I'm honored to be here, but please don't hack me. Computers already hate me, so I need to make friends with as many people in this room as possible. I want to bring one small trifle from my biography, which is interesting for the American audience. I was born and raised in the very south of the country, right next to Georgia. This is actually true. Wait a second, I told you computers hate me!
One slide is lost, but this is really the very south of the USSR, where I was born in a republic that was located right next to the Republic of Georgia (approx. translator: the name of the state of Georgia and the republic of Georgia sound the same in English).
Speaking of my homeland, the funny thing is that my last book, Deep Thinking, was written about artificial intelligence, about my own experience of battling computers, and the book, written two years earlier, was called Winter Is Coming. It wasn't a synopsis of Game of Thrones, it was about Vladimir Putin and the fight for the free world, but when I was touring the presentation of this book, everyone wanted to ask me about chess and the IBM Deep Blue computer. Now, when I present the book "Deep Thinking", everyone wants to ask me about Putin. But I'm trying to stay on topic and I'm sure there will be a few questions after this presentation that I'll be happy to answer. I'm not a politician, so I don't shy away from answering questions.
It may seem strange that the game of chess, which originated thousands of years ago, God knows when, is the perfect analogy for artificial intelligence, because when we talk about AI, we must remember that the letter I stands for "intelligence" and there is nothing that demonstrates it's better than chess.
Many people believe that chess is nothing more than a pastime that people indulge in in cafes. If you look at the creations of Hollywood, then everyone plays chess - aliens, X-men, Wizard, vampires. My favorite Humphrey Bogart movie Casablanca is also about chess, and when I watch this movie, I always want to stand in such a position to look inside the screen and see Bogart's board. He plays the French Defence, which was very popular in the early 40s. I think Bogart was a pretty decent chess player.
I want to mention that Alfred Binet, one of the co-inventors of the IQ test in the late 19th century, admired the intelligence of chess players and studied it for many years. Therefore, it is not surprising that the game of chess attracts those who wanted to create smart machines. However, it often happens that intelligent machines, such as von Kempelen's "Turk", are just a grandiose swindle. But at the end of the 18th century, this chess machine was a great miracle, it toured Europe and America and fought strong and weak players, such as Franklin and Napoleon, but of course, it was all a hoax. The Turk was not a real machine, it was an original mechanical system of sliding panels and mirrors, inside which a strong player was hiding - a man.
What's interesting is that a hundred or two hundred years later, over the past twenty years, the opposite has been observed - we see in tournaments that human players try to hide computer devices in their pockets. So now we have to look for a computer hidden in the human body.
However, the history of mechanical devices is comparatively little known. The first mechanical device for playing chess appeared in 1912, it played with a single mechanical part, it could checkmate with a rook, but it could not be called the prototype of the first computer.
Interestingly, computer design pioneers such as Alan Turing and Claude Shannon showed great interest in chess. They believed that playing chess could unlock the secrets of artificial intelligence. And if one day a computer beats an ordinary chess player or a world chess champion, this will be a manifestation of the evolution of AI.
If you remember, Alan Turing created the first computer program for playing chess in 1952, which was a great achievement, but even more significant was the fact that there were no computers at that time. It was just the algorithm he used to play chess and it acted like a human computer processor. It is important to remember that the founding fathers of computers determined the way that AI was to develop, following the processes of human thought. The opposite way is what we call a brute-force attack, or a quick enumeration of possible moves.
I didn't hear anything about the competition with computers in 1985, but in this photo you can see 32 boards, and although I played with people, it was actually a real game against computers. At that time, there were 4 leading manufacturers of chess computers, which just introduced them to the world. Perhaps some of you still have such computers, now they are real rarities. Each manufacturer had 8 computer modules, so in reality I played with 32 opponents and won all the games.
It is very important that this was not a surprise, but a natural result, and every time I look at this photo of my victory, I remember this time as the golden age of chess machines, when they were weak and my hair - thick.
So it was June 1985, and 12 years later I was playing against just one computer. There was a rematch in 1997 because I won the first match, which was in 1996 in Philadelphia. I lost that rematch, but to be fair, the turning point in computer chess was not in 1997, but in 1996, when I won the match but lost the first game. Then I won 3 games and the score became 4:2 in my favor.
In fact, the important fact here is that a computer at that time was capable of becoming a world chess champion if it played in an ordinary chess tournament. I did not expect from IBM that they could do such a serious technical job in a year to strengthen their computer. But my biggest mistake, with the exception of a huge increase in the price of IBM stock, which jumped from a few points to a billion dollars 2 weeks after the match, was the inability to read the fine print. Because one of the problems I had in 1996 playing with the Deep Blue computer was that it was a black box for me. I knew nothing about the opponent, about how he thinks, what tactics he uses. Usually, when you prepare for a game, you study your opponent, whether it be a chess match or a football match, and by observing the way you play, you study his strategy. But there was no information about Deep Blue's "manner of play".
I tried to be smarter and stated that by the next match I should have access to the games played by Deep Blue. They replied: “Of course!”, but at the same time they added in small print:
“…only during official competitions.”
And this despite the fact that Deep Blue has not played a single game outside the laboratory walls. So in 1997 I played against a black box, and it turned out the opposite of the events of 1996 - I won the first game, but lost the match.
By the way, where were you hackers 20 years ago when I needed you so much? True, when I glance over the rows of those present, I understand that many of you, probably, were not yet born then.
My biggest mistake was that I took the match with Deep Blue as a great scientific and social experiment. I thought it would be great because it would really find an area where human intuition can be compared to the "brute force" of computer calculations. However, Deep Blue, with its phenomenal computational speed of about 2 million chess positions per second, which was not bad for 1997, was anything but artificial intelligence. His game did nothing to unravel the mystery of human intelligence.
It was no more intelligent than an ordinary alarm clock, but it doesn't make me feel any better that I lost to a $10 million alarm clock.
I remember the press conference during the opening ceremony of the match, when the man who led the IBM project said that this would mark the end of scientific experiments and the victory of science. Since we had one win and one loss, I wanted to play a third match to find out who was still stronger, but they dismantled the computer, apparently to remove the only impartial witness. I tried to find out what happened to Deep Blue, but I couldn't find out. Later, I learned that he had made a new career and was now preparing sushi in one of the terminals of the JFK airport.
I love sushi, but I don't need a computer there. So, that's where my story with computer chess ended pretty quickly. But those of you who also play chess or other games know how vulnerable we are compared to computers because we are not as stable, impartial and make mistakes. Even the highest level players make mistakes, for example, during a 50 or 45-move championship match, at least one tiny mistake is inevitable. If real people are playing, it doesn't really matter, but if you make a mistake when playing with a machine, then it is possible that you will not lose, but you will not win either, because the machine will be able to avoid defeat.
At some point, I realized that it was just a matter of time, because we cannot achieve the same level of vigilance and accuracy that is necessary to defeat the computer, because the machine is unusually stable in its actions. Years later, we have seen cars consistently win matches. I repeat once again - this is all about just playing chess, which is very vulnerable to the brute-force game method, when the computer goes through a lot of options at great speed and chooses the most optimal one. This is not artificial intelligence, so people make the mistake of saying that a human chess player has been defeated by artificial intelligence.
Later I played a few more matches against computers. I once analyzed these games using modern chess engines and it was quite a painful experience. It was a journey into the past and I had to admit how badly I had in those matches because I could only blame myself. However, at that time the computer "demon" was not so powerful, you may not believe, but the free chess application on your mobile device is stronger today than Deep Blue was. Of course, if you have a chess engine like asmFish or Comodo and the latest laptop, this system will be even more powerful.
When I played against Deep Blue, I think it was game 5, the computer put perpetual check in the endgame, and everyone started saying that it was a great win and that the computer showed a phenomenal quality of play. But today, with a modern computer, it just looks ridiculous. Our entire match can be played in 30 seconds, up to a minute depending on the performance of your laptop. I made a mistake at the beginning, then I tried to save the game, Deep Blue made a few counter moves and won. These are the rules of the game and there is nothing wrong with that.
In 2003 I played 2 more matches against the X3D Frintz computer, both of them ended in a draw. The organizers made me wear 3D glasses because the computer had a 3D interface.
But anyway, the story was over and I was thinking about the future. Look at this photo, which was taken at the beginning of this century.
If you look at these children, you can see that they are playing on rare computers. Today my children won't even understand what it is. Here are some complex keyboards, but now they just slide their fingers over the touchscreen.
The important thing is that smarter machines make our tasks much easier. I probably shouldn't say this because you know it better than anyone else. Thus, with the help of Peppa Pig and technical challenges, the way is cleared for true creativity.
I thought about how you can combine the power of a computer and a person? We can take chess as an example, because chess has a solution. You know perfectly well in which areas the computer is strong, and in which it is inferior to a person. And then a concept came to my mind, which I called "advanced chess".
Following the Russian proverb: “if you can't win, join!”, I called advanced chess a game where one person with a computer fights against another person with a computer.
In 1998 I played with a representative of the chess elite from Bulgaria, and the most interesting thing is that both of us could not play well, because we could not maximize the effect of working together with a computer. I wondered why two great players could not benefit from the cooperation with AI. The answer came later with the introduction of the so-called freestyle with a limit on the number of hints from the computer. You can play by connecting to a supercomputer via the Internet, you can use your own computer or many computers. I want to point out that the human-computer pair will always outperform any supercomputer. The reason is very simple - the computer compensates for our absent-mindedness, and we are in a good position to switch to the computer, because this eliminates the vulnerability when another computer can take advantage of our human weakness.
But there is nothing sensational about it. The sensation was that the winners of the competition were not top-class players, but relatively weak chess players with ordinary computers, but who managed to create an improved interaction process. This is hard to articulate because it sounds paradoxical: a weak player plus a normal computer plus an improved process outperforms a strong player with a powerful computer but a weak interaction process. The interface is everything!
The interesting thing is that you don't need a strong player at all, you don't need Garry Kasparov to find the best move from the side of the machine, and there is a simple answer to that. If today we consider the relative strengths of a person and a computer, we can go beyond the limits of chess, but let's start with them anyway, because there are numbers in chess. So, my highest chess rating was 2851 until I lost to Magnus Carlsen, and at the end of my chess career it was 2812. Today, Magnus Carlsen leads the rating above 2800 points. Approximately 50 players have a rating between 2700 and 2800 points. This is the elite of the chess world. Nowadays, the power of a computer is in the range of 3200 points, and taking into account specialized software, its rating can reach 3300-3400 points.
Now you understand why you don't need a strong player? Because a player of my level will try to push the computer to act in one direction or another, instead of being a mere operator with it. Therefore, a weaker chess player who does not have such "arrogance" and such conceit as a world chess champion will interact with a computer much more efficiently and form a more productive "human-computer" combination.
I think that this is a very important discovery not only for chess, but also, for example, for medicine. As you know, computers in many cases are able to make a more accurate diagnosis than the best doctors do. So what would you like more: a good doctor in the form of a computer, or a good nurse who will simply follow the instructions and write a little guide based on the recommendations of the machine?
I don't know the exact numbers, let's say 60-65% of people will choose a doctor and 85% will go to a computer, but psychologically, if you are a good doctor, you can't accept that. If you look at today's technological progress, we can say that computers make a true diagnosis in 80 - 85 - 90% of cases, but 10% still remain for people! And this can make a huge difference, because when a bullet is deflected by only 1 degree when fired, it can fly several hundred meters from the target. The question is whether we are able to direct the full power of computer computing in the right direction.
Therefore, I still believe that all fears about the fact that machines will soon replace all of us, and this will be the end of the world, Armageddon, are just rumors. Because, as I said, this is about human creativity, and the uniqueness of computer intelligence is that it just enhances our creativity, releases it and tells us how to use it in the best possible way.
Sometimes, to find the answer to a question, it is worth moving away from the world of science and delving into the world of art. Somehow I found a great paradox stated by the great artist Pablo Picasso: “Computers are useless. The only thing they can do is give answers.” I think there is great wisdom in this and it sounds reassuring, because machines do provide answers, and these answers are exhaustive!
However, Picasso was not satisfied with exhaustive answers because he was an artist. It has to do with the constant rethinking of art, which is exactly what we do all the time – ask questions. Can computers ask questions?
I once visited the hedge fund Bridgewater Associates, where I was going to talk to Dave Ferrucci, one of the developers of IBM's Watson supercomputer. We were talking about whether machines can ask questions, and Dave said, "Yes, computers can ask questions, but they don't know which questions really matter." That's the point. Thus, we are still in the game and we have a chance to move on, because the game of man and computer is not over yet.
On this slide, you see several photographs of possible uses for autonomous computers, machines that can program themselves, that is, have the ability to learn.
One of the photos shows Demis Hassabis with his self-learning neural network AlphaGo. In fact, this is probably the first machine that can be called a prototype of artificial intelligence.
As I said, Deep Blue is a brute-force bust, Watson is perhaps a transitional link, but not yet AI. AlphaGo is a deep learning program that improves itself by finding relevant patterns by playing millions and millions of games.
I can say that in the case of AlphaGo we are dealing with a real "black box" for the first time. Because, for example, if we spend a hundred years studying thousands of miles of Deep Blue game logs, we will eventually get to the original idea of why this decision was made and this particular move was made. As for AlphaGo, I'm sure that even Demis Hassabis himself can't say why version 6 is better than version 9, or vice versa, referring to the decision taken by this machine.
On the one hand, this is a great achievement, but on the other hand, it can be a problem, because if the machine made a mistake, you will not be able to know about it. However, in any case, this is a movement towards the creation of real AI.
I was speaking at Google headquarters the other day, and they gave me a tour of Google X. This one was very interesting, because the company is moving forward in the direction of AI, solving the problems of creating a self-driving car or autonomous drones that deliver goods on their own. However, no less a problem than the technical support of AI is the problem of regulating its activities. People talk about how AI can completely replace them, putting them out of work. However, let's call on the history of human civilization to help - this has happened for hundreds and thousands of years!
24:35 min
Thank you for staying with us. Do you like our articles? Want to see more interesting content? Support us by placing an order or recommending to friends, 30% discount for Habr users on a unique analogue of entry-level servers, which was invented by us for you: (available with RAID1 and RAID10, up to 24 cores and up to 40GB DDR4).
Dell R730xd 2 times cheaper? Only here in the Netherlands! Dell R420 - 2x E5-2430 2.2Ghz 6C 128GB DDR3 2x960GB SSD 1Gbps 100TB - from $99! Read about
Source: habr.com