<s>
AlphaGo	B-Application
versus	I-Application
Lee	I-Application
Sedol	I-Application
,	O
also	O
known	O
as	O
the	O
Google	B-Application
DeepMind	I-Application
Challenge	I-Application
Match	I-Application
,	O
was	O
a	O
five-game	O
Go	O
match	O
between	O
top	O
Go	O
player	O
Lee	O
Sedol	O
and	O
AlphaGo	B-Application
,	O
a	O
computer	B-Application
Go	I-Application
program	O
developed	O
by	O
Google	B-Application
DeepMind	I-Application
,	O
played	O
in	O
Seoul	O
,	O
South	O
Korea	O
between	O
9	O
and	O
15	O
March	O
2016	O
.	O
</s>
<s>
AlphaGo	B-Application
won	O
all	O
but	O
the	O
fourth	O
game	O
;	O
all	O
games	O
were	O
won	O
by	O
resignation	O
.	O
</s>
<s>
The	O
match	O
has	O
been	O
compared	O
with	O
the	O
historic	O
chess	B-Application
match	O
between	O
Deep	B-General_Concept
Blue	I-General_Concept
and	O
Garry	O
Kasparov	O
in	O
1997	O
.	O
</s>
<s>
Since	O
AlphaGo	B-Application
won	O
,	O
Google	B-Application
DeepMind	I-Application
stated	O
that	O
the	O
prize	O
will	O
be	O
donated	O
to	O
charities	O
,	O
including	O
UNICEF	O
,	O
and	O
Go	O
organisations	O
.	O
</s>
<s>
After	O
the	O
match	O
,	O
The	O
Korea	O
Baduk	O
Association	O
awarded	O
AlphaGo	B-Application
the	O
highest	O
Go	O
grandmaster	O
rank	O
–	O
an	O
"	O
honorary	O
9	O
dan	O
"	O
.	O
</s>
<s>
It	O
was	O
given	O
in	O
recognition	O
of	O
AlphaGo	B-Application
's	O
"	O
sincere	O
efforts	O
"	O
to	O
master	O
Go	O
.	O
</s>
<s>
It	O
has	O
long	O
been	O
considered	O
a	O
difficult	O
challenge	O
in	O
the	O
field	O
of	O
artificial	B-Application
intelligence	I-Application
(	O
AI	B-Application
)	O
and	O
is	O
considerably	O
more	O
difficult	O
to	O
solve	O
than	O
chess	B-Application
.	O
</s>
<s>
Many	O
in	O
the	O
field	O
of	O
artificial	B-Application
intelligence	I-Application
consider	O
Go	O
to	O
require	O
more	O
elements	O
that	O
mimic	O
human	O
thought	O
than	O
chess	B-Application
.	O
</s>
<s>
Prior	O
to	O
2015	O
,	O
the	O
best	O
Go	B-Application
programs	I-Application
only	O
managed	O
to	O
reach	O
amateur	O
dan	O
level	O
.	O
</s>
<s>
Prior	O
to	O
AlphaGo	B-Application
,	O
some	O
researchers	O
had	O
claimed	O
that	O
computers	O
would	O
never	O
defeat	O
top	O
humans	O
at	O
Go	O
.	O
</s>
<s>
Elon	O
Musk	O
,	O
an	O
early	O
investor	O
of	O
Deepmind	B-Application
,	O
said	O
in	O
2016	O
that	O
experts	O
in	O
the	O
field	O
thought	O
AI	B-Application
was	O
10	O
years	O
away	O
from	O
achieving	O
a	O
victory	O
against	O
a	O
top	O
professional	O
Go	O
player	O
.	O
</s>
<s>
The	O
match	O
AlphaGo	B-Application
versus	I-Application
Lee	I-Application
Sedol	I-Application
is	O
comparable	O
to	O
the	O
1997	O
chess	B-Application
match	O
when	O
Garry	O
Kasparov	O
lost	O
to	O
IBM	O
computer	O
Deep	B-General_Concept
Blue	I-General_Concept
.	O
</s>
<s>
Kasparov	O
's	O
loss	O
to	O
Deep	B-General_Concept
Blue	I-General_Concept
is	O
considered	O
the	O
moment	O
a	O
computer	O
became	O
better	O
than	O
humans	O
at	O
chess	B-Application
.	O
</s>
<s>
AlphaGo	B-Application
is	O
significantly	O
different	O
from	O
previous	O
AI	B-Application
efforts	O
.	O
</s>
<s>
Instead	O
of	O
using	O
probability	O
algorithms	O
hard-coded	O
by	O
human	O
programmers	O
,	O
AlphaGo	B-Application
uses	O
neural	O
networks	O
to	O
estimate	O
its	O
probability	O
of	O
winning	O
.	O
</s>
<s>
AlphaGo	B-Application
accesses	O
and	O
analyses	O
the	O
entire	O
online	O
library	O
of	O
Go	O
;	O
including	O
all	O
matches	O
,	O
players	O
,	O
analytics	O
,	O
and	O
literature	O
;	O
as	O
well	O
as	O
games	O
played	O
by	O
AlphaGo	B-Application
against	O
itself	O
and	O
other	O
players	O
.	O
</s>
<s>
Once	O
setup	O
,	O
AlphaGo	B-Application
is	O
independent	O
of	O
the	O
developer	O
team	O
and	O
evaluates	O
the	O
best	O
pathway	O
to	O
solving	O
Go	O
(	O
i.e.	O
</s>
<s>
By	O
using	O
neural	O
networks	O
and	O
Monte	B-Application
Carlo	I-Application
tree	I-Application
search	I-Application
,	O
AlphaGo	B-Application
calculates	O
colossal	O
numbers	O
of	O
likely	O
and	O
unlikely	O
probabilities	O
many	O
moves	O
into	O
the	O
future	O
.	O
</s>
<s>
AlphaGo	B-Application
defeated	O
European	O
champion	O
Fan	O
Hui	O
,	O
a	O
2	O
dan	O
professional	O
,	O
5	O
–	O
0	O
in	O
October	O
2015	O
,	O
the	O
first	O
time	O
an	O
AI	B-Application
had	O
beaten	O
a	O
human	O
professional	O
player	O
at	O
the	O
game	O
on	O
a	O
full-sized	O
board	O
without	O
a	O
handicap	O
.	O
</s>
<s>
Computer	O
programs	O
Zen	O
and	O
Crazy	B-Application
Stone	I-Application
have	O
previously	O
defeated	O
human	O
players	O
ranked	O
9	O
dan	O
professional	O
with	O
handicaps	O
of	O
four	O
or	O
five	O
stones	O
.	O
</s>
<s>
Canadian	O
AI	B-Application
specialist	O
Jonathan	O
Schaeffer	O
,	O
commenting	O
after	O
the	O
win	O
against	O
Fan	O
,	O
compared	O
AlphaGo	B-Application
with	O
a	O
"	O
child	O
prodigy	O
"	O
that	O
lacked	O
experience	O
,	O
and	O
considered	O
,	O
"	O
the	O
real	O
achievement	O
will	O
be	O
when	O
the	O
program	O
plays	O
a	O
player	O
in	O
the	O
true	O
top	O
echelon.	O
"	O
</s>
<s>
Hajin	O
Lee	O
,	O
a	O
professional	O
Go	O
player	O
and	O
the	O
International	O
Go	O
Federation	O
's	O
secretary-general	O
,	O
commented	O
that	O
she	O
was	O
"	O
very	O
excited	O
"	O
at	O
the	O
prospect	O
of	O
an	O
AI	B-Application
challenging	O
Lee	O
,	O
and	O
thought	O
the	O
two	O
players	O
had	O
an	O
equal	O
chance	O
of	O
winning	O
.	O
</s>
<s>
In	O
the	O
aftermath	O
of	O
his	O
match	O
against	O
AlphaGo	B-Application
,	O
Fan	O
Hui	O
noted	O
that	O
the	O
game	O
had	O
taught	O
him	O
to	O
be	O
a	O
better	O
player	O
,	O
and	O
to	O
see	O
things	O
he	O
had	O
not	O
previously	O
seen	O
.	O
</s>
<s>
Go	O
experts	O
found	O
errors	O
in	O
AlphaGo	B-Application
's	O
play	O
against	O
Fan	O
,	O
in	O
particular	O
relating	O
to	O
a	O
lack	O
of	O
awareness	O
of	O
the	O
entire	O
board	O
.	O
</s>
<s>
AlphaGo	B-Application
's	O
original	O
training	O
dataset	O
started	O
with	O
games	O
of	O
strong	O
amateur	O
players	O
from	O
internet	O
Go	O
servers	O
,	O
after	O
which	O
AlphaGo	B-Application
trained	O
by	O
playing	O
against	O
itself	O
for	O
tens	O
of	O
millions	O
of	O
games	O
.	O
</s>
<s>
AlphaGo	B-Application
is	O
a	O
computer	O
program	O
developed	O
by	O
Google	B-Application
DeepMind	I-Application
to	O
play	O
the	O
board	O
game	O
Go	O
.	O
</s>
<s>
AlphaGo	B-Application
's	O
algorithm	O
uses	O
a	O
combination	O
of	O
machine	O
learning	O
and	O
tree	B-Algorithm
search	I-Algorithm
techniques	O
,	O
combined	O
with	O
extensive	O
training	O
,	O
both	O
from	O
human	O
and	O
computer	O
play	O
.	O
</s>
<s>
AlphaGo	B-Application
was	O
initially	O
trained	O
to	O
mimic	O
human	O
play	O
by	O
attempting	O
to	O
match	O
the	O
moves	O
of	O
expert	O
players	O
from	O
recorded	O
historical	O
games	O
,	O
using	O
a	O
KGS	B-Application
Go	I-Application
Server	I-Application
database	O
of	O
around	O
30	O
million	O
moves	O
from	O
160,000	O
games	O
by	O
KGS	O
6	O
to	O
9	O
dan	O
human	O
players	O
.	O
</s>
<s>
As	O
one	O
of	O
the	O
creators	O
of	O
AlphaGo	B-Application
explained	O
:	O
</s>
<s>
In	O
the	O
match	O
against	O
Lee	O
,	O
AlphaGo	B-Application
used	O
about	O
the	O
same	O
computing	O
power	O
as	O
it	O
had	O
in	O
the	O
match	O
against	O
Fan	O
Hui	O
,	O
where	O
it	O
used	O
1,202	O
CPUs	B-Device
and	O
176	O
GPUs	B-Architecture
.	O
</s>
<s>
The	O
Economist	O
reported	O
that	O
it	O
used	O
1,920	O
CPUs	B-Device
and	O
280	O
GPUs	B-Architecture
.	O
</s>
<s>
Google	O
has	O
also	O
stated	O
that	O
its	O
proprietary	O
tensor	B-Device
processing	I-Device
units	I-Device
were	O
used	O
in	O
the	O
match	O
against	O
Lee	O
Sedol	O
.	O
</s>
<s>
Lee	O
Sedol	O
initially	O
predicted	O
he	O
would	O
defeat	O
AlphaGo	B-Application
in	O
a	O
"	O
landslide	O
"	O
.	O
</s>
<s>
Aja	O
Huang	O
,	O
a	O
DeepMind	B-Application
team	O
member	O
and	O
amateur	O
6-dan	O
Go	O
player	O
,	O
placed	O
stones	O
on	O
the	O
Go	O
board	O
for	O
AlphaGo	B-Application
,	O
which	O
ran	O
through	O
the	O
Google	B-General_Concept
Cloud	I-General_Concept
Platform	I-General_Concept
with	O
its	O
server	O
located	O
in	O
the	O
United	O
States	O
.	O
</s>
<s>
Game	O
Date	O
Black	O
White	O
Result	O
Moves	O
1	O
9	O
March	O
2016	O
Lee	O
Sedol	O
Lee	O
Sedol	O
resigned	O
186	O
Game	O
1	O
2	O
10	O
March	O
2016	O
Lee	O
Sedol	O
Lee	O
Sedol	O
resigned	O
211	O
Game	O
2	O
3	O
12	O
March	O
2016	O
Lee	O
Sedol	O
Lee	O
Sedol	O
resigned	O
176	O
Game	O
3	O
4	O
13	O
March	O
2016	O
AlphaGo	B-Application
AlphaGo	B-Application
resigned	O
180	O
Game	O
4	O
5	O
15	O
March	O
2016	O
Lee	O
Sedol	O
Lee	O
Sedol	O
resigned	O
280	O
Game	O
5	O
Result	O
:	O
AlphaGo	B-Application
4	O
–	O
1	O
Lee	O
Sedol	O
note	O
1	O
:	O
For	O
Game	O
Five	O
,	O
under	O
the	O
official	O
rules	O
,	O
it	O
was	O
intended	O
that	O
the	O
colour	O
assignments	O
would	O
be	O
done	O
at	O
random	O
.	O
</s>
<s>
AlphaGo	B-Application
(	O
white	O
)	O
won	O
the	O
first	O
game	O
.	O
</s>
<s>
Lee	O
appeared	O
to	O
be	O
in	O
control	O
throughout	O
much	O
of	O
the	O
match	O
,	O
but	O
AlphaGo	B-Application
gained	O
the	O
advantage	O
in	O
the	O
final	O
20	O
minutes	O
and	O
Lee	O
resigned	O
.	O
</s>
<s>
Lee	O
stated	O
afterwards	O
that	O
he	O
had	O
made	O
a	O
critical	O
error	O
at	O
the	O
beginning	O
of	O
the	O
match	O
;	O
he	O
said	O
that	O
the	O
computer	O
's	O
strategy	O
in	O
the	O
early	O
part	O
of	O
the	O
game	O
was	O
"	O
excellent	O
"	O
and	O
that	O
the	O
AI	B-Application
had	O
made	O
one	O
unusual	O
move	O
that	O
no	O
human	O
Go	O
player	O
would	O
have	O
made	O
.	O
</s>
<s>
David	O
Ormerod	O
,	O
commenting	O
on	O
the	O
game	O
at	O
Go	O
Game	O
Guru	O
,	O
described	O
Lee	O
's	O
seventh	O
stone	O
as	O
"	O
a	O
strange	O
move	O
to	O
test	O
AlphaGo	B-Application
's	O
strength	O
in	O
the	O
opening	O
"	O
,	O
characterising	O
the	O
move	O
as	O
a	O
mistake	O
and	O
AlphaGo	B-Application
's	O
response	O
as	O
"	O
accurate	O
and	O
efficient	O
"	O
.	O
</s>
<s>
He	O
described	O
AlphaGo	B-Application
's	O
position	O
as	O
favourable	O
in	O
the	O
first	O
part	O
of	O
the	O
game	O
,	O
considering	O
that	O
Lee	O
started	O
to	O
come	O
back	O
with	O
move	O
81	O
,	O
before	O
making	O
"	O
questionable	O
"	O
moves	O
at	O
119	O
and	O
123	O
,	O
followed	O
by	O
a	O
"	O
losing	O
"	O
move	O
at	O
129	O
.	O
</s>
<s>
Professional	O
Go	O
player	O
Cho	O
Hanseung	O
commented	O
that	O
AlphaGo	B-Application
's	O
game	O
had	O
greatly	O
improved	O
from	O
when	O
it	O
beat	O
Fan	O
Hui	O
in	O
October	O
2015	O
.	O
</s>
<s>
According	O
to	O
9-dan	O
Go	O
grandmaster	O
Kim	O
Seong-ryong	O
,	O
Lee	O
seemed	O
stunned	O
by	O
AlphaGo	B-Application
's	O
strong	O
play	O
on	O
the	O
102nd	O
stone	O
.	O
</s>
<s>
After	O
watching	O
AlphaGo	B-Application
make	O
the	O
game	O
's	O
102nd	O
move	O
,	O
Lee	O
mulled	O
over	O
his	O
options	O
for	O
more	O
than	O
10	O
minutes	O
.	O
</s>
<s>
AlphaGo	B-Application
(	O
black	O
)	O
won	O
the	O
second	O
game	O
.	O
</s>
<s>
Lee	O
stated	O
afterwards	O
that	O
"	O
AlphaGo	B-Application
played	O
a	O
nearly	O
perfect	O
game	O
"	O
,	O
"	O
from	O
very	O
beginning	O
of	O
the	O
game	O
I	O
did	O
not	O
feel	O
like	O
there	O
was	O
a	O
point	O
that	O
I	O
was	O
leading	O
"	O
.	O
</s>
<s>
One	O
of	O
the	O
creators	O
of	O
AlphaGo	B-Application
,	O
Demis	O
Hassabis	O
,	O
said	O
that	O
the	O
system	O
was	O
confident	O
of	O
victory	O
from	O
the	O
midway	O
point	O
of	O
the	O
game	O
,	O
even	O
though	O
the	O
professional	O
commentators	O
could	O
not	O
tell	O
which	O
player	O
was	O
ahead	O
.	O
</s>
<s>
Michael	O
Redmond	O
(	O
9p	O
)	O
noted	O
that	O
AlphaGo	B-Application
's	O
19th	O
stone	O
(	O
move	O
37	O
)	O
was	O
"	O
creative	O
"	O
and	O
"	O
unique	O
"	O
.	O
</s>
<s>
An	O
Younggil	O
(	O
8p	O
)	O
called	O
AlphaGo	B-Application
's	O
move	O
37	O
"	O
a	O
rare	O
and	O
intriguing	O
shoulder	O
hit	O
"	O
but	O
said	O
Lee	O
's	O
counter	O
was	O
"	O
exquisite	O
"	O
.	O
</s>
<s>
He	O
stated	O
that	O
control	O
passed	O
between	O
the	O
players	O
several	O
times	O
before	O
the	O
endgame	O
,	O
and	O
especially	O
praised	O
AlphaGo	B-Application
's	O
moves	O
151	O
,	O
157	O
,	O
and	O
159	O
,	O
calling	O
them	O
"	O
brilliant	O
"	O
.	O
</s>
<s>
AlphaGo	B-Application
showed	O
anomalies	O
and	O
moves	O
from	O
a	O
broader	O
perspective	O
which	O
professional	O
Go	O
players	O
described	O
as	O
looking	O
like	O
mistakes	O
at	O
the	O
first	O
sight	O
but	O
an	O
intentional	O
strategy	O
in	O
hindsight	O
.	O
</s>
<s>
As	O
one	O
of	O
the	O
creators	O
of	O
the	O
system	O
explained	O
,	O
AlphaGo	B-Application
does	O
not	O
attempt	O
to	O
maximize	O
its	O
points	O
or	O
its	O
margin	O
of	O
victory	O
,	O
but	O
tries	O
to	O
maximize	O
its	O
probability	O
of	O
winning	O
.	O
</s>
<s>
If	O
AlphaGo	B-Application
must	O
choose	O
between	O
a	O
scenario	O
where	O
it	O
will	O
win	O
by	O
20	O
points	O
with	O
80	O
percent	O
probability	O
and	O
another	O
where	O
it	O
will	O
win	O
by	O
1	O
and	O
a	O
half	O
points	O
with	O
99	O
percent	O
probability	O
,	O
it	O
will	O
choose	O
the	O
latter	O
,	O
even	O
if	O
it	O
must	O
give	O
up	O
points	O
to	O
achieve	O
it	O
.	O
</s>
<s>
In	O
particular	O
,	O
move	O
167	O
by	O
AlphaGo	B-Application
seemed	O
to	O
give	O
Lee	O
a	O
fighting	O
chance	O
and	O
was	O
declared	O
to	O
look	O
like	O
an	O
obvious	O
mistake	O
by	O
commentators	O
.	O
</s>
<s>
An	O
Younggil	O
stated	O
"	O
So	O
when	O
AlphaGo	B-Application
plays	O
a	O
slack	O
looking	O
move	O
,	O
we	O
may	O
regard	O
it	O
as	O
a	O
mistake	O
,	O
but	O
perhaps	O
it	O
should	O
more	O
accurately	O
be	O
viewed	O
as	O
a	O
declaration	O
of	O
victory	O
?	O
"	O
</s>
<s>
AlphaGo	B-Application
(	O
white	O
)	O
won	O
the	O
third	O
game	O
.	O
</s>
<s>
After	O
the	O
second	O
game	O
,	O
there	O
had	O
still	O
been	O
strong	O
doubts	O
among	O
players	O
whether	O
AlphaGo	B-Application
was	O
truly	O
a	O
strong	O
player	O
in	O
the	O
sense	O
that	O
a	O
human	O
might	O
be	O
.	O
</s>
<s>
According	O
to	O
An	O
Younggil	O
(	O
8p	O
)	O
and	O
David	O
Ormerod	O
,	O
the	O
game	O
showed	O
that	O
"	O
AlphaGo	B-Application
is	O
simply	O
stronger	O
than	O
any	O
known	O
human	O
Go	O
player.	O
"	O
</s>
<s>
AlphaGo	B-Application
was	O
seen	O
to	O
capably	O
navigate	O
tricky	O
situations	O
known	O
as	O
ko	O
that	O
did	O
not	O
come	O
up	O
in	O
the	O
previous	O
two	O
matches	O
.	O
</s>
<s>
An	O
and	O
Ormerod	O
consider	O
move	O
148	O
to	O
be	O
particularly	O
notable	O
:	O
in	O
the	O
middle	O
of	O
a	O
complex	O
ko	O
fight	O
,	O
AlphaGo	B-Application
displayed	O
sufficient	O
"	O
confidence	O
"	O
that	O
it	O
was	O
winning	O
the	O
fight	O
to	O
play	O
a	O
large	O
move	O
elsewhere	O
.	O
</s>
<s>
Lee	O
,	O
playing	O
black	O
,	O
opened	O
with	O
a	O
High	O
Chinese	O
formation	O
and	O
generated	O
a	O
large	O
area	O
of	O
black	O
influence	O
,	O
which	O
AlphaGo	B-Application
invaded	O
at	O
move	O
12	O
.	O
</s>
<s>
AlphaGo	B-Application
had	O
gained	O
control	O
of	O
the	O
game	O
by	O
move	O
48	O
,	O
and	O
forced	O
Lee	O
onto	O
the	O
defensive	O
.	O
</s>
<s>
Lee	O
counterattacked	O
at	O
moves	O
77/79	O
,	O
but	O
AlphaGo	B-Application
's	O
response	O
was	O
effective	O
and	O
its	O
move	O
90	O
succeeded	O
in	O
simplifying	O
the	O
position	O
.	O
</s>
<s>
Lee	O
attacked	O
again	O
at	O
moves	O
115	O
and	O
125	O
,	O
but	O
AlphaGo	B-Application
's	O
responses	O
were	O
again	O
effective	O
.	O
</s>
<s>
Lee	O
chose	O
to	O
play	O
a	O
type	O
of	O
extreme	O
strategy	O
,	O
known	O
as	O
amashi	O
,	O
in	O
response	O
to	O
AlphaGo	B-Application
's	O
apparent	O
preference	O
for	O
Souba	O
Go	O
(	O
attempting	O
to	O
win	O
by	O
many	O
small	O
gains	O
when	O
the	O
opportunity	O
arises	O
)	O
,	O
taking	O
territory	O
at	O
the	O
perimeter	O
rather	O
than	O
the	O
center	O
.	O
</s>
<s>
By	O
doing	O
so	O
,	O
his	O
apparent	O
aim	O
was	O
to	O
force	O
an	O
"	O
all	O
or	O
nothing	O
"	O
style	O
of	O
situation	O
–	O
a	O
possible	O
weakness	O
for	O
an	O
opponent	O
strong	O
at	O
negotiation	O
types	O
of	O
play	O
,	O
and	O
one	O
which	O
might	O
make	O
AlphaGo	B-Application
's	O
capability	O
of	O
deciding	O
slim	O
advantages	O
largely	O
irrelevant	O
.	O
</s>
<s>
In	O
the	O
early	O
game	O
,	O
Lee	O
concentrated	O
on	O
taking	O
territory	O
in	O
the	O
edges	O
and	O
corners	O
of	O
the	O
board	O
,	O
allowing	O
AlphaGo	B-Application
to	O
gain	O
influence	O
in	O
the	O
top	O
and	O
centre	O
.	O
</s>
<s>
Lee	O
then	O
invaded	O
AlphaGo	B-Application
's	O
region	O
of	O
influence	O
at	O
the	O
top	O
with	O
moves	O
40	O
to	O
48	O
,	O
following	O
the	O
amashi	O
strategy	O
.	O
</s>
<s>
AlphaGo	B-Application
responded	O
with	O
a	O
shoulder	O
hit	O
at	O
move	O
47	O
,	O
subsequently	O
sacrificing	O
four	O
stones	O
elsewhere	O
,	O
and	O
gaining	O
the	O
initiative	O
with	O
moves	O
47	O
to	O
53	O
and	O
69	O
.	O
</s>
<s>
Lee	O
tested	O
AlphaGo	B-Application
with	O
moves	O
72	O
to	O
76	O
without	O
provoking	O
an	O
error	O
,	O
and	O
by	O
this	O
point	O
in	O
the	O
game	O
commentators	O
had	O
begun	O
to	O
feel	O
Lee	O
's	O
play	O
was	O
a	O
lost	O
cause	O
.	O
</s>
<s>
AlphaGo	B-Application
responded	O
poorly	O
on	O
move	O
79	O
,	O
at	O
which	O
time	O
it	O
estimated	O
it	O
had	O
a	O
70%	O
chance	O
to	O
win	O
the	O
game	O
.	O
</s>
<s>
AlphaGo	B-Application
's	O
initial	O
response	O
in	O
moves	O
83	O
to	O
85	O
was	O
appropriate	O
,	O
but	O
at	O
move	O
87	O
,	O
its	O
estimate	O
of	O
its	O
chances	O
to	O
win	O
suddenly	O
plummeted	O
,	O
provoking	O
it	O
to	O
make	O
a	O
series	O
of	O
very	O
bad	O
moves	O
from	O
black	O
87	O
to	O
101	O
.	O
</s>
<s>
Despite	O
good	O
tactics	O
during	O
moves	O
131	O
to	O
141	O
,	O
AlphaGo	B-Application
proved	O
unable	O
to	O
recover	O
during	O
the	O
endgame	O
and	O
resigned	O
.	O
</s>
<s>
AlphaGo	B-Application
's	O
resignation	O
was	O
triggered	O
when	O
it	O
evaluated	O
its	O
chance	O
of	O
winning	O
to	O
be	O
less	O
than	O
20%	O
;	O
this	O
is	O
intended	O
to	O
match	O
the	O
decision	O
of	O
professionals	O
who	O
resign	O
rather	O
than	O
play	O
to	O
the	O
end	O
when	O
their	O
position	O
is	O
felt	O
to	O
be	O
irrecoverable	O
.	O
</s>
<s>
Lee	O
commented	O
after	O
the	O
match	O
that	O
he	O
considered	O
AlphaGo	B-Application
was	O
strongest	O
when	O
playing	O
white	O
(	O
second	O
)	O
.	O
</s>
<s>
David	O
Ormerod	O
of	O
Go	O
Game	O
Guru	O
stated	O
that	O
although	O
an	O
analysis	O
of	O
AlphaGo	B-Application
's	O
play	O
around	O
79	O
–	O
87	O
was	O
not	O
yet	O
available	O
,	O
he	O
believed	O
it	O
was	O
a	O
result	O
of	O
a	O
known	O
weakness	O
in	O
play	O
algorithms	O
which	O
use	O
Monte	B-Application
Carlo	I-Application
tree	I-Application
search	I-Application
.	O
</s>
<s>
AlphaGo	B-Application
(	O
white	O
)	O
won	O
the	O
fifth	O
game	O
.	O
</s>
<s>
Lee	O
,	O
playing	O
black	O
,	O
opened	O
in	O
a	O
similar	O
fashion	O
to	O
the	O
first	O
game	O
and	O
then	O
began	O
to	O
stake	O
out	O
territory	O
in	O
the	O
right	O
and	O
top	O
left	O
corners	O
–	O
a	O
similar	O
strategy	O
to	O
the	O
one	O
he	O
employed	O
successfully	O
in	O
game	O
4	O
–	O
while	O
AlphaGo	B-Application
gained	O
influence	O
in	O
the	O
centre	O
of	O
the	O
board	O
.	O
</s>
<s>
The	O
game	O
remained	O
even	O
until	O
white	O
moves	O
48	O
to	O
58	O
,	O
which	O
AlphaGo	B-Application
played	O
in	O
the	O
bottom	O
right	O
.	O
</s>
<s>
Michael	O
Redmond	O
(	O
9p	O
)	O
speculated	O
that	O
perhaps	O
AlphaGo	B-Application
had	O
missed	O
black	O
's	O
"	O
tombstone	O
squeeze	O
"	O
tesuji	O
.	O
</s>
<s>
AlphaGo	B-Application
then	O
started	O
to	O
develop	O
the	O
top	O
of	O
the	O
board	O
as	O
well	O
as	O
the	O
centre	O
,	O
and	O
defended	O
successfully	O
against	O
an	O
attack	O
by	O
Lee	O
in	O
moves	O
69	O
to	O
81	O
that	O
David	O
Ormerod	O
characterised	O
as	O
over-cautious	O
.	O
</s>
<s>
By	O
white	O
90	O
,	O
AlphaGo	B-Application
had	O
regained	O
equality	O
,	O
and	O
then	O
played	O
a	O
series	O
of	O
moves	O
described	O
by	O
Ormerod	O
as	O
"	O
unusual	O
...	O
but	O
subtly	O
impressive	O
"	O
which	O
gained	O
a	O
small	O
advantage	O
.	O
</s>
<s>
Lee	O
tried	O
a	O
Hail	O
Mary	O
pass	O
with	O
moves	O
167	O
and	O
169	O
but	O
AlphaGo	B-Application
's	O
defence	O
was	O
successful	O
.	O
</s>
<s>
AlphaGo	B-Application
's	O
victory	O
was	O
a	O
major	O
milestone	O
in	O
artificial	B-Application
intelligence	I-Application
research	I-Application
.	O
</s>
<s>
Most	O
experts	O
thought	O
a	O
Go	O
program	O
as	O
powerful	O
as	O
AlphaGo	B-Application
was	O
at	O
least	O
five	O
years	O
away	O
;	O
some	O
experts	O
thought	O
that	O
it	O
would	O
take	O
at	O
least	O
another	O
decade	O
before	O
computers	O
would	O
beat	O
Go	O
champions	O
.	O
</s>
<s>
Most	O
observers	O
at	O
the	O
beginning	O
of	O
the	O
2016	O
matches	O
expected	O
Lee	O
to	O
beat	O
AlphaGo	B-Application
.	O
</s>
<s>
With	O
games	O
such	O
as	O
checkers	O
,	O
chess	B-Application
,	O
and	O
now	O
Go	O
won	O
by	O
computer	O
players	O
,	O
victories	O
at	O
popular	O
board	O
games	O
can	O
no	O
longer	O
serve	O
as	O
major	O
milestones	O
for	O
artificial	B-Application
intelligence	I-Application
in	O
the	O
way	O
that	O
they	O
used	O
to	O
.	O
</s>
<s>
Deep	B-General_Concept
Blue	I-General_Concept
's	O
Murray	O
Campbell	O
called	O
AlphaGo	B-Application
's	O
victory	O
"	O
the	O
end	O
of	O
an	O
era	O
...	O
board	O
games	O
are	O
more	O
or	O
less	O
done	O
and	O
it	O
's	O
time	O
to	O
move	O
on.	O
"	O
</s>
<s>
When	O
compared	O
with	O
Deep	B-General_Concept
Blue	I-General_Concept
or	O
with	O
Watson	B-Application
,	O
AlphaGo	B-Application
's	O
underlying	O
algorithms	O
are	O
potentially	O
more	O
general-purpose	O
,	O
and	O
may	O
be	O
evidence	O
that	O
the	O
scientific	O
community	O
is	O
making	O
progress	O
toward	O
artificial	O
general	O
intelligence	O
.	O
</s>
<s>
Some	O
commentators	O
believe	O
AlphaGo	B-Application
's	O
victory	O
makes	O
for	O
a	O
good	O
opportunity	O
for	O
society	O
to	O
start	O
discussing	O
preparations	O
for	O
the	O
possible	O
future	O
impact	O
of	O
machines	O
with	O
general	O
purpose	O
intelligence	O
.	O
</s>
<s>
In	O
March	O
2016	O
,	O
AI	B-Application
researcher	O
Stuart	O
Russell	O
stated	O
that	O
"	O
AI	B-Application
methods	O
are	O
progressing	O
much	O
faster	O
than	O
expected	O
,	O
(	O
which	O
)	O
makes	O
the	O
question	O
of	O
the	O
long-term	O
outcome	O
more	O
urgent	O
,	O
"	O
adding	O
that	O
"	O
in	O
order	O
to	O
ensure	O
that	O
increasingly	O
powerful	O
AI	B-Application
systems	O
remain	O
completely	O
under	O
human	O
control	O
...	O
there	O
is	O
a	O
lot	O
of	O
work	O
to	O
do.	O
"	O
</s>
<s>
Some	O
scholars	O
,	O
such	O
as	O
physicist	O
Stephen	O
Hawking	O
,	O
warn	O
that	O
some	O
future	O
self-improving	O
AI	B-Application
could	O
gain	O
actual	O
general	O
intelligence	O
,	O
leading	O
to	O
an	O
unexpected	O
AI	B-Application
takeover	O
;	O
other	O
scholars	O
disagree	O
:	O
AI	B-Application
expert	O
Jean-Gabriel	O
Ganascia	O
believes	O
that	O
"	O
Things	O
like	O
'	O
common	O
sense	O
 '	O
...	O
may	O
never	O
be	O
reproducible	O
"	O
,	O
and	O
says	O
"	O
I	O
do	O
n't	O
see	O
why	O
we	O
would	O
speak	O
about	O
fears	O
.	O
</s>
<s>
The	O
DeepMind	B-Application
AlphaGo	I-Application
Team	O
received	O
the	O
Inaugural	O
IJCAI	O
Marvin	O
Minsky	O
Medal	O
for	O
Outstanding	O
Achievements	O
in	O
AI	B-Application
.	O
</s>
<s>
"	O
AlphaGo	B-Application
is	O
a	O
wonderful	O
achievement	O
,	O
and	O
a	O
perfect	O
example	O
of	O
what	O
the	O
Minsky	O
Medal	O
was	O
initiated	O
to	O
recognise	O
"	O
,	O
said	O
Professor	O
Michael	O
Wooldridge	O
,	O
Chair	O
of	O
the	O
IJCAI	O
Awards	O
Committee	O
.	O
</s>
<s>
"	O
What	O
particularly	O
impressed	O
IJCAI	O
was	O
that	O
AlphaGo	B-Application
achieves	O
what	O
it	O
does	O
through	O
a	O
brilliant	O
combination	O
of	O
classic	O
AI	B-Application
techniques	O
as	O
well	O
as	O
the	O
state-of-the-art	O
machine	O
learning	O
techniques	O
that	O
DeepMind	B-Application
is	O
so	O
closely	O
associated	O
with	O
.	O
</s>
<s>
It	O
’s	O
a	O
breathtaking	O
demonstration	O
of	O
contemporary	O
AI	B-Application
,	O
and	O
we	O
are	O
delighted	O
to	O
be	O
able	O
to	O
recognise	O
it	O
with	O
this	O
award	O
"	O
.	O
</s>
<s>
Many	O
top	O
Go	O
players	O
characterized	O
AlphaGo	B-Application
's	O
unorthodox	O
plays	O
as	O
seemingly-questionable	O
moves	O
that	O
initially	O
befuddled	O
onlookers	O
,	O
but	O
made	O
sense	O
in	O
hindsight	O
:	O
"	O
All	O
but	O
the	O
very	O
best	O
Go	O
players	O
craft	O
their	O
style	O
by	O
imitating	O
top	O
players	O
.	O
</s>
<s>
AlphaGo	B-Application
seems	O
to	O
have	O
totally	O
original	O
moves	O
it	O
creates	O
itself.	O
"	O
</s>
<s>
AlphaGo	B-Application
appeared	O
to	O
have	O
unexpectedly	O
become	O
much	O
stronger	O
,	O
even	O
when	O
compared	O
with	O
its	O
October	O
2015	O
match	O
against	O
Fan	O
Hui	O
where	O
a	O
computer	O
had	O
beaten	O
a	O
Go	O
professional	O
for	O
the	O
first	O
time	O
ever	O
without	O
the	O
advantage	O
of	O
a	O
handicap	O
.	O
</s>
<s>
China	O
's	O
number	O
one	O
player	O
,	O
Ke	O
Jie	O
,	O
who	O
was	O
at	O
the	O
time	O
the	O
top-ranked	O
player	O
worldwide	O
,	O
initially	O
claimed	O
that	O
he	O
would	O
be	O
able	O
to	O
beat	O
AlphaGo	B-Application
,	O
but	O
declined	O
to	O
play	O
against	O
it	O
for	O
fear	O
that	O
it	O
would	O
"	O
copy	O
my	O
style	O
"	O
.	O
</s>
<s>
Toby	O
Manning	O
,	O
the	O
referee	O
of	O
AlphaGo	B-Application
's	O
match	O
against	O
Fan	O
Hui	O
,	O
and	O
Hajin	O
Lee	O
,	O
secretary	O
general	O
of	O
the	O
International	O
Go	O
Federation	O
,	O
both	O
reason	O
that	O
in	O
the	O
future	O
,	O
Go	O
players	O
will	O
get	O
help	O
from	O
computers	O
to	O
learn	O
what	O
they	O
have	O
done	O
wrong	O
in	O
games	O
and	O
improve	O
their	O
skills	O
.	O
</s>
<s>
Lee	O
apologized	O
for	O
his	O
losses	O
,	O
stating	O
after	O
game	O
three	O
that	O
"	O
I	O
misjudged	O
the	O
capabilities	O
of	O
AlphaGo	B-Application
and	O
felt	O
powerless.	O
"	O
</s>
<s>
In	O
response	O
to	O
the	O
match	O
the	O
South	O
Korean	O
government	O
announced	O
on	O
17	O
March	O
2016	O
that	O
it	O
would	O
invest	O
$863	O
million	O
(	O
1	O
trillion	O
won	O
)	O
in	O
artificial-intelligence	B-Application
(	O
AI	B-Application
)	O
research	O
over	O
the	O
next	O
five	O
years	O
.	O
</s>
<s>
An	O
award-winning	O
documentary	O
film	O
about	O
the	O
matches	O
,	O
AlphaGo	B-Application
,	O
was	O
made	O
in	O
2017	O
.	O
</s>
<s>
On	O
13	O
March	O
2020	O
,	O
the	O
film	O
was	O
made	O
free	O
online	O
on	O
the	O
DeepMind	B-Application
YouTube	B-General_Concept
channel	I-General_Concept
.	O
</s>
