribute	O
to	O
a	O
series	O
of	O
AGI	B-Application
conferences	I-Application
.	O
</s>
<s>
In	O
the	O
introduction	O
to	O
his	O
2006	O
book	O
,	O
Goertzel	O
says	O
that	O
estimates	O
of	O
the	O
time	O
needed	O
before	O
a	O
truly	O
flexible	O
AGI	B-General_Concept
is	O
built	O
vary	O
from	O
10	O
years	O
to	O
over	O
a	O
century	O
.	O
</s>
<s>
the	O
consensus	O
in	O
the	O
AGI	B-General_Concept
research	O
community	O
seemed	O
to	O
be	O
that	O
the	O
timeline	O
discussed	O
by	O
Ray	O
Kurzweil	O
in	O
The	O
Singularity	O
is	O
Near	O
(	O
i.e.	O
</s>
<s>
Mainstream	O
AI	B-Application
researchers	O
have	O
given	O
a	O
wide	O
range	O
of	O
opinions	O
on	O
whether	O
progress	O
will	O
be	O
this	O
rapid	O
.	O
</s>
<s>
A	O
2012	O
meta-analysis	O
of	O
95	O
such	O
opinions	O
found	O
a	O
bias	O
towards	O
predicting	O
that	O
the	O
onset	O
of	O
AGI	B-General_Concept
would	O
occur	O
within	O
16	O
–	O
26	O
years	O
for	O
modern	O
and	O
historical	O
predictions	O
alike	O
.	O
</s>
<s>
In	O
2017	O
,	O
researchers	O
Feng	O
Liu	O
,	O
Yong	O
Shi	O
,	O
and	O
Ying	O
Liu	O
conducted	O
intelligence	O
tests	O
on	O
publicly	O
available	O
and	O
freely	O
accessible	O
weak	O
AI	B-Application
such	O
as	O
Google	O
AI	B-Application
,	O
Apple	O
's	O
Siri	O
,	O
and	O
others	O
.	O
</s>
<s>
At	O
the	O
maximum	O
,	O
these	O
AI	B-Application
reached	O
an	O
IQ	O
value	O
of	O
about	O
47	O
,	O
which	O
corresponds	O
approximately	O
to	O
a	O
six-year-old	O
child	O
in	O
first	O
grade	O
.	O
</s>
<s>
In	O
2020	O
,	O
OpenAI	O
developed	O
GPT-3	B-General_Concept
,	O
a	O
language	O
model	O
capable	O
of	O
performing	O
many	O
diverse	O
tasks	O
without	O
specific	O
training	O
.	O
</s>
<s>
According	O
to	O
Gary	O
Grossman	O
in	O
a	O
VentureBeat	O
article	O
,	O
while	O
there	O
is	O
consensus	O
that	O
GPT-3	B-General_Concept
is	O
not	O
an	O
example	O
of	O
AGI	B-General_Concept
,	O
it	O
is	O
considered	O
by	O
some	O
to	O
be	O
too	O
advanced	O
to	O
classify	O
as	O
a	O
narrow	O
AI	B-Application
system	O
.	O
</s>
<s>
In	O
the	O
same	O
year	O
Jason	O
Rohrer	O
used	O
his	O
GPT-3	B-General_Concept
account	O
to	O
develop	O
a	O
chatbot	O
,	O
and	O
provided	O
a	O
chatbot-developing	O
platform	O
called	O
"	O
Project	O
December	O
"	O
.	O
</s>
<s>
OpenAI	O
asked	O
for	O
changes	O
to	O
the	O
chatbot	O
to	O
comply	O
with	O
their	O
safety	O
guidelines	O
;	O
Rohrer	O
disconnected	O
Project	O
December	O
from	O
the	O
GPT-3	B-General_Concept
API	O
.	O
</s>
<s>
In	O
2022	O
,	O
DeepMind	B-Application
developed	O
Gato	B-Application
,	O
a	O
"	O
general-purpose	O
"	O
system	O
capable	O
of	O
performing	O
more	O
than	O
600	O
different	O
tasks	O
.	O
</s>
<s>
In	O
2023	O
,	O
Microsoft	O
Research	O
published	O
a	O
study	O
on	O
an	O
early	O
version	O
of	O
OpenAI	O
's	O
GPT-4	O
,	O
contending	O
that	O
it	O
exhibited	O
more	O
general	O
intelligence	O
than	O
previous	O
AI	B-Application
models	O
and	O
demonstrated	O
human-level	O
performance	O
in	O
tasks	O
spanning	O
multiple	O
domains	O
,	O
such	O
as	O
mathematics	O
,	O
coding	O
,	O
and	O
law	O
.	O
</s>
<s>
One	O
possible	O
approach	O
to	O
achieving	O
AGI	B-General_Concept
is	O
whole	O
brain	O
emulation	O
:	O
A	O
brain	O
model	O
is	O
built	O
by	O
scanning	B-Algorithm
and	O
mapping	O
a	O
biological	O
brain	O
in	O
detail	O
and	O
copying	O
its	O
state	O
into	O
a	O
computer	O
system	O
or	O
another	O
computational	O
device	O
.	O
</s>
<s>
The	O
computer	O
runs	O
a	O
simulation	B-Application
model	O
sufficiently	O
faithful	O
to	O
the	O
original	O
that	O
it	O
behaves	O
in	O
practically	O
the	O
same	O
way	O
as	O
the	O
original	O
brain	O
.	O
</s>
<s>
Whole	O
brain	O
emulation	O
is	O
discussed	O
in	O
computational	O
neuroscience	O
and	O
neuroinformatics	O
,	O
in	O
the	O
context	O
of	O
brain	O
simulation	B-Application
for	O
medical	O
research	O
purposes	O
.	O
</s>
<s>
It	O
is	O
discussed	O
in	O
artificial	B-Application
intelligence	I-Application
research	I-Application
as	O
an	O
approach	O
to	O
strong	O
AI	B-Application
.	O
</s>
<s>
Neuroimaging	B-Algorithm
technologies	O
that	O
could	O
deliver	O
the	O
necessary	O
detailed	O
understanding	O
are	O
improving	O
rapidly	O
,	O
and	O
futurist	O
Ray	O
Kurzweil	O
in	O
the	O
book	O
The	O
Singularity	O
Is	O
Near	O
predicts	O
that	O
a	O
map	O
of	O
sufficient	O
quality	O
will	O
become	O
available	O
on	O
a	O
similar	O
timescale	O
to	O
the	O
computing	O
power	O
required	O
to	O
emulate	O
it	O
.	O
</s>
<s>
For	O
low-level	O
brain	O
simulation	B-Application
,	O
an	O
extremely	O
powerful	O
computer	O
would	O
be	O
required	O
.	O
</s>
<s>
The	O
human	O
brain	O
has	O
a	O
huge	O
number	O
of	O
synapses	B-Application
.	O
</s>
<s>
Each	O
of	O
the	O
1011	O
(	O
one	O
hundred	O
billion	O
)	O
neurons	O
has	O
on	O
average	O
7,000	O
synaptic	O
connections	O
(	O
synapses	B-Application
)	O
to	O
other	O
neurons	O
.	O
</s>
<s>
The	O
brain	O
of	O
a	O
three-year-old	O
child	O
has	O
about	O
1015	O
synapses	B-Application
(	O
1	O
quadrillion	O
)	O
.	O
</s>
<s>
Estimates	O
vary	O
for	O
an	O
adult	O
,	O
ranging	O
from	O
1014	O
to	O
5×1014	O
synapses	B-Application
(	O
100	O
to	O
500	O
trillion	O
)	O
.	O
</s>
<s>
(	O
For	O
comparison	O
,	O
if	O
a	O
"	O
computation	O
"	O
was	O
equivalent	O
to	O
one	O
"	O
floating-point	O
operation	O
"	O
–	O
a	O
measure	O
used	O
to	O
rate	O
current	O
supercomputers	B-Architecture
–	O
then	O
1016	O
"	O
computations	O
"	O
would	O
be	O
equivalent	O
to	O
10	O
petaFLOPS	O
,	O
achieved	O
in	O
2011	O
,	O
while	O
1018	O
was	O
achieved	B-General_Concept
in	I-General_Concept
2022	I-General_Concept
.	O
)	O
</s>
<s>
The	O
artificial	B-Algorithm
neuron	I-Algorithm
model	O
assumed	O
by	O
Kurzweil	O
and	O
used	O
in	O
many	O
current	O
artificial	B-Architecture
neural	I-Architecture
network	I-Architecture
implementations	O
is	O
simple	O
compared	O
with	O
biological	O
neurons	O
.	O
</s>
<s>
A	O
brain	O
simulation	B-Application
would	O
likely	O
have	O
to	O
capture	O
the	O
detailed	O
cellular	O
behaviour	O
of	O
biological	O
neurons	O
,	O
presently	O
understood	O
only	O
in	O
broad	O
outline	O
.	O
</s>
<s>
Some	O
research	O
projects	O
are	O
investigating	O
brain	O
simulation	B-Application
using	O
more	O
sophisticated	O
neural	O
models	O
,	O
implemented	O
on	O
conventional	O
computing	O
architectures	O
.	O
</s>
<s>
The	B-Application
Artificial	I-Application
Intelligence	I-Application
System	O
project	O
implemented	O
non-real	O
time	O
simulations	O
of	O
a	O
"	O
brain	O
"	O
(	O
with	O
1011	O
neurons	O
)	O
in	O
2005	O
.	O
</s>
<s>
The	O
Blue	O
Brain	O
project	O
used	O
one	O
of	O
the	O
fastest	B-Operating_System
supercomputer	I-Operating_System
architectures	O
,	O
IBM	O
's	O
Blue	B-Operating_System
Gene	I-Operating_System
platform	O
,	O
to	O
create	O
a	O
real	O
time	O
simulation	B-Application
of	O
a	O
single	O
rat	O
neocortical	O
column	O
consisting	O
of	O
approximately	O
10,000	O
neurons	O
and	O
108	O
synapses	B-Application
in	O
2006	O
.	O
</s>
<s>
A	O
longer-term	O
goal	O
is	O
to	O
build	O
a	O
detailed	O
,	O
functional	O
simulation	B-Application
of	O
the	O
physiological	O
processes	O
in	O
the	O
human	O
brain	O
:	O
"	O
It	O
is	O
not	O
impossible	O
to	O
build	O
a	O
human	O
brain	O
and	O
we	O
can	O
do	O
it	O
in	O
10	O
years	O
,	O
"	O
Henry	O
Markram	O
,	O
director	O
of	O
the	O
Blue	O
Brain	O
Project	O
,	O
said	O
in	O
2009	O
at	O
the	O
TED	O
conference	O
in	O
Oxford	O
.	O
</s>
<s>
The	O
actual	O
complexity	O
of	O
modeling	O
biological	O
neurons	O
has	O
been	O
explored	O
in	O
OpenWorm	B-Application
project	I-Application
that	O
aimed	O
at	O
complete	O
simulation	B-Application
of	O
a	O
worm	O
that	O
has	O
only	O
302	O
neurons	O
in	O
its	O
neural	B-Architecture
network	I-Architecture
(	O
among	O
about	O
1000	O
cells	O
in	O
total	O
)	O
.	O
</s>
<s>
The	O
animal	O
's	O
neural	B-Architecture
network	I-Architecture
was	O
well	O
documented	O
before	O
the	O
start	O
of	O
the	O
project	O
.	O
</s>
<s>
However	O
,	O
although	O
the	O
task	O
seemed	O
simple	O
at	O
the	O
beginning	O
,	O
the	O
models	O
based	O
on	O
a	O
generic	O
neural	B-Architecture
network	I-Architecture
did	O
not	O
work	O
.	O
</s>
<s>
Goertzel	O
proposes	O
virtual	O
embodiment	O
(	O
like	O
in	O
Second	B-Application
Life	I-Application
)	O
as	O
an	O
option	O
,	O
but	O
it	O
is	O
unknown	O
whether	O
this	O
would	O
be	O
sufficient	O
.	O
</s>
<s>
Desktop	O
computers	O
using	O
microprocessors	O
capable	O
of	O
more	O
than	O
109	O
cps	O
(	O
Kurzweil	O
's	O
non-standard	O
unit	O
"	O
computations	O
per	O
second	O
"	O
,	O
see	B-Application
above	O
)	O
have	O
been	O
available	O
since	O
2005	O
.	O
</s>
<s>
According	O
to	O
the	O
brain	O
power	O
estimates	O
used	O
by	O
Kurzweil	O
(	O
and	O
Moravec	O
)	O
,	O
such	O
a	O
computer	O
should	O
be	O
capable	O
of	O
supporting	O
a	O
simulation	B-Application
of	O
a	O
bee	O
brain	O
,	O
but	O
despite	O
some	O
interest	O
no	O
such	O
simulation	B-Application
exists	O
.	O
</s>
<s>
The	O
neuron	O
model	O
seems	O
to	O
be	O
oversimplified	O
(	O
see	B-Application
next	O
section	O
)	O
.	O
</s>
<s>
Even	O
if	O
our	O
understanding	O
of	O
cognition	O
advances	O
sufficiently	O
,	O
early	O
simulation	B-Application
programs	O
are	O
likely	O
to	O
be	O
very	O
inefficient	O
and	O
will	O
,	O
therefore	O
,	O
need	O
considerably	O
more	O
hardware	O
.	O
</s>
<s>
One	O
estimate	O
puts	O
the	O
human	O
brain	O
at	O
about	O
100	O
billion	O
neurons	O
and	O
100	O
trillion	O
synapses	B-Application
.	O
</s>
<s>
Glial	O
cell	O
synapses	B-Application
are	O
currently	O
unquantified	O
but	O
are	O
known	O
to	O
be	O
extremely	O
numerous	O
.	O
</s>
<s>
In	O
1980	O
,	O
philosopher	O
John	O
Searle	O
coined	O
the	O
term	O
"	O
strong	O
AI	B-Application
"	O
as	O
part	O
of	O
his	O
Chinese	O
room	O
argument	O
.	O
</s>
<s>
He	O
wanted	O
to	O
distinguish	O
between	O
two	O
different	O
hypotheses	O
about	O
artificial	B-Application
intelligence	I-Application
:	O
</s>
<s>
Strong	O
AI	B-Application
hypothesis	O
:	O
An	O
artificial	B-Operating_System
intelligence	I-Operating_System
system	I-Operating_System
can	O
"	O
think	O
"	O
—	O
have	O
"	O
a	O
mind	O
"	O
and	O
"	O
consciousness	O
"	O
.	O
</s>
<s>
Weak	O
AI	B-Application
hypothesis	O
:	O
An	O
artificial	B-Operating_System
intelligence	I-Operating_System
system	I-Operating_System
can	O
(	O
only	O
)	O
act	O
like	O
it	O
thinks	O
and	O
has	O
a	O
mind	O
and	O
consciousness	O
.	O
</s>
<s>
The	O
behaviour	O
of	O
a	O
"	O
weak	O
AI	B-Application
"	O
machine	O
would	O
be	O
precisely	O
identical	O
to	O
a	O
"	O
strong	O
AI	B-Application
"	O
machine	O
,	O
but	O
the	O
latter	O
would	O
also	O
have	O
subjective	O
conscious	O
experience	O
.	O
</s>
<s>
This	O
usage	O
is	O
also	O
common	O
in	O
academic	O
AI	B-Application
research	I-Application
and	O
textbooks	O
.	O
</s>
<s>
Mainstream	O
AI	B-Application
is	O
most	O
interested	O
in	O
how	O
a	O
program	O
behaves	O
.	O
</s>
<s>
According	O
to	O
Russell	O
and	O
Norvig	O
,	O
"	O
as	O
long	O
as	O
the	O
program	O
works	O
,	O
they	O
do	O
n't	O
care	O
if	O
you	O
call	O
it	O
real	O
or	O
a	O
simulation.	O
"	O
</s>
<s>
For	O
AI	B-Application
research	I-Application
,	O
Searle	O
's	O
"	O
weak	O
AI	B-Application
hypothesis	O
"	O
is	O
equivalent	O
to	O
the	O
statement	O
"	O
artificial	O
general	O
intelligence	O
is	O
possible	O
"	O
.	O
</s>
<s>
Thus	O
,	O
according	O
to	O
Russell	O
and	O
Norvig	O
,	O
"	O
most	O
AI	B-Application
researchers	O
take	O
the	O
weak	O
AI	B-Application
hypothesis	O
for	O
granted	O
,	O
and	O
do	O
n't	O
care	O
about	O
the	O
strong	O
AI	B-Application
hypothesis.	O
"	O
</s>
<s>
Thus	O
,	O
for	O
academic	O
AI	B-Application
research	I-Application
,	O
"	O
Strong	O
AI	B-Application
"	O
and	O
"	O
AGI	B-General_Concept
"	O
are	O
two	O
very	O
different	O
things	O
.	O
</s>
<s>
In	O
contrast	O
to	O
Searle	O
and	O
mainstream	O
AI	B-Application
,	O
some	O
futurists	O
such	O
as	O
Ray	O
Kurzweil	O
use	O
the	O
term	O
"	O
strong	O
AI	B-Application
"	O
to	O
mean	O
"	O
human	O
level	O
artificial	O
general	O
intelligence	O
"	O
.	O
</s>
<s>
This	O
is	O
not	O
the	O
same	O
as	O
Searle	O
's	O
strong	O
AI	B-Application
,	O
unless	O
you	O
assume	O
that	O
consciousness	O
is	O
necessary	O
for	O
human-level	O
AGI	B-General_Concept
.	O
</s>
<s>
Academic	O
philosophers	O
such	O
as	O
Searle	O
do	O
not	O
believe	O
that	O
is	O
the	O
case	O
,	O
and	O
to	O
most	O
artificial	B-Application
intelligence	I-Application
researchers	O
the	O
question	O
is	O
out-of-scope	O
.	O
</s>
<s>
Other	O
aspects	O
of	O
the	O
human	O
mind	O
besides	O
intelligence	O
are	O
relevant	O
to	O
the	O
concept	O
of	O
strong	O
AI	B-Application
,	O
and	O
these	O
play	O
a	O
major	O
role	O
in	O
science	O
fiction	O
and	O
the	O
ethics	O
of	O
artificial	B-Application
intelligence	I-Application
:	O
</s>
<s>
These	O
traits	O
have	O
a	O
moral	O
dimension	O
,	O
because	O
a	O
machine	O
with	O
this	O
form	O
of	O
strong	O
AI	B-Application
may	O
have	O
rights	O
,	O
analogous	O
to	O
the	O
rights	O
of	O
non-human	O
animals	O
.	O
</s>
<s>
Preliminary	O
work	O
has	O
been	O
conducted	O
on	O
integrating	O
with	O
existing	O
legal	O
and	O
social	O
frameworks	O
,	O
focusing	O
on	O
the	O
legal	O
position	O
and	O
rights	O
of	O
'	O
strong	O
 '	O
AI	B-Application
.	O
</s>
<s>
It	O
remains	O
to	O
be	O
shown	O
whether	O
any	O
of	O
these	O
traits	O
are	O
necessary	O
for	O
strong	O
AI	B-Application
.	O
</s>
<s>
It	O
is	O
possible	O
that	O
some	O
of	O
these	O
traits	O
naturally	O
emerge	O
from	O
a	O
fully	O
intelligent	B-Application
machine	I-Application
.	O
</s>
<s>
Although	O
the	O
role	O
of	O
consciousness	O
in	O
strong	O
AI/AGI	O
is	O
debatable	O
,	O
many	O
AGI	B-General_Concept
researchers	O
regard	O
research	O
that	O
investigates	O
possibilities	O
for	O
implementing	O
consciousness	O
as	O
vital	O
.	O
</s>
<s>
Since	O
the	O
launch	O
of	O
AI	B-Application
research	I-Application
in	O
1956	O
,	O
progress	O
in	O
this	O
field	O
of	O
creating	O
machines	O
skilled	O
with	O
intelligent	O
action	O
at	O
the	O
human	O
level	O
has	O
slowed	O
.	O
</s>
<s>
One	O
possible	O
reason	O
is	O
that	O
computers	O
lack	O
sufficient	O
memory	O
,	O
processing	O
power	O
,	O
or	O
chip	O
flexibility	O
to	O
accommodate	O
computer-science-oriented	O
and/or	O
neuroscience-oriented	O
AGI	B-General_Concept
platforms	O
.	O
</s>
<s>
Also	O
the	O
complexity	O
involved	O
in	O
AI	B-Application
research	I-Application
likely	O
also	O
limits	O
the	O
progress	O
of	O
strong	O
AI	B-Application
research	I-Application
.	O
</s>
<s>
Conceptual	O
limitations	O
are	O
another	O
possible	O
reason	O
for	O
the	O
slowness	O
in	O
AI	B-Application
research	I-Application
.	O
</s>
<s>
AI	B-Application
researchers	O
may	O
need	O
to	O
modify	O
the	O
conceptual	O
framework	O
of	O
their	O
discipline	O
in	O
order	O
to	O
provide	O
a	O
stronger	O
foundation	O
to	O
the	O
quest	O
of	O
achieving	O
strong	O
AI	B-Application
.	O
</s>
<s>
This	O
means	O
situating	O
a	O
strong	O
AI	B-Application
in	O
a	O
sociocultural	O
context	O
where	O
human-like	O
AI	B-Application
derives	O
from	O
human-like	O
experiences	O
.	O
</s>
<s>
AI	B-Application
researchers	O
have	O
been	O
able	O
to	O
create	O
computers	O
that	O
can	O
perform	O
jobs	O
that	O
are	O
complicated	O
for	O
people	O
to	O
do	O
,	O
such	O
as	O
mathematics	O
,	O
but	O
they	O
have	O
struggled	O
to	O
develop	O
a	O
computer	O
that	O
is	O
capable	O
of	O
carrying	O
out	O
tasks	O
that	O
are	O
simple	O
for	O
humans	O
to	O
do	O
,	O
such	O
as	O
walking	O
(	O
Moravec	O
's	O
paradox	O
)	O
.	O
</s>
<s>
The	O
idea	O
of	O
whether	O
thoughts	O
and	O
the	O
creator	O
of	O
those	O
thoughts	O
are	O
isolated	O
or	O
must	O
be	O
socially	O
situated	O
has	O
intrigued	O
AI	B-Application
researchers	O
.	O
</s>
<s>
The	O
problems	O
encountered	O
in	O
AI	B-Application
research	I-Application
over	O
the	O
past	O
decades	O
have	O
further	O
impeded	O
the	O
progress	O
of	O
AGI	B-General_Concept
research	O
and	O
development	O
by	O
generating	O
a	O
degree	O
of	O
distrust	O
in	O
the	O
field	O
.	O
</s>
<s>
The	O
failed	O
predictions	O
of	O
success	O
from	O
AI	B-Application
researchers	O
and	O
the	O
lack	O
of	O
a	O
complete	O
understanding	O
of	O
human	O
behaviours	O
have	O
diminished	O
optimism	O
in	O
the	O
idea	O
of	O
creating	O
human-level	O
AI	B-Application
.	O
</s>
<s>
The	O
waxing	O
and	O
waning	O
progress	O
of	O
AI	B-Application
research	I-Application
has	O
brought	O
both	O
improvement	O
and	O
disappointment	O
.	O
</s>
<s>
Most	O
investigators	O
are	O
optimistic	O
about	O
achieving	O
AGI	B-General_Concept
in	O
the	O
21st	O
century	O
.	O
</s>
<s>
Other	O
possible	O
reasons	O
have	O
been	O
proposed	O
for	O
the	O
slow	O
progress	O
towards	O
strong	O
AI	B-Application
.	O
</s>
<s>
Many	O
researchers	O
tend	O
to	O
underestimate	O
any	O
doubt	O
that	O
is	O
involved	O
with	O
future	O
predictions	O
of	O
AI	B-Application
,	O
but	O
without	O
taking	O
issues	O
like	O
human	O
brain	O
modelling	O
seriously	O
,	O
AGI	B-General_Concept
researchers	O
then	O
overlook	O
solutions	O
to	O
problematic	O
questions	O
.	O
</s>
<s>
Clocksin	O
says	O
AI	B-Application
researchers	O
may	O
be	O
using	O
the	O
wrong	O
techniques	O
for	O
computer	O
programs	O
and	O
for	O
the	O
implementation	O
of	O
equipment	O
.	O
</s>
<s>
When	O
AI	B-Application
researchers	O
first	O
began	O
to	O
aim	O
for	O
AGI	B-General_Concept
,	O
a	O
main	O
interest	O
was	O
to	O
emulate	O
and	O
investigate	O
human	O
reasoning	O
.	O
</s>
<s>
In	O
response	O
,	O
the	O
practice	O
of	O
abstraction	O
,	O
which	O
people	O
tend	O
to	O
redefine	O
when	O
working	O
with	O
a	O
particular	O
context	O
in	O
research	O
,	O
provides	O
AI	B-Application
researchers	O
with	O
the	O
option	O
to	O
concentrate	O
on	O
just	O
a	O
few	O
concepts	O
.	O
</s>
<s>
The	O
most	O
productive	O
use	O
of	O
abstraction	O
in	O
AI	B-Application
research	I-Application
comes	O
from	O
planning	O
and	O
problem	O
solving	O
.	O
</s>
<s>
Human	O
heuristics	B-Algorithm
remain	O
superior	O
to	O
computer	O
performance	O
in	O
some	O
domains	O
.	O
</s>
<s>
But	O
specific	O
functions	O
programmed	O
into	O
increasingly	O
powerful	O
computers	O
may	O
be	O
able	O
to	O
implement	O
heuristics	B-Algorithm
that	O
eventually	O
allow	O
AI	B-Application
to	O
match	O
human	O
intelligence	O
.	O
</s>
<s>
While	O
heuristics	B-Algorithm
are	O
not	O
necessarily	O
a	O
fundamental	O
barrier	O
to	O
achieving	O
strong	O
AI	B-Application
,	O
to	O
be	O
a	O
challenge	O
.	O
</s>
<s>
AI	B-Application
researchers	O
have	O
debated	O
whether	O
or	O
not	O
machines	O
should	O
be	O
created	O
with	O
emotions	O
.	O
</s>
<s>
There	O
are	O
no	O
emotions	O
in	O
typical	O
models	O
of	O
AI	B-Application
,	O
and	O
some	O
researchers	O
say	O
programming	O
emotions	O
into	O
machines	O
allows	O
such	O
machines	O
.	O
</s>
<s>
Emotion	O
is	O
likely	O
to	O
continue	O
to	O
be	O
a	O
topic	O
of	O
strong	O
AI	B-Application
research	I-Application
.	O
</s>
<s>
,	O
AGI	B-General_Concept
remains	O
speculative	O
.	O
</s>
<s>
AI	B-Application
pioneer	O
Herbert	O
A	O
.	O
Simon	O
speculated	O
in	O
1965	O
that	O
"	O
machines	O
will	O
be	O
capable	O
,	O
within	O
twenty	O
years	O
,	O
of	O
doing	O
any	O
work	O
a	O
man	O
can	O
do	O
"	O
.	O
</s>
<s>
Writing	O
in	O
The	O
Guardian	O
,	O
roboticist	O
Alan	O
Winfield	O
claimed	O
the	O
gulf	O
between	O
modern	O
computing	O
and	O
human-level	O
artificial	B-Application
intelligence	I-Application
is	O
as	O
wide	O
as	O
the	O
gulf	O
between	O
current	O
space	O
flight	O
and	O
practical	O
faster-than-light	O
spaceflight	O
.	O
</s>
<s>
Most	O
AI	B-Application
researchers	O
believe	O
strong	O
AI	B-Application
can	O
be	O
achieved	O
in	O
the	O
future	O
,	O
but	O
some	O
people	O
,	O
like	O
Hubert	O
Dreyfus	O
and	O
Roger	O
Penrose	O
,	O
deny	O
the	O
possibility	O
of	O
achieving	O
strong	O
AI	B-Application
.	O
</s>
<s>
John	O
McCarthy	O
is	O
among	O
those	O
who	O
believe	O
human-level	O
AI	B-Application
will	O
be	O
accomplished	O
,	O
but	O
that	O
the	O
present	O
level	O
of	O
progress	O
is	O
such	O
that	O
a	O
date	O
cannot	O
accurately	O
be	O
predicted	O
.	O
</s>
<s>
AI	B-Application
experts	O
 '	O
views	O
on	O
the	O
feasibility	O
of	O
AGI	B-General_Concept
wax	O
and	O
wane	O
.	O
</s>
<s>
Four	O
polls	O
conducted	O
in	O
2012	O
and	O
2013	O
suggested	O
that	O
the	O
median	O
guess	O
among	O
experts	O
for	O
when	O
they	O
would	O
be	O
50%	O
confident	O
AGI	B-General_Concept
would	O
arrive	O
was	O
2040	O
to	O
2050	O
,	O
depending	O
on	O
the	O
poll	O
,	O
with	O
the	O
mean	O
being	O
2081	O
.	O
</s>
<s>
Further	O
current	O
AGI	B-General_Concept
progress	O
considerations	O
can	O
be	O
found	O
above	O
Tests	O
for	O
confirming	O
human-level	O
AGI	B-General_Concept
.	O
</s>
<s>
A	O
report	O
by	O
Stuart	O
Armstrong	O
and	O
Kaj	O
Sotala	O
of	O
the	O
Machine	B-Application
Intelligence	I-Application
Research	O
Institute	O
found	O
that	O
"	O
over	O
 [ a ] 	O
60	O
year	O
time	O
frame	O
there	O
is	O
a	O
strong	O
bias	O
towards	O
predicting	O
the	O
arrival	O
of	O
human	O
level	O
AI	B-Application
as	O
between	O
15	O
and	O
25	O
years	O
from	O
the	O
time	O
the	O
prediction	O
was	O
made	O
"	O
.	O
</s>
<s>
They	O
analyzed	O
95	O
predictions	O
made	O
between	O
1950	O
and	O
2012	O
on	O
when	O
human	O
level	O
AI	B-Application
will	O
come	O
about	O
.	O
</s>
<s>
The	O
thesis	O
that	O
AI	B-Application
poses	O
an	O
existential	O
risk	O
for	O
humans	O
,	O
and	O
that	O
this	O
risk	O
needs	O
much	O
more	O
attention	O
than	O
it	O
currently	O
gets	O
,	O
has	O
been	O
endorsed	O
by	O
many	O
public	O
figures	O
including	O
Elon	O
Musk	O
,	O
Bill	O
Gates	O
,	O
and	O
Stephen	O
Hawking	O
.	O
</s>
<s>
AI	B-Application
researchers	O
like	O
Stuart	O
J	O
.	O
Russell	O
,	O
Roman	O
Yampolskiy	O
,	O
and	O
Alexey	O
Turchin	O
,	O
also	O
support	O
the	O
basic	O
thesis	O
of	O
a	O
potential	O
threat	O
to	O
humanity	O
.	O
</s>
<s>
A	O
2021	O
systematic	O
review	O
of	O
the	O
risks	O
associated	O
with	O
AGI	B-General_Concept
,	O
while	O
noting	O
the	O
paucity	O
of	O
data	O
,	O
found	O
the	O
following	O
potential	O
threats	O
:	O
"	O
AGI	B-General_Concept
removing	O
itself	O
from	O
the	O
control	O
of	O
human	O
owners/managers	O
,	O
being	O
given	O
or	O
developing	O
unsafe	O
goals	O
,	O
development	O
of	O
unsafe	O
AGI	B-General_Concept
,	O
AGIs	B-General_Concept
with	O
poor	O
ethics	O
,	O
morals	O
and	O
values	O
;	O
inadequate	O
management	O
of	O
AGI	B-General_Concept
,	O
and	O
existential	O
risks	O
"	O
.	O
</s>
<s>
Many	O
scholars	O
who	O
are	O
concerned	O
about	O
existential	O
risk	O
advocate	O
(	O
possibly	O
massive	O
)	O
research	O
into	O
solving	O
the	O
difficult	O
"	O
control	O
problem	O
"	O
to	O
answer	O
the	O
question	O
:	O
what	O
types	O
of	O
safeguards	O
,	O
algorithms	O
,	O
or	O
architectures	O
can	O
programmers	O
implement	O
to	O
maximise	O
the	O
probability	O
that	O
their	O
recursively-improving	O
AI	B-Application
would	O
continue	O
to	O
behave	O
in	O
a	O
friendly	O
,	O
rather	O
than	O
destructive	O
,	O
manner	O
after	O
it	O
reaches	O
superintelligence	O
?	O
</s>
<s>
Solving	O
the	O
control	O
problem	O
is	O
complicated	O
by	O
the	O
AI	B-Application
arms	O
race	O
,	O
which	O
will	O
almost	O
certainly	O
see	B-Application
the	O
militarization	O
and	O
weaponization	O
of	O
AGI	B-General_Concept
by	O
more	O
than	O
one	O
nation-state	O
,	O
resulting	O
in	O
AGI-enabled	O
warfare	O
,	O
and	O
in	O
the	O
case	O
of	O
AI	B-Application
misalignment	O
,	O
AGI-directed	O
warfare	O
,	O
potentially	O
against	O
all	O
humanity	O
.	O
</s>
<s>
The	O
thesis	O
that	O
AI	B-Application
can	O
pose	O
existential	O
risk	O
also	O
has	O
detractors	O
.	O
</s>
<s>
Much	O
criticism	O
argues	O
that	O
AGI	B-General_Concept
is	O
unlikely	O
in	O
the	O
short	O
term	O
.	O
</s>
<s>
Former	O
Baidu	B-Application
Vice	O
President	O
and	O
Chief	O
Scientist	O
Andrew	O
Ng	O
says	O
worrying	O
about	O
AI	B-Application
existential	O
risk	O
is	O
"	O
like	O
worrying	O
about	O
overpopulation	O
on	O
Mars	O
when	O
we	O
have	O
not	O
even	O
set	O
foot	O
on	O
the	O
planet	O
yet.	O
"	O
</s>
