COMPASSION

Affirmation of life is the spiritual act by which man ceases to live thoughtlessly and begins to devote himself to his life
with reverence in order to give it true value.
— Albert Schweitzer

11/24/2017

Garry Kasparov and DeepMind’s CEO Demis Hassabis @ Google




Garry Kasparov and DeepMind’s CEO Demis Hassabis discuss Garry’s new book “Deep Thinking”, his match with Deep Blue and his thoughts on the future of AI in the world of chess.
Get the book here: https://goo.gl/OwuOcW
Event moderated by Demis Hassabis, CEO, DeepMind. 
** About the book, Deep Thinking: Where Machine Intelligence Ends and Human Creativity Begins ** 
In May 1997, the world watched as Garry Kasparov, the greatest chess player in the world, was defeated for the first time by the IBM supercomputer Deep Blue.
It was a watershed moment in the history of technology: machine intelligence had arrived at the point where it could best human intellect. It wasn't a coincidence that Kasparov became the symbol of man's fight against the machines.
Chess has long been the fulcrum in development of machine intelligence; the hoax automaton 'The Turk' in the 18th century and Alan Turing's first chess program in 1952 were two early examples of the quest for machines to think like humans -- a talent we measured by their ability to beat their creators at chess.
As the pre-eminent chessmaster of the 80s and 90s, it was Kasparov's blessing and his curse to play against each generation's strongest computer champions, contributing to their development and advancing the field.

Like all passionate competitors, Kasparov has taken his defeat and learned from it.
He has devoted much energy to devising ways in which humans can partner with machines in order to produce results better than either can achieve alone.
During the twenty years since playing Deep Blue, he's played both with and against machines, learning a great deal about our vital relationship with our most remarkable creations.
Ultimately, he's become convinced that by embracing the competition between human and machine intelligence, we can spend less time worrying about being replaced and more thinking of new challenges to conquer.
In this breakthrough book, Kasparov tells his side of the story of Deep Blue for the first time -- what it was like to strategize against an implacable, untiring opponent -- the mistakes he made and the reasons the odds were against him.
But more than that, he tells his story of AI more generally, and how he's evolved to embrace it, taking part in an urgent debate with philosophers worried about human values, programmers creating self-learning neural networks, and engineers of cutting edge robotics.






00:00
[Music]
00:09
well thank you everyone for coming
00:11
it's a really special privilege and
00:14
honor for me actually to talk to Gary in
00:17
my opinion my humble opinion the
00:19
greatest chess player of all time and
00:22
you know I've really enjoyed his book
00:26
which I reviewed recently and you know
00:29
I've is impressed with Gary's
00:31
understanding of artificial intelligence
00:32
and the latest advances in that so you
00:35
know it's gonna be great to talk about
00:36
that as well as chess today so welcome
00:38
thank you all matters or we might call
00:40
your review and over the album
00:42
protection against all the tech guys
00:43
that was supposed to be for not being an
00:47
expert I'm glad as I can be of use so
00:49
before we get talking about you know the
00:53
blue match and I'm sure everyone's going
00:55
to want to hear about your insights on
00:56
that and also machine learning more
00:58
generally and I wanted to begin by
00:59
asking you you know about growing up as
01:02
a chess champion in Soviet Union did you
01:04
always want to be a chess player world
01:07
champion Oh Chet did you consider
01:08
anything else or were you from a very
01:11
young age decide that this was your path
01:13
I learned how to play chess when I was
01:18
five or six sorry I couldn't give you an
01:22
exact moment nobody was there to tweet
01:23
about it this was late sixty-eight
01:27
neighbourly 69 you're watching my
01:29
parents trying to solve chess problem
01:31
and them I loved the game at first sight
01:33
and then ever since and I'm still in or
01:36
was the game and there I could feel that
01:40
was the match made in heaven
01:42
and everybody around also could see that
01:45
chess was a perfect fit for for my mind
01:48
skills yeah
01:49
and actually you talk about in the book
01:50
about chess informing all of your
01:53
thinking and the rest of your life I do
01:54
what would you mean by that what skills
01:56
can you see yourself using in the rest
01:58
of your life that you know yes naturally
02:02
if you are engaged in a competitive
02:04
sport at all the age you
02:09
you see many things just a reflection of
02:14
your chess games your engagement because
02:18
you have to play up to win it means you
02:20
have to change certain um certain habits
02:23
and customs and what was important for
02:28
me that's what I learned from my mother
02:29
is that my game was not just about
02:32
winning it was also about making a
02:34
difference and that's what helped me to
02:36
make a transition later on in my life
02:38
from playing chess being gone number one
02:41
chess player for 20 years to other
02:44
things that I'm doing now not pretending
02:46
that I could be number one and repeat my
02:47
outstanding achievements in the game of
02:49
chess but still recognizing that I'm
02:51
quite useful because I'm trying to bring
02:57
my chess experience my what I learned
02:59
from the game of chess my analytical
03:02
skills to make the difference elsewhere
03:04
so maybe I will wish it we should talk
03:06
about obviously the heart of your book
03:08
which is that the deeply batch and you
03:11
know I was fascinated to see or your
03:13
take on you know having going through
03:15
the alphago matches ourselves on the
03:16
other side you'll take on it from from
03:18
the player's perspective it's an
03:21
interesting story because when we played
03:22
our first match I always want to remind
03:24
people of the two matches and I won the
03:26
first one exactly Richard Richard
03:28
Meckler very clear so directly check now
03:34
the first match in Philadelphia
03:37
it just it was organized as quite a
03:39
low-level event the corporation was not
03:42
even involved it was a cm behind it and
03:45
the team always wanted to challenge me
03:48
and I had an experience of playing with
03:51
our played against them in 1989 when
03:53
they had a deep stock are the the
03:55
prototype from Carnegie Mellon that they
03:58
brought to IBM is turned into a new
04:01
project and everything has changed at
04:05
the game one which by the way if we are
04:07
talking about a watershed moment that
04:09
was in February 1996 in Philadelphia
04:11
when I lost Game one of that match
04:13
because the rest is you may argue matter
04:15
of technique matter of time it was like
04:17
signing on the wall if the machine could
04:20
be the world champion one game then
04:21
eventually eventually here I fought back
04:24
I won the match I won Game two and Game
04:26
five and Game six but it was pretty
04:28
clear that the rest would be matter of
04:30
time and the first game had some kind of
04:34
records up in following on internet I
04:37
think they did the numbers they had they
04:39
would hide the numbers later on in
04:41
Atlanta that IBM run around the website
04:46
there and some way the corporation lugar
04:50
already discovered the huge potential at
04:52
the at a relatively low cost so and
04:56
while the rest of the match was still in
04:58
our player it's just what people chess
05:01
then it turned into a big corporate in
05:03
devil and them and I look it's what
05:07
under the bridge
05:08
20 years ago I love the match but I
05:11
think and I've discovered in book I made
05:13
many mistakes in preparation and one of
05:15
the biggest mistake and that's quite you
05:17
know that's why I was so upset with
05:19
myself is that I didn't reach the speed
05:22
I be m.d blue as just new opponent the
05:26
way I created Anatoly Karpov or vision
05:28
and or not referred for me it was a
05:30
steal and it I was still part of a great
05:33
social and scientific experiment at the
05:36
end of the 20th century something that
05:38
could help us to understand more about
05:39
how we humans make decisions how
05:42
machines can play with us
05:43
it's it was not just you know winning or
05:46
losing big mistake
05:47
yeah now for I be able to exact about
05:50
winning or losing care yeah and the one
05:53
of one of the big mistakes that I made
05:55
as just as I while signing the contract
05:58
you always have to read the fine print
06:00
yeah and every day reacts anything and
06:03
because it's when people ask me what IBM
06:06
cheated no they just bend the rules in
06:09
their favor at states they followed the
06:11
the letter but of the spirit of the
06:13
agreement and for instance one of the
06:15
big issues after the match after our
06:18
first match in Philadelphia for me was
06:19
how can I prepare if I didn't have any
06:22
games that this is the normal way to
06:24
prepare you look at the games of your
06:25
opponent and deep-blue in Philadelphia
06:27
was the black box I had no idea what it
06:29
was capable of there were so few games
06:32
that machine played in other people get
06:35
other computer
06:35
but it was not it was not the machine
06:38
that that saves me now one here more
06:41
than one year 15 months between the
06:43
first match and the second match and and
06:45
I was under impression that it's it's it
06:47
will be altitude is fairly because after
06:50
my first victory in Philadelphia I went
06:52
to Yorktown Heights I said with the team
06:53
they had this you know similar
06:55
arrangements for all ID lateraled well
06:57
so the atmosphere was very friendly yeah
06:59
uh-huh a change and and then we just
07:03
it's the when I eat though just expected
07:06
them games to be provided and then
07:08
there's a wait a second but can you read
07:09
the games played in official tournaments
07:12
and of course the blue hasn't played a
07:14
single game out of the lab right so
07:16
which means that in main 98 in May 1997
07:20
I face the face another black box that I
07:22
knew was much stronger than its it was
07:26
before but still no idea what what to
07:28
expect I knew they had a professional
07:30
team so then they made a massive
07:31
preparation and I have to admit my
07:34
probation was quite lousy because again
07:36
I did only just before the match a week
07:38
before the match a realize you know
07:40
that's how difficult the challenge could
07:42
be but also the one of the one of the
07:45
key elements of the of of this contract
07:49
that I told overlooked was about
07:51
machines rebooting that's a big issue
07:54
yeah you understand what it is it's here
07:56
I don't have to explain but general
07:58
audience doesn't understand that the
07:59
moment you will be the computer you will
08:00
never be able to reconstruct the game so
08:02
which means that is the whole idea that
08:03
it matches fair and square and and and
08:07
you can always go back and to see why
08:08
the blue made this move or that though
08:11
it's all and also there may be a baby
08:14
children realize that I didn't realize
08:15
it I read the book that in fact several
08:17
times there will immediately crashed and
08:19
anyway we don't know the enemy and then
08:22
they came up with a different move but I
08:23
in a couple of it's a source it doesn't
08:25
matter what the means was is just the
08:27
crash you know it's if you play the
08:29
match anything buddy-buddy problems was
08:31
agreed crush you all to get heart attack
08:34
sorry go to Syria straight a lot yeah
08:40
nobody see where had Ken Thompson you
08:43
know that's the it's great to games or
08:45
expert just you know just who was there
08:47
in Philadelphia helping me he was also
08:48
from
08:49
screen but on the screen you could see
08:51
the D blues communication back to the
08:53
programmer but you didn't see whether
08:55
they said anything the other way is the
08:58
other way again I don't know but it's
09:00
it's definitely created a lot of tension
09:02
in the match and and we now after losing
09:05
game game to its essence another story
09:07
of its own so I I was very upset and I
09:10
demanded log and by the way if they
09:13
wanted to play a fair a fair game all
09:16
they had to do is to produce the blog
09:18
the logs and to prove that my suspicions
09:21
were not you know founded they didn't do
09:23
it
09:23
yeah they just wanted me just to to to
09:25
be inflamed because they realized that
09:27
while the boom was not that strong at
09:29
that time I still think I was stronger
09:31
yeah you would around 20 years later you
09:33
can look at the games and you can take a
09:36
chess engine on your laptop and you'll
09:38
find out that many mistakes are made for
09:39
both sides
09:40
I mean one of the one of the most
09:42
amazing that would look in game cubed
09:43
game 5 the endgame region endgame I was
09:46
slightly better and everybody at that
09:48
time in 99 yourself believe that was a
09:50
brilliant escape by deep blue now in 30
09:52
seconds to one minute depends on the
09:54
strength of you of the speed of your
09:56
computer chess engines like stock fish
10:00
or comedy will tell you that the end
10:02
game was draw the blue made a bad
10:04
mistake and then I missed a win which no
10:06
would so in 1997
10:08
it's including including the blue so
10:10
that tells you that these are all the
10:11
strengths and I think that if we play
10:13
that the should match the rubber match I
10:15
slide against winning again it wouldn't
10:18
change this or the long-term outcome but
10:20
how long do you think you could have
10:21
held them off for a full potential maybe
10:24
two three years that's what I mean
10:25
people would make some to three angles I
10:26
played two more matches with with a deep
10:29
prison and give junior in m73 both ended
10:33
in the draw and so that was a balance in
10:35
the next five years but in 1997 they
10:37
realized that if putting pressure on
10:39
only one human player in a match they
10:42
could they could achieve the result if
10:43
you cannot make your play stronger you
10:45
can definitely no in methylating the
10:48
other player and took him of the bow so
10:51
so what sort of moving more to the
10:53
present day now you know how do you do
10:56
you think how chess computers change
10:58
chess do you think it's for the better
10:59
it's just different what what do you
11:02
think
11:02
that inhalation it's the this something
11:05
that you said that you know it just is
11:06
quite striking because you said is it
11:08
for better or worse it's happening you
11:11
hear it it says the technology is no
11:13
neither good nor bad it's agnostic you
11:15
know you can do many great things with
11:17
your mobile phone but you can also
11:19
create a terrorist network so it says
11:20
it's it's happening and we just have to
11:23
adjust and as for the game of chess it
11:26
different because young generation of
11:28
chess players they learn very
11:30
differently from us I remember I had
11:32
books and not so many books you can use
11:34
new books you can you can buy in the
11:36
civil union every book was cherished and
11:39
I had my my notebooks so this is and I
11:42
when I when I when I went to the top and
11:44
played World Championship matches I had
11:46
also notebooks and just all my recording
11:49
my analysis and that I treasured them I
11:52
remembered as you know I had a couple of
11:53
couple of quite big notebooks with
11:57
analysis and they were already know this
11:58
top secret and I believed in 8th 1985 in
12:02
1986 1987 that was real pressure that
12:05
was a powerful weapon it's like you know
12:06
magic Shore of Maryland now when you
12:09
look at this allows with computer you
12:10
under settled broken knife but also when
12:15
you look at at young young chess players
12:18
and under the umbrella of sport officers
12:20
foundation I have involved in working
12:23
with them and I'm talking about Kate's
12:24
off international masters run master
12:27
level it's it's such a difference in the
12:31
way they approach the game the way to
12:33
look at the pieces if it happened time
12:37
and again we reach in certain position
12:39
analyzing the game and they say bad knew
12:43
I made a mistake here so fine why oh and
12:47
then it's a long line so the Machine
12:49
showed that I understand I could see the
12:52
screen but why you think this move is
12:55
wrong and they don't understand the
12:57
question
12:57
because machine said so because it's on
13:00
the screen now so it's it's somehow they
13:02
their be the the minds being hijacked by
13:05
by the power of the machine and one of
13:07
the reasons Magnus Carlsen was so
13:09
successful and still a dominant force in
13:11
wall of chess and I remember after
13:14
working with him in 2009
13:16
10 for more than a year he never looked
13:20
at the machine as an ultimate source of
13:22
wisdom for Amos working was more like a
13:24
calculator - to verify his own
13:28
understanding and evaluation of the
13:30
position this is a big challenge but I
13:32
believe it's not only chess it's
13:33
elsewhere many people just staring at
13:36
the computers as eyes or just being you
13:39
know being caught by by the screen
13:41
expecting just to find the solution
13:43
they're just ahead of thinking for
13:44
themselves exactly that's why I always
13:46
bring care as a piece of wisdom the
13:49
classical phrase from Pablo Picasso that
13:51
computers are useless because they can
13:53
only give you answers yeah so but
13:55
everything begins with a question coming
13:57
you know essentially - about Magnus
13:59
Carlsen you say that he although he's
14:01
growing up in the computer chess area
14:02
he's one of the most human-like in
14:04
recorded or intuitive players around
14:06
right so it's kind of interesting he's
14:08
consistent
14:09
it's a yes it's human because at the end
14:13
of the day 20 years after my match was d
14:17
blue more people playing chess than ever
14:19
before and the chess is still very
14:21
popular because at the end of the day
14:23
it's it's a fight between two
14:25
individuals and what has changed is not
14:29
it's not this game itself but the way
14:31
people are watching it 20 years ago or
14:34
30 years ago 40 years ago the World
14:37
Championship match was kind of analogous
14:39
in event of absolute quality even
14:41
carpooling Kasparov played the game and
14:43
one made a terrible blunder it could
14:45
take time in the press center for
14:47
grandmasters can find out too - -
14:49
whispering mistake and it's like you
14:53
know that it's something that you just
14:54
worship today you know I when I'm
14:57
watching what I'm watching the games but
14:59
Magnus Carlsen Caruana and you have
15:02
thousands of amateurs from all over the
15:04
world watching did you keep describing a
15:06
mistake this technical machine shows
15:08
immediately the devaluation drop so it's
15:11
some kind of respect has disappeared yes
15:13
from that is a real shame but but but
15:16
also but also it added interest because
15:19
people can follow the exact to do this
15:21
and they do have access to the computer
15:23
and they don't have to then already
15:25
strong players to understand what is now
15:27
what is happening and
15:29
one other what an interesting thing you
15:31
said actually about the chess computers
15:33
and I wonder if it's going to happen
15:34
we'd go as well in the countries that
15:36
are not traditionally good at chess or
15:38
go because they have access to these
15:40
machines maybe kids in those countries
15:42
can now get very strong right like
15:44
Magnus in the way or I don't know
15:45
whether that's you know got sure Magnus
15:47
its rise Medora cries was due to
15:52
computers and maybe it's because even
15:54
you GT in in this environment you don't
15:57
have to spend so much time learning from
15:58
other players so the process of maturing
16:02
for the chess player is much shorter
16:04
you have grandmasters at 1450 today I
16:08
know much more Bobby Fischer knew 40
16:10
years ago
16:10
just because they played better games
16:12
they could travel around they could
16:13
watch the games
16:14
so chess chess is a perfect match for
16:17
for internet because you can follow the
16:19
games you can learn you can analyze so
16:21
the many things you can do that
16:23
dramatically increase the pace of
16:25
learning and getting to the top so you
16:29
invented I think the concept of advanced
16:31
chess right man and computer human and
16:34
computer issue first human and computer
16:36
versus computer are you still stronger
16:39
where have you tried that recently with
16:40
the the latest chess engines a you is
16:42
that not all true now oh yes there while
16:45
licking my wounds after the blue match
16:47
so I thought how about bring you
16:49
together just out of curiosity because I
16:52
saw wait a second if I just can play
16:54
with losing machine just against another
16:56
player so maybe you can play a perfect
16:58
chess now the interesting thing is when
17:00
we played this match was Veselin Topalov
17:02
another another top player in 1998 I can
17:06
tell you the quality was not very high
17:08
because it was a limited amount of time
17:10
and we it was so new for us how to use
17:13
the machine at and eventually I realized
17:16
and we had many events the so-called
17:18
freestyle event on the internet that
17:20
proved that it could sound quite ironic
17:23
but you don't need a very strong player
17:25
to get the best result of human plasma
17:28
shield combination it could sound like a
17:30
heresy now but is I would say that you
17:34
don't you don't want a strong player
17:36
it's you need a good operator ignition
17:39
player but someone
17:40
we'll follow that machine I should guide
17:42
the machine but not will you to use the
17:44
machine to back up his or her own ideas
17:47
because instinctively if I came up with
17:51
a computer I'll try to see I try to make
17:54
my own Mugen you don't have to only need
17:56
is just to maximize the effect of
17:58
machines play because machines are so
18:00
strong now only all you have to do is
18:03
just to guide sometimes you can feel now
18:06
just you know slight correction move
18:07
here move there so it's it's it's
18:10
something that requires a very different
18:13
kind of qualities it's more about
18:15
interface so you don't need great
18:18
knowledge of the game my weight it may
18:20
help but it's to the other side it may
18:23
preclude you from sort of using machines
18:26
power bands you try to play your own
18:28
game which could be detrimental so it's
18:30
it's something that you know I think you
18:33
touched on in a few places it's become
18:35
known as Kasparov slow now right
18:36
something like that the process is
18:38
actually I won't explain what that in
18:39
and again and again I just I relied on
18:41
results of the of the Freestyle
18:43
tournaments and what's happened is there
18:46
that as predicted a human plus machine

--------------------

18:46
that as predicted a human plus machine
18:50
beat supercomputer quite handily but the
18:54
most unexpected story was that it's as I
18:57
described a relatively weak human or
19:00
group of humans plus machine or machines
19:03
mclarson plus a better process they of
19:07
course they beat supercomputer not more
19:09
remarkably they being a strong human
19:11
plus a machine plus inferior process so
19:14
that's let me do consider conclusion
19:15
that's all about interface there's so
19:18
many ways of of of empowering machines
19:23
without creativity so not our creativity
19:26
with machines brute force of course
19:27
actually do it other way around
19:29
and then the result is equally good new
19:31
phenomenon so it strikes me in the whole
19:33
book you're very optimistic about
19:35
technology in general work you know in
19:36
terms of like what it might be able to
19:38
do what what is this kind of process is
19:40
that a kind of blueprint for you know in
19:42
advance chess of how you see things
19:44
going forward in other areas of life
19:46
with machines and humans working
19:48
together in a complementary way I
19:51
believe the future was a soul fulfilling
19:53
prophecy
19:54
and I cannot stand all these you know
19:57
doom and gloom predictions it's it was
20:00
quite amazing when you just look at ad
20:02
the change of the trend in science
20:05
fiction from fifties and sixties where
20:07
it was all about optimism us teaming up
20:11
with computers
20:11
Robert cyborgs flying to two other not
20:16
just to other planets but to other star
20:18
systems and then it's a chance to a very
20:23
dystopian vision of the Terminator and
20:28
the matrix by the way there's a sting
20:31
about the terminator you know I just
20:33
it's I I just wish I just got on here
20:36
just having a lecture in Dallas Texas
20:38
all these months I looked at the
20:40
Terminator I said know what guys I can
20:43
tell you that's that's another proof of
20:44
what you call Kasparov law because we're
20:47
all watched the first one human versus
20:50
machine but if you follow the number two
20:52
and number three that was exactly what I
20:55
said a human that was machine plus a
20:58
better process within a supercomputer
21:02
very it's me yeah and and I think it's
21:06
just it's it's what we learn from
21:07
chances that there are many ways of us
21:09
of us sort of gain income getting some
21:12
something new something positive out of
21:14
the corporation and by the way these
21:15
things are going to happen anyway so
21:17
let's see what's the point of trying to
21:19
slow down which is a natural cycle that
21:22
we have technology replacing certain
21:25
elements of human activities for
21:29
centuries technology was there machines
21:31
have been replacing blue-collar jobs now
21:33
the difference is down machines of sweat
21:35
ninka people with college degree
21:37
political inputs and twitter accounts
21:38
that's that's why we hear all the
21:41
stories about it but that's that's
21:43
actually normal I think that's that's
21:45
that's all progress and it's if machines
21:48
taking over certain they're certain
21:50
menial parts of elements are or aspects
21:54
of cognition that's that story end of
21:57
the world it's still many things that
21:59
humans can do all we need is just to
22:01
look for new challenges and for new
22:03
frontiers
22:04
so we just come back from China for
22:07
their alphago match again circa je and
22:09
one thing that happens in Goa which is
22:11
slightly different than chess is and go
22:12
there's a tradition of players thinking
22:14
about how far off from optimal play are
22:17
they and in theoretical God player
22:19
optimal play so how far you do you think
22:21
even the top chess computers are from
22:23
optimal chess I mean what do you think
22:25
that the top a low rating would be
22:26
possible to play at chess at is do you
22:30
have an idea no I don't know any idea
22:32
because I just would briefly discussing
22:34
in her lunch is that when you look at
22:36
the app at the data endgame databases
22:40
and now we have all seven pieces that's
22:42
I don't hundred terabytes or whatever
22:44
and so every position has been
22:47
calculated from well to the very end and
22:49
in many cases you just you know we have
22:51
a position that says Made in 492 moves
22:55
and I bet you that in the first 450
22:59
moves you will not see the difference so
23:01
I could see poor always falling 20 moves
23:02
yes but it's just now I don't know what
23:06
it tells about the game we play because
23:08
the average human game is 50 moves now
23:10
when you look at average machine games
23:12
it's maybe 80 90 moves it doesn't mean
23:15
that the game should be too long or what
23:17
we know that you know the game of chess
23:19
is an alternate endgame we study two
23:21
pieces so that's why I don't see any
23:22
chance in any future that machine will
23:26
play a twenty four and we'll announce
23:29
made in 16,000 7:55 and it's not going
23:32
to happen the number of legal moves in
23:34
the game of chess 10 power 45 that's
23:36
enough to to feel safe but it's not
23:40
about solving the game it's about
23:41
winning the game and and I've even still
23:44
you know some improvement the machines
23:45
could get better and better this is the
23:47
I mean basically sky's the limit and
23:49
today I still think Magnus was white in
23:53
his good day we'll probably secure a
23:55
draw against the machine but winning
23:58
against the computer today it's
24:00
virtually impossible be the level of
24:02
precision that require it hasn't
24:04
required their level of vigilance it's
24:06
just it's impossible so which we know
24:09
and we're not used to to play with such
24:11
attention so what machines will get will
24:14
get better and by the reshape we see and
24:17
improvement all the time I remember this
24:19
as well by writing my books my great
24:21
predecessors and then my matches against
24:22
Karpov and then my own by my best games
24:25
and some of the games the same games
24:27
analyzed 2-3 years later which is this
24:29
new version of this imagine an engine
24:32
and just I could see that you know that
24:35
some of these some of the some of the
24:37
moves that I treated as great in say
24:42
2009 in 2012 I was powerful powerful
24:45
computer I had my doubts yeah very cool
24:48
so though I've got so many more
24:49
questions but I know I should I should
24:51
give some time a time answer is moving
24:53
forward so I don't know I should let the
24:54
audience ask any question if you put
24:57
your hand up high so I can see Wow
25:00
silence we covered everything well let
25:05
me ask you another quick let's take a
25:06
little questionnaire there yeah ok ok so
25:11
now I want to ask you I was too young
25:15
when the deep-blue match happened so I
25:16
don't have any personal member in the
25:18
really young yet but when I read stories
25:23
of it I kind of struck me that the match
25:27
seemed like everyone was it had great
25:30
publicity but people really wanted to
25:32
see whether well to believe and whether
25:35
he would lose against the machine and I
25:37
found this really my question is
25:40
basically do you feel like this was the
25:43
case or do you feel it actually felt
25:44
like a normal just match where people
25:46
would see who would win or rather did
25:48
you see like you said support on your
25:50
side as well
25:51
oh we are at little support I can tell
25:53
you that most people who wanted me to
25:57
lose the rise in world chess now because
25:59
I was a world champion for 12 years and
26:01
I'd won the first event I have a lost so
26:03
that's naturally a lot of people wanted
26:05
me to lose one day and since I was
26:07
unbeatable in the inhuman chest so they
26:09
they had they had a host and a machine
26:15
but the atmosphere there was phenomenal
26:20
and it sub D it was a reflection of the
26:23
of the famous cover of the Newsweek the
26:25
brains love scanned and I remember when
26:28
I want get one of the match
26:30
was a big celebration akeno's Rahman CNN
26:33
they they talk to you the the two
26:37
presenters they they talked about it and
26:39
said look it's a Russian playing an
26:41
American machine but I'm looking for
26:43
Russian and the questions to rebel from
26:55
the alphago team humans seem to be more
27:00
efficient in playing both stress and go
27:03
probably and that they evaluate much
27:05
fewer variations and positions than than
27:08
computers do I mean by many orders of
27:11
magnitude probably can can you give us
27:13
an intuition of how this difference can
27:16
come about what what are humans actually
27:19
so good at in chess that they can can do
27:22
this so efficiently and that they only
27:24
need to examine so few variations as
27:27
compared to computers now we can talk
27:31
about sort of general rules but also you
27:34
should remember that there are different
27:36
playing styles because the way carpet or
27:39
myself will look at the same positions
27:41
it'd be very different because I will
27:43
make a look for an opportunity to break
27:46
through just to sharpen the game to
27:49
create complications and carpel will be
27:51
looking for sort of long-term strategic
27:53
advantage that could manifest in the
27:55
endgame except those are differences
27:58
what brings us together is that as you
28:01
just said we didn't have to analyze
28:04
millions of lines we couldn't so we
28:08
could look for one or two options how do
28:11
we know that those two options are the
28:14
best I don't know I just simply I know
28:17
this what it is and then again it's the
28:20
it's theater then we had another subtle
28:23
difference I will probably try to go as
28:27
deep as I can calculating Karpov will
28:29
try to look for an option where he
28:31
doesn't have to calculate at all so
28:32
relying on his understanding because
28:34
that many patterns you can regular as
28:36
patterns and then bringing patterns
28:37
together you can have a picture big
28:39
picture that's that's what she was a
28:41
unique at and that's that's why for
28:44
instance
28:44
if you are if you team up with a
28:46
computer sometimes they will of
28:48
calculating is that's I wouldn't go
28:50
there
28:50
it's and then it's quite interesting
28:53
that to check whether machines
28:54
calculation proves it
28:56
but in many cases that hang out the
28:58
right and and I'm special at a time when
29:01
machine goes very deep and then which is
29:05
it horizon and that he also should look
29:08
at the position and say it smells
29:11
there's something wrong I don't know
29:13
exactly what is wrong but something
29:14
something is wrong there are some
29:18
situations where you have to calculate
29:20
when you sacrifice something
29:22
your sacrifice material and that's it's
29:23
it's take it or break it so you cannot
29:26
afford to be the duty to use your common
29:29
sense because you have material down so
29:32
my game is Veselin Topalov another one I
29:34
played in 1999 that's my longest
29:36
combination so I I cannot tell you that
29:40
I saw though every line there it was not
29:43
be true but the combination the the
29:46
final position that I saw it's like a no
29:47
light link just very quickly what will
29:50
happen at the end included the the
29:54
should apply legs Kristin foot moves
29:58
so that's ironically that because I saw
30:02
this the final position as later machine
30:05
proved that I could win earlier and top
30:08
of mister chance to you not to escape
30:09
but xst the end game that he could
30:12
probably defend but otherwise it's just
30:15
it's it's what you describe is just
30:18
another proof of the moronic paradox
30:20
that's the I was going to bring up you
30:22
should explain what that it can you talk
30:24
about that is that machines are very
30:27
good at what humans are not so good and
30:29
other round so it's a chess chess it's
30:32
it's interesting that chess was probably
30:36
because go and shogi they were just
30:38
played elsewhere but for for the Western
30:41
science chess was an ultimate test for
30:44
for artificial intelligence and that was
30:47
another result of 99-97 match that the
30:50
expectations of the founding fathers of
30:53
computer science like Alan Turing
30:54
optional Norbit winner that machine
30:57
beating strong chess player and of
30:59
course the world champion would be it
31:01
this is the EDD moment for AI I have to
31:05
say they were wrong
31:05
so existence and people was as
31:07
intelligent as your alarm clock yeah
31:10
tell edge all along I actually have a
31:13
theory about the more average paradoxes
31:15
the in explanation for that if you have
31:17
hand-built systems like the blue walls
31:19
then as the programmers you have to
31:23
understand clearly enough what you're
31:24
trying to codify explicitly so you can
31:27
codify in a rules or heuristics like the
31:29
Guru was and the problem is is that for
31:31
many things that we take for granted as
31:33
humans like vision or riding a bike all
31:36
these things we do implicitly we don't
31:38
explicitly understand well enough how we
31:40
do those things so we can't codify it
31:42
and that but that's why I think that
31:44
learning systems the kinds of things are
31:46
alphago might end up being more powerful
31:48
because they could learn how to
31:50
experience how to do those things like
31:51
humans do
31:52
oh it's seeds it's one of the one of the
31:54
rules that are learned you know from my
31:56
experience is that anything that we do
31:59
and we know how would you machines a
32:01
little better because we can communicate
32:03
it so it's one were not it codified so
32:06
no big questions now whether machines
32:09
could ever do the interior was lesser
32:11
things that we do without knowing that
32:14
yeah yeah no exactly that's the big
32:15
question right I mean I think you say in
32:17
your book up to now anybody who tempted
32:20
learning systems including your great
32:22
teacher Botvinnik were fell short
32:24
against the against the basically be
32:27
Hank I years ago 40 years ago because
32:29
it's in the beginning there was a big
32:31
debate and and I think that state states
32:33
I've been curious people know by the way
32:37
that you're the first program in 1952
32:39
there was a chess program and if the
32:42
trick was that there was no computer
32:43
yeah the only game that they did the
32:46
Turing problem yes exactly just he put
32:49
on a piece of paper and calculated the
32:51
moves and and when I spoke at this
32:54
antenna Rizzo I asked my friends from
32:55
Germany legs would reconstruct cities
32:57
and put in a computer code you guys play
32:59
a Turing machine it's pretty weak but
33:01
it's 1952 yeah so but they believed and
33:04
I think they believed that the way to to
33:09
make machines playing chess it's not
33:11
force but understanding but it failed
33:14
viscosity is very quickly because brute
33:16
force kept coming and the studio that's
33:19
all I got was like avalanche they they
33:21
couldn't stand a chance so that's why
33:23
the all attempts including one of my
33:26
great teacher make up a clinic to come
33:28
up with this with this parallel concept
33:30
of learning failed and but in the
33:33
sixties early seventies it was just it
33:35
was it the store was over now it seems
33:38
that we're just like an incisions we go
33:40
back to to these to this to this notion
33:42
and maybe it will prove to be superior
33:44
hopefully alphago will make above manic
33:46
happy then you know it's it's sort of a
33:48
learning system right so it turns out
33:50
the go-go needed to need it to happen
33:52
what do you think the difference is
33:53
between go and Chester required go to
33:56
have that you know how to have this
33:57
other approach you know they couldn't do
33:58
it handcrafted approach now look it's a
34:03
tough question because I have nearly
34:05
absolute knowledge of the game of chess
34:07
and almost zero knowledge of the game
34:09
we've got yeah so but what I know is
34:14
that it's the God doesn't have the same
34:17
tactical configuration so you you you
34:22
have to it's all about strategy just
34:24
years it's a long term and and that's
34:27
why it's far more difficult for machines
34:29
to learn how to do that but also if
34:32
machines could do a certain level then
34:34
they could be deadly for humans because
34:36
it's state they suddenly become superior
34:39
so my and I again I'm not sure and since
34:43
I'm sticking to great expert I still
34:44
think that relatively up it if you
34:47
compare the strengths I think the chess
34:50
playing engines are relatively stronger
34:52
than alphago in absolute writings
34:56
but but then you look at it as but then
34:58
the gap between it's just because the
35:01
mistakes made by human players and god
35:02
they're deadly they are just here they
35:05
they could take it off or more openness
35:07
for the machine so in chess the human
35:10
game is always unstable so it's not as a
35:12
state as a machine but I think in in God
35:15
the depths of off of the mistake - it
35:19
could be could be far more significant
35:22
than inches I guess we'll have to put
35:23
that to the test by
35:25
alphago chess to play chess right and I
35:27
will have it okay hello be interesting
35:30
yeah so how soon alphago can crush the
35:33
strongest Church it just engines again
35:36
it's it's worse said it's not about
35:37
alphago but it's about the league shall
35:39
be Gabe Go Go and chess but I you know
35:42
would you leave the pretty surprised if
35:44
a learning system could beat the
35:45
handcraft the system delivering chess I
35:47
look at this is that's that would be
35:48
another level of experiment because the
35:51
the current systems a it's not primitive
35:53
goof force anymore now it says that's
35:56
why I said it this today any and by the
36:00
way that's that's the moment I say
36:01
people just still look at me in
36:02
disbelief from non-professional
36:05
audiences I said a free chest app or
36:07
your mobile for the stronger that the
36:08
blue is your soul loser yes I'm so loser
36:14
it doesn't change the fact that it's it
36:20
now very good so I don't if there's a
36:22
last question from the audience we have
36:23
time for one or two more yes from the
36:25
lady Duvall get to the front here if you
36:32
could would you play deep blue a Dan oh
36:38
yeah this be a couple of problems one
36:41
I'm a retired and I don't know
36:42
commercial just to be boozed it
36:46
yeah I wonder boy in 1998 and I I
36:52
I wish I had a chance but um that's
36:56
that's it you know that's that's that's
36:57
old history it's them it's a still milk
37:01
what on the bridge here you name it but
37:05
I laid a plate other computers and I as
37:08
long as I was active chess player I
37:11
never ducked a challenge and and that's
37:14
why you know this is study this book
37:17
begins was a story of me playing
37:21
32 chess computers in 1985 its
37:25
simultaneous exhibition I'm not sure
37:28
that anyone oh you old the chess machine
37:32
that from an antique chest machines
37:35
anyway here I travel
37:39
Loretha so I play that played it's 32
37:42
machines and there were four
37:44
manufacturers eight machines each and I
37:47
want all the game's most amazing thing
37:50
that nobody was surprised and it can
37:53
tell you the progress it's just from
37:56
that match in 1985 just there plated in
37:59
June just a few months before I want the
38:01
title beating Karpov to my match people
38:05
just 12 years it tells you that
38:06
something's happening and but I couldn't
38:09
help but reminding people about this
38:11
1995 because I say that with the Golden
38:13
Age yeah machines were weak my hair or
38:16
strong hahahaha
38:18
yes we're entering another very clear
38:20
thing I think interesting era so I think
38:22
let's all thank Gary for an amazing
38:24
discussion and thank you very much
38:30
[Applause]
38:39
you

Don't fear intelligent machines. Work with them | Garry























No comments:

Post a Comment