Kasparov

Kasparov on interplay between machine learning and humans | Business Insider | May 24th, 2017

5.24.2017

by Elena Holodny

May 24, 2017

READ ORIGINAL ARTICLE AT BUSINESS INSIDER

Garry Kasparov, one of the greatest chess players of all time, is famous for his pair of faceoffs against the IBM supercomputer Deep Blue.

Kasparov won the first match against the computer, 4-2, in 1996, but lost in the rematch, 3½-2½, in 1997. He recently published a book, “Deep Thinking,” about the experience.

Business Insider recently spoke with Kasparov about Deep Blue, his thoughts on AI, and machine advancements over the past 20 years — and how he sees the interplay between machine intelligence and humanity.

This interview has been edited for clarity and length.

Elena Holodny: What’s the biggest misconception about AI?

Garry Kasparov: AI as a concept is surrounded by mythology.

Most of the things we mention we understand. You know, if we say “white,” we all see it’s white. If we talk about elements of computer science or some general items, we are in agreement. There’s no need to go into definition.

Now with AI, the moment we mention “AI,” you should spend a considerable amount of time with the person understanding what is AI for this very person. And that was one of the purposes of my book, just to remove the mythology, to look at the problem objectively, without the utopian expectations or without dystopian fears.

And also to understand, what is the nature of human intelligence. Obviously I understand my own limitations. But it is important to recognize when we say AI, what do we mean, and what do we expect?

One of the biggest problems that arise in the beginning of the conversation is: Do you mean intelligence as a result of the AI or as a process? Because when you look back at my matches that I’ve played with chess computers, now, if we stick with the intelligence as a result, then by the definition of its output, Deep Blue was intelligent because it played grand-master-level chess. Now, if you look at the process, if you try to understand the intricate details of human intelligence, now Deep Blue, this phenomenal speed of 200 million positions per second, offers you no information because it was as intelligent as your alarm clock.

And it’s a big issue. Many people simply don’t recognize that this discussion — still open-ended discussion — can bring us to very different conclusions. And probably if we’re talking about misconception, the first misconception is that people simply don’t agree what they mean by saying “AI” and why AI could be good or bad for us in the future.

Holodny: You draw the distinction between the process of thinking versus the result. Obviously computers are good at calculating and humans are very good at analogical thinking and seeing patterns.

Kasparov: We have different ways of evaluating the positions.

For instance, if you try to oversimplify it, in the game of chess, there’s a certain position and I have to make my choice. My decision will be based on, very roughly, 1% of calculation — probably even less — and 99% or more of understanding, of looking at patterns, drawing information from my previous experience.

Now the machine will be exactly the opposite. It will be 99% calculation and some percent of understanding, though this understanding is growing.

Today, chess programs — they are far more sophisticated than Deep Blue. A free chess app on your mobile is better than Deep Blue, stronger than Deep Blue. So maybe it’s no longer 99-to-1, but still it’s the decision, the core of the machine’s decision, is always based on calculation.

And that’s why we have to realize that all experiments that are related to the games when you have humans versus machines in the games — whether it’s chess or “Go” or any other game — machines will prevail not because they can solve the game. Chess is mathematically unsolvable. The number of legal moves is about 1045 …. But at the end of the day, the machine doesn’t have to solve the game. The machine has to win the game. And to win the game, it just has to make fewer mistakes than humans. Which is not that difficult since humans are humans and vulnerable, and we don’t have the same steady hand as the computer.

So chess, as we found out, could be crunched once the hardware got fast enough, the database got big enough, and the algorithm got smart enough. But again, even if you move from Deep Blue and chess to AlphaGo, which is more complicated, more strategic, and, I would say, looks more like our expectations about AI, we’re still staying in the territory of games where the machine prevails because humans make mistakes.

It’s not that machines are impeccable. Looking, for instance, at the games we played in 1997, and using modern computers, I found out that it’s not just I who made mistakes, but Deep Blue made quite a few serious mistakes … serious mistakes that could bring the game from a drawing position to a losing one. And I’m sure in 20 years, we’ll have even more powerful machines, and will look and say, oh by the way, maybe it’s not that easy.

We should simply accept the fact that the way machines make decisions is different, and rather look at the result. If machines are providing results that we are looking for, you would mind how much human understanding was used in the process. And more likely we should look for the way of combining human skills and machine skills. And that, I believe, is the future role of humanity, is just to make sure it will be using this immense power of brute force of calculation for our benefit.

Holodny: A chess piece’s relative value can change over the course of the game, or a weaker piece could be in a stronger position than a stronger piece. How does a computer think about shifts like that?

Kasparov: The machine doesn’t care about psychological problems like sacrificing a stronger piece. It looks at immediate returns. So the smart algorithms and very fast hardware, they allow a machine to look quite deep, to actually see the consequences.

What you mentioned is still one of the weakest elements, because long-term compensations — sacrificing some material for long-term strategic advantages — that could be challenging for a machine because it still has to see immediate returns. But in most cases, these kinds of sacrifices, they are within the machine’s reach. And as long as it can see that at move four or five it will get something in exchange, it’s not a big deal.

For example, from the machine’s perspective, the solution is very, very simple: You just have to sacrifice the queen. For the machine, it doesn’t matter because it immediately sees that in two moves, it will win and this is the only way to win the game.

But still, for a human player, just to give up a queen for nothing, even for one move — you simply don’t look at that. Humans have some kinds of dead zones. I don’t look there because it’s against what I learned: You don’t give up the queen.

Machines look at everything, so that’s another big advantage. And as I said, the areas where machines are relatively vulnerable, they are quite narrow. But still, if you bring human-plus-machine versus the most powerful machine, the former combination will win because human advice in this very special situation could be vital.

Holodny: It’s interesting because that’s true of other industries too. In neuroradiology, a human is less accurate than a machine, but a human and a machine are more accurate than a machine.

Kasparov: Yes, that exactly. A machine helps us to annihilate our weaknesses. We don’t have a steady hand. We can lose all vigilance. We can be distracted by something that is not that relevant. But we have intuition. We can feel certain things. And with a machine you can check whether it’s right or wrong. That’s why by bringing [the two] together, you create a very, very powerful combination.

Now, what is the most important element of this combination? It’s an interface. Let me stick with chess, but I’m sure you can look for other examples. [Sometimes] you have a relatively weak player, not a top player, because she or he could be a better operator. Because with the machine it’s very important to help the machine, rather than trying to play on your own. So you don’t need too many of your ideas. Yeah, you have to look, but still, if you have a powerful interface, especially if you deal with more than one computer; if you start bringing them together and checking different lines, [then] the operator has an advantage — a good operator — over a very strong player. Because at the end of the day, all that’s needed is human skills to maximize the machine’s output, rather than great human understanding to be assisted by the machine.

Holodny: It’s easy to see how human’s intuition can be weaker than a computer, but do you think there are examples when it’s an advantage to act on intuition?

Kasparov: Again, we are on a very slippery ground of semantics. You know, what is intuition? Some of it’s based on experience.

Holodny: Yeah, “Napoleon’s glance,” for example.

Kasparov: Yeah, but taking just, you know, pure human decision versus machine, I think that if you look for … If you run a test, and if you have enough samples, I think the machine will prevail eventually. But there are certain moments where, intuitively, I would bet on intuition — especially if we are in the area where machine expertise is limited.

Holodny: Like an early chess program?

Kasparov: Let’s talk about 2017. Forget about early chess programs. I reached a conclusion that anything that we know how we do, machines will do better. Now, the key element of this phrase is, “We know how we do it.” Because we do many things without knowing exactly how we do them. So this is the area where machines are vulnerable, because it still has to learn from some kind of experience. It needs something — at least the rules of the game. You have to bring in something that will help the machine to start learning. It’s like square one. If there’s nothing there, if you can’t explain it, that’s a problem.

One of my optimistic prophecies is based on the assumption that machines could have the best algorithms in the universe, but it will never have purpose. And the problem for us to explain purpose to a machine is because we don’t know what our purpose is. We have the purpose, but we still … When we look at this global picture, a universal picture, to understand what is our purpose being here on this planet? We don’t know. So that means we’re still searching, and will not be able to pass this message to the machine. And it’s a problem for us, some kind of comfort, though. People say it’s more like preaching … OK, maybe preaching, yes.

Because, as I mentioned in the book and all my lectures, is that people’s minds are polluted by these dark pictures of the future from Hollywood: “The Terminator,” the Skynet, “The Matrix.” It’s world where there’s no room for humans, or they have to fight against the machines. I think it’s just a way, way, way, way in the future. Is it going to happen? I don’t know. For me, these debates are not similar, but they resemble debates about how the sun will turn into a supernova in 4 to 5 billion years. Frankly, I don’t care. [Laughs]

Holodny: How does it feel playing against AI where you don’t perceive any emotional changes during the game — meaning the psychological element of the game doesn’t exist — versus a person? For example, you versus Deep Blue.

Kasparov: That was quite an experience, and I’ve been playing machines for many years. In the book, I started describing the story in Hamburg, in 1985 … and I’m still trying to figure it out whether it’s my curse or my blessing that when I became world champion, machines were weak — just a laughing stock — and when I left chess in 2005, machines were unbeatable. So I didn’t just witness that [but] I was an active part of this process. And, in fact, after matches with Deep Blue, I played two more matches with other programs, with a German program and an Israeli program in 2003, and both matches ended in a draw.

To sum up objectively, I think I was still stronger maybe the next year. If IBM didn’t retire the machine and we played, I think I had a chance of winning. But from the historical perspective, it was immaterial. I can go even further saying that since the machine managed to win the first game in the Philadelphia match [1996] — the match that I won eventually — in my view that’s the bigger milestone than even 1997, because if the machine was able to win one game, the rest is a matter of time. One year, two years, five years … but we were there. We were on that road. So that eventually the machine will be able to win the match. So it was clear that the machine had reached the level where beating the strongest players would be just a matter of a couple more decimal places in speed and a few more smart ideas for algorithms.

But going back to this match, I’m leaning toward blessing, because it doesn’t happen often that you’re part of history. So even if it’s not one of the most comfortable parts, it’s still history. And I think, you know, what’s happened there is we helped chess players — and the game of chess — to understand and to test many new ideas.

It’s interesting that the greatest minds of computer science, the founding fathers, like Alan Turing and Claude Shannon and Norbert Wiener, they all looked at chess as the ultimate test. So they thought, “Oh, if a machine can play chess, and beat strong players, set aside a world champion, that would be the sign of a dawn of the AI era.” With all due respect, they were wrong. It’s an important step forward, but we’re still, still far away, and that’s why I think it’s the best lesson from this match and from the game of chess is that we could see much clearer how humans and machines can cooperate because that’s the way to move forward. And I’m always saying that it’s for us to find new challenges.

So, somehow, AI is playing an important role of breaking up the ice of complacency. We have a comfortable life, we just don’t want to take risks. AI is threatening too many comfortable jobs to make people think about taking risks again.

Holodny: You argue that machines and humans working together is a stronger combination, but you also argue at the end of your book more generally that machines can make humans more human. Can you explain your thinking here?

Kasparov: It’s an interesting debate. I was part of this debate at the University of Toronto about six months ago. So, there was this Oxford-style debate. The house motion was: “Will machines make us more human or not?” I defended the house motion — just, it’s what it was. And I have to say it was very close. It started, I think, at something like 51-49, in favor, and we lost, 49-51.

But it’s amazing, the paradox of this debate: A very good debater on the other side played Trump on us. Because, you don’t do it in debates; you don’t start discussing the definition of the house motion [and] that’s what he did, which is just a violation of the rules. And you don’t attack opponents, but that’s another story. But it’s amazing that we lost the debate because he used very human arguments.

It’s quite funny, but it’s also interesting because I was learning from this process, and my argument was that, look, if we keep an optimistic view about the future, we definitely have to look for our place. Not to be redundant. And that means we have to emphasize what makes us unique.

As I said, everything we know how to do, machines will do better. So what about other things?

Machines taking over jobs — it’s the history of civilization. Replacing farm animals, old forms of manual labor, now taking over small, menial aspects of cognition. But there’s still plenty of room for creativity, for curiosity — many things that are related to passion, like art. But also, things about human communication and challenges, massive challenges that we left behind because we didn’t want to take so much risk, such as space exploration, deep ocean exploration.

For us to make sure that we have full lives, meaningful lives, we will have to elevate our unique human qualities. And I think we know now, we can see clearly what makes us different from machines. And that’s why the future is enhanced humanity.

The opposite argument is that we will end up being cyborgs. It’s an interesting argument and it’s amazing: Every debate moves from technology to philosophy. Now, we’ll use some kind of devices that we’ll implant — we’ll see better, we’ll run faster, we’ll lift more weight. But it doesn’t change humanity — it’s like taking a drug. So this is doping. This is a form of electronic doping, but it doesn’t change you. Even if you have a few implants that are making you faster, you are still the same human.

Now, the idea that you can remove intelligence from a human body and put it somewhere else — that’s an interesting question. But this is something that is far from being understood. Because then the natural question is, can we imagine our intelligence, our brains functioning outside of our bodies? Is it all connected? How does it work? Because it’s not just simply brain, but it’s also the way we move, and — that’s what makes us different — also the passion and the character.

So there are many things that you cannot break down in numbers. And as long as we don’t know how to describe the way our intelligence functions, the fear that human bodies will become irrelevant in the process — this fear is a gross exaggeration.

So that’s why for me, I believe more than ever, machines will put new challenges, and that means we’ll have to be more creative and more human, because that’s the way to make the difference.

Holodny: OK, just some fun questions for the end. What do you think is the biggest misconception about chess?

Kasparov: I’m afraid the misconception is very much based on the image projected by some of the great books, like Nabokov’s “The Luzhin Defence” or Stefan Zweig’s “Chess Story” — it’s about chess players being kind of nerds. Just playing in the dark corner of a café and just being geniuses, but totally removed from the real world. Yes, we have cases in our game — Paul Morphy, Bobby Fischer. But when you look at the numbers of chess players who were detached from reality and compare to the general, chess is a totally sane game.

This misconception has been gradually removed because more and more kids are playing chess. But still it’s very much alive. The game — it’s a nonmainstream game. And the irony is that when you look at Hollywood, it kept using chess as the symbol of intelligence for its heroes, for its top characters, all the time. So it’s from “Casablanca” to “Harry Potter.” You always have chess as a very important element to demonstrate intelligence, while in normal life people think it’s just a weird intelligence — like AI. [Laughs]

Holodny: What advice would you give to young chess players who want to play seriously?

Kasparov: Look, I would say, you have a unique chance of learning more about the game of chess with your computer than Bobby Fischer, or even myself, could manage throughout our entire lives. What is very important is that you will use this power productively and you will not be hijacked by the computer screen. Always keep your personality intact.

Remember that the machine is there to help you, because at the end of the day, you’re not playing freestyle chess, advanced chess, human-plus-machine. If you are playing against other humans, it’s about winning the game. The machine will not be assisting you, unless you are cheating of course. [Laughs] And since the machine is not there, you have to make sure that everything you learn from the computer will not badly affect the way you play the real game.

Holodny: What separates a good chess player from a great one? Or a grand master from a world champion?

Kasparov: Oh it’s a tiny difference.

Holodny: Is it something you perceive as a player?

Kasparov: Somehow it’s your ability to operate different kinds of positions. If you want, again, to oversimplify: A very strong player can manage and can just know how to manage a thousand positions. I get it; it’s a very arbitrary number. So then you have the world champion who could do more. But, again, any increase in numbers creates, sort of, a new level of playing.

And then you go to the very top, and the difference is so minimal, but it does exist. So even a few players who never became world champion, like Vassily Ivanchuk, for instance, I think they belong to the same category.

I would say that when you go to the very, very top, it’s an ability to come up with new ideas, with something new, to make the difference. Every world champion, every player who was traversing this universe, managed to bring something new to the game. This ability to always find some unconventional ways. That makes the final difference.

Holodny: Is chess beautiful to you?

Kasparov: Oh absolutely. It’s endlessly beautiful. I haven’t stopped enjoying great games, and when I see a nice endgame, for instance, especially when there are very few pieces left. Oh, I can’t help think, “Wow, how beautiful it is.”

And now, because of computers, we have a new technique of composing studies, endgame studies. They look at the positions from the large databases, and all pieces’ positions are already calculated by the machine. More than 100 terabytes of information. And then they create studies that lead to these positions. Sometimes it’s really beautiful.

So there are so many great things that you can discover. As I said, the number of legal moves is infinite, but if you find this, the little jewels in the tons of garbage.

Holodny: It’s interesting that this is maybe the one thing machines actually can’t do.

Kasparov: But again, you need humans to actually look for the jewels.

Holodny: Exactly.

Kasparov: To understand how beautiful it is. The geometry — it’s just amazing. You know, if I’m in a bad mood, I always look at the chessboard, just to find something that can cheer me up.

Holodny: So I guess it gives you purpose?

Kasparov: [Laughs] Yes.

 

Garry’s Timeline

Follow Garry's extraordinary path through years of relentless activism.

View the full Biography