Episode 3: Building Human-AI Collaborations with Garry Kasparov
Watch
Listen
Follow
Transcript
The Evolution of Chess and Machines
Sean Merat
Well, Mr. Kasparov, thank you so much for joining us and spending the morning. It's just going to be a short discussion.
Going back to 1997, your match with Deep Blue, of course it was, you know, everyone tends to talk about the second match, but I'd just really love to understand, what was your mindset at the time and how has it evolved in the last almost 30 years?
Garry Kasparov
Yes, thank you for reminding our audience that it was the second match, the rematch, because I won the first one in 1996 in Philadelphia.
But the history of human-machine competition in chess started early. Of course, for a long time, machines were not just a good match to top players. But in 1989, I had a first experience playing Deep Thought, which was a prototype that later became Deep Blue. It was developed by the young scientists in Carnegie Mellon University, later purchased by IBM, the whole project, and turned into this. It's a mega project of parallel processors, that as one of the subproducts, had Deep Blue. Deep Blue was not a chess-playing…just a machine only. It was a part of these parallel processors, which was a pretty successful project itself.
In 1989 I won handedly, but since the beginning of the '90s, many chess players already experienced some problems with chess engines, because we have some strong chess engines, the German program Fritz, you know, one of the first, you know, in the field. In 1994, we played World Blitz Championship, which is five-minute chess, and Intel Blitz Championship, and I won the event but we tied with Fritz, so then I won the tiebreak match.
But the fact is that it was a tie and the machine did so well against the best players in Blitz, that was already the first signal. And then we played rapid chess, 25-minute chess, and also many of us, myself included, suffered some painful defeats occasionally. But we were under the impression—totally wrong of course—that if we add more time, we would have more time just to avoid any blunders, just to control the flow of the game, ignoring the simple fact, we have more time, the machine has more time.
So the 1996 match that I won in Philadelphia was, you know, probably was a watershed moment because while I won the match, I lost game one in the match. So this is, it was a six-game match, I lost game one, then I won game two and game five and game six, so winning 4–2, with two draws in the middle. But the fact that machine could beat the sitting world champion in what we call classical chess, six hours plus, that was already like a sign on the wall, you know, in capital letters: "IT'S COMING."
The Shift from Competition to Collaboration
So the 1997 match somehow, you know, just was a shock for the general public. But analyzing now, just 28 years later, so the outcome, actually 30 years of this, you know, this relations, humans and machines, and also knowing that by 2005 the competition was literally over. We still could compete. The '97 match was not the end of the story. It took another eight years before machines totally dominated the field. I actually played in 2003, two more matches against all the versions of Deep Fritz and then Israeli program, Junior, both ended as a tie. But I knew back in '97–'98, the day was coming. So that's why it was already a moment for me to recognize that at the end of the competition, so stop competing, start collaborating, start looking for the best ways to bring together machines, brute force, and pattern recognition to human creativity and intuition. And that's how I came up with the idea of Advanced Chess: human plus machine versus another human plus machine.
Sean Merat
Right, the Kasparov Law.
Garry Kasparov
Kasparov Law came up later because we found out by making experiments, by playing online, so that it's in this competition, in advanced chess, human plus machine or another human or groups of humans or plus machine or groups of machines.
So the decisions, the quality of decision and the outcome depends not only on the quality of a human player, not even just talent, not on the speed of the machine, but on the synergy, on the superior process, so weaker player with slower machine could beat stronger player with a faster machine if he or she will have better collaboration, because Kasparov at the end of the day, it's about finding the best formula that will increase the output.
The Future of Chess: Human-Machine Collaboration
Sean Merat
Right. So the process itself, how would you think about chess at the time, and you had the foresight to think about this collaboration outperforming the…so how did you see the future in collaboration in chess?
Garry Kasparov
Look, chess has changed dramatically, because today—we all know that even Magnus Carlsen, the strongest player, he's no match for a computer.
Sean Merat
He would probably lose to a….
Garry Kasparov
Maybe he could compete, maybe. But, still, it's unlikely.
But when you look at the chess engine at your laptop, the difference between this, you know, "chess player," silicon chess player, and Magnus Carlsen is about the same as between Usain Bolt and Ferrari. The first 50 meters they run, you know, just on par, then, just you know, it's over.
But machines played a very important role in changing our professional preparation. Because when you look at the games today, many of the traditional openings that have been played when I was active, they have gone, because you don't want to take the risk, because it's not about, you know, just trying to prepare some very sharp lines, or this slight, slight, you know, sleep, just not recognizing, you know, what can be done by the opponent who has a machine and is more agile in using the computer. So you could be dead just without making a single move in a real game; it could be decided by the machine's preparation. So it's very logical that top players, they avoid very sharp lines.
We had a few interesting duels five, six years ago, but it became very apparent that it's too risky because preparation basically decides everything. So they shifted to more, like, we call positional chess, so when the main battle is pushed from the opening to the middle game, so you can slowly develop your pieces and create a situation where machine preparation is not going to solve everything.
Some people are just complaining, saying, look, it's not the same excitement when Kasparov played Karpov, so we just had, this, all this fun. But I can see the positive side, because with machines we always have both negatives and positives. And somehow it's a bit odd for me, just, you know, as a former World Chess Champion, to see that my colleagues, you know, just playing the game and you can see on the screen that they're not doing the best moves, and every amateur on the planet now can see that Magnus Carlsen is is not playing as well as machine recommends.
The Impact of Technology on Chess
But the fact is that you can have, now, millions of people watching the game without waiting for a special commentary from a grandmaster, basically having an understanding. That's a big win for chess. It's, computers, yeah, they somehow downgraded the glory of the top players in the eyes of the beholden—but on the other side, they brought millions into the game because now you can follow, and you can see instantly whether Magnus or just another top player is doing well or not. So you are engaged, so you don't have a special preparation or just, you know, the special commentary to follow the match.
Sean Merat
Yeah. I think, I love, I love your framing, that, you know, the ideal engine, because it's a finite number of moves, regardless of how big it is, the machine, the engine, the person who makes less mistakes will eventually….
Garry Kasparov
But you said, you said, finite number of moves. Okay. Technically, any number is a finite number. So, this number, number of legal moves in the game of chess, contains 46 zeros, according to Claude Shannon. I don't think we can call it the number that, you know, makes any sense to us.
So yeah, it's, it's finite on paper, but for us, it's not, it's not something that, you know, could have any practical, even in the remote future, practical implications. Because in chess, we have now, all the positions with seven pieces, seven pieces, positions, being resolved. It's the database, I think, it's over 100 terabytes. That's the—but that's seven, seven pieces, just seven pieces. Now imagine that, this is the game with ultimate endgame with 32 pieces. I don't think we ever crossed to eight, because it's….
Sean Merat
Because it's exponential.
Garry Kasparov
Well exponential, just beyond our, it's even with all the powers we have. So, again, it's, chess tells you that the future of this collaboration, recognizing that you don't expect machine to solve all the problems. You expect machine to make fewer mistakes, to limit your mistakes, and also to offer you immense power, but that's for you how to guide this power.
This is, this is the biggest, you know, advantage that we may have, is that it's too—competitive advantage—if you know how to harness this power. Imagine you have a tsunami, and you can just change it a little bit so that, it will, the destruction it brings, you know, will change. So here, is, just you know, you have this immense power, and slightly shifting it, you know, just, it's like, a bullet in the barrel of a very powerful rifle, so it could have a tremendous outcome when this bullet hits the target two miles away.
Harnessing AI for a Competitive Advantage
So, again, recognizing that, A) we do belong to less, fewer decimal places. Machines can do 95% of tasks better than us. Recognize that—that's key. Psychologically, don't compete with machines where, you know, we are already inferior. And B) find the area where human contribution for this specific task could be absolutely unique. That's where you have your competitive advantage, because, again, this, there's so many areas where the machine's immense power could do a lot of good, providing it has proper guidance.
Sean Merat
Proper process, that you….
Garry Kasparov
Well, again, it's about superior process. It's about recognizing where your human creativity, your human intuition, your human experience could actually, could help machine to, again, not to compete with its recommendation but basically, slightly shifting it in the direction you want it to go.
Sean Merat
Correct. Which is immense power by itself.
Garry Kasparov
Absolutely. This is recognizing that it's, again, the slightest tweak, you know, could have a tremendous result. Again, it's, it's, I would say it's a very creative process, because, yes, we have similar machines, we have similar tasks, but they are all not the same, and by the way, we are also different. So, this is again, recognizing, you know, what kind of contribution I can make for this task. And also, at what point I want to get in.
This is, again, even timing, could, could mean everything, because, it's, it's so easy, just, just to be mesmerized by the computer. You look at the screen, they come and come, and come, and come, and come again. Yet—stop! The moment you stop and, you know, just say, "okay, now I push the button." Just imagine the difference. You know, even if you do it, you know, two seconds later or two seconds earlier, that could have, could just have tremendous implications. Again, the human contribution, no matter how small, could still be a decisive factor in a very competing field, whether it's insurance or any other area where AI is, has been playing already, a major role.
Challenges in the Insurance Industry
Sean Merat
Yeah, I couldn't agree more.
I think it's, it's, you know, the applicability to insurance, it's, it's really, it's so easy for executives, I find, to really buy into the hype, trying to find the solution that solves all problems. Whereas, I agree with you that the foundation may be flawed if you're not thinking about the process.
So on that note, and I know we're coming at about time, but, what would be your suggestion to these executives to implement these models accordingly, so that, so that their really focus remains on the process itself, not so much on solving everything?
Garry Kasparov
Again, every industry has its own problems and challenges, and obviously the solutions, creative solutions coming from AI or AI collaboration with experienced humans, you know, differs, differ.
But, specifically in insurance, I think that's one of the interesting challenges, will be how to adjust your rates to the change in probabilities. For instance, in car insurance, you have driverless cars; the moment you have enough driverless cars on the road, that are being programmed specifically, let's say the life of a passenger is number-one priority. So you will see the change in probabilities of the outcomes of the accidents.
So, this is an amazing, an amazing challenge. But if you recognize, the, it's earlier than others, again, you'll see, okay, enough cars on the road for me to start offering these new, new packages. So that's an advantage. And again, it's, it's not, you know, something you can find in the books. You basically have to create books, because, it's, it's, you're talking about immense probabilities, but again, it's all statistics, it's insurance, because you work with probabilities. And now, this is, this is recognizing when it changes and it quite dramatically changes because, again, it's the, it's uncertainty that always follows accidents on the road will be dramatically reduced because you already may know the, okay, not the, not 100%, but, you know, you can be quite confident about the outcome, because machines will be programmed a certain way, and maybe the owner will have a choice how to program it.
Effective Collaboration Between Humans and AI
But, again, great, great new challenge, virtually unlimited field of possibilities, and here we know you need the most effective collaboration between human, expert, and AI. So this is, it's a key. It's if you, if you try to rely on all the computers, you're on losing side. If you downplay the role of the computers, you're on the losing side. So you have to find the most effective collaboration between human and computers to constantly adjust your, in case of insurance, your rates to, to be ahead of the curve.
Sean Merat
Right. It's giving that tsunami a direction.
Garry Kasparov
Absolutely.
Sean Merat
Building that premise so that you can use the machines to fill in the blanks.
I recognize we're at time. I really appreciate you taking the time. It's an honor to speak with you and thank you again.
Garry Kasparov
Thank you. Thank you for having me.
Sean Merat
Thank you.