The Machine Has a Steady Hand: Garry Kasparov on Deep Blue, AlphaZero, and Putin

Lex Fridman asks a question Kasparov says he has never heard: looking inward at a career that dominated chess for two decades, what was the bigger motivator, the love of winning or the hatred of losing? Losing, Kasparov answers, was always painful, almost a physical pain, because he knew that if he lost it was because he had made a mistake. He always believed the result of the game had to be decided by the quality of his play. It sounds arrogant, he admits. It also helped him move forward, because there was always room for improvement.

Fear of a mistake, he says, actually guarantees mistakes. The difference at the very top is the ability to decide without predictable consequences. You are at a crossroads: right, left, straight, or turn back. You have some intuition about each road, never enough for a fully calculated choice. At that level chess is inner strength: I will make this decision, I will stand firm, I will not waste time looking over my shoulder. Returning to Fridman's question, he says neither love of winning nor hatred of losing was the driving force. Both were important psychological elements. The engine was a passion for making a difference — for creating something new in an opening, in a middlegame plan, in a game that had not existed before.

That same appetite, he says, is what let him leave the game. He knew he would not be number one in the next life. He could still be active in Russia's democratic movement, in writing, in talking to people about how humans decide. Influence would not be as decisive as it was over a chessboard. It could still be strong enough to help. Fridman pushes on demons. Kasparov will not mystify them. Even with Anatoly Karpov across the board, and with all respect for the opponent, he believed it was up to him to make the difference. He was not invincible. He made blunders, more of them with age. He still wanted to be the decisive factor. At fifty-six he had just been crushed by Fabiano Caruana at Chess960, Fischer Random, after holding winning positions he would have converted fifteen years earlier. He can fight an opponent, he told an interviewer, but not his biological clock. He still got Caruana almost on the ropes. His wife offered a joke worthy of the goddess of chess: maybe she did not want him to win, because if a retiree beat one of the world's top players, people would say the game had stopped progressing.

Asked for a moment of brilliance he is proud of, Kasparov starts with Mikhail Tal, the magician from Riga, then widens the frame. Every world champion was a creator. Some were flashy sacrificers; Tigran Petrosian was the opposite, slow and strangling. Amateurs love the games where pieces hang and the weaker side delivers a final blow. There is another kind of beauty: positional maneuvering, looking for weaknesses, gradually squeezing. Kasparov thinks he made more difference in the game than he imagined when he started, and he left when he knew he could no longer bring the same kind of new knowledge to it.

He does not lead with the famous 1999 game against Veselin Topalov, where his king walked from its own camp across the board. Amateurs pick that one because it is rare and visual. He goes to 1985 against Karpov. Game 24 was the decisive game of the match; he won and became world champion at twenty-two. What he wants remembered is a unique maneuver found in a short game, a move that looked ugly on display and is now typical. Game 16, with Black, paralyzed Karpov's army in its own camp. The score was equal; under the rules the champion retained the title in a tie, and nine games remained. It was also a bluff. Neither he nor Karpov saw the refutation of the opening idea. Kasparov played impeccably after it. Then they discovered it did not stand up. He did not know he was bluffing. That, he says, is the difference between that era and this one. Today a machine shows you the refutation in a split second.

Analysis in the 1980s and 1990s was imperfect because the analysts were human. The hours still counted. You could spend a night with coaches, find nothing that appeared on the board, and somehow those hours helped. He reaches for a phrase and then qualifies it: the work seemed to transform into a kind of spiritual energy — not religious, just the only words he has — an invisible connection between dedication in the study and brilliance with the clock ticking. Even when there was no direct line from the preparation to the win, the connection was there.

Where does Magnus Carlsen sit among the all-time greats? Kasparov calls the ranking unfair. Each new generation knows much more about the game than the last. People who say old Kasparov, Fischer, or Magnus was the greatest disregard the fact that the great players of the past, by today's standards, knew so little. Any child who spends a few years with a chess computer knows more than a former world champion did, simply by access. A better measure is the gap between the champion and the rest of the field, plus time. By gap, Fischer might be on top, for a short period. Kasparov was on top longer, and unlike Fischer he beat the next generation. Vladimir Kramnik was already twelve years younger. At forty-one, when Kasparov left, he was still dominant.

Carlsen's consistency, he says, is phenomenal. The reason he seems unbeatable is that he is a lethal combination of Fischer and Karpov, which is unusual. Fischer's style was dynamic, fighting to the last resource. Karpov used every piece with maximum effect from minimum means. Carlsen squeezes every stone for a drop of water and can do it for as many hours as Fischer, with the same meticulous deadliness as Karpov. Physical condition matters. Studies show chess players burn thousands of calories in a game. As for whether peak Kasparov of the late 1980s would beat Carlsen in 2019, he calls the question irrelevant. Kasparov in 1989 played great chess and still knew very little compared with Carlsen in 2019 — who, Kasparov notes, learned from him as well. Empower 1989 Kasparov with thirty years of later knowledge and you no longer have Kasparov. You have someone else.

He is just as cautious with soccer. Fans pick Messi because they are younger and did not see Maradona. Kasparov saw them. If he has to choose, he goes with Maradona. Brazil in 1970 could have won without Pelé and still been great. Argentina in 1986 without Maradona would have been nothing. Messi, at the time of the conversation, still had not won the World Cup. Ratings have their own inflation: when Kasparov crossed 2800 in 1990 there was one other player in that category, Karpov; now there are more than fifty. His 2851, he suspects, is as valuable as Carlsen's peak 2882. Then he waves the hypotheticals away.

Fridman calls the loss to IBM's Deep Blue in May 1997 one of the seminal moments in civilization: centuries of treating chess as the peak of intellectual mastery, then a machine beating the human champion, inspiring a generation of AI researchers including Fridman himself. On Kasparov's face, at the time, it looked like a tragedy. Why so painful, if the moment was that large? Kasparov repeats the earlier point. Losing was physically painful. Then he corrects a detail the culture has been sloppy about. The 1997 match was not the first time he lost to a machine. It was the first match he lost. He had lost games. He had lost rapid events. He had lost a short match. The nature of a full formal match was different.

He was angry because he suspected the loss was not only the result of his bad play — though, looking at the games today, he made tons of mistakes. He believed other factors that had nothing to do with chess were in play. Twenty-two years on, he is willing to analyze the match. He agrees with almost everything Fridman said, with one exception: considering chess the pinnacle of intellectual activity was our mistake. It is a game. All a machine has to do in this game is make fewer mistakes, not solve it. According to Claude Shannon, the number of legal positions is 10 to the 46th, too many zeros for any computer to finish the job in the next few billion years. It does not have to. It just has to be more consistent than we are.

What followed in Go, shogi, and other closed games was the same demonstration. The moment you build a closed system — rules clear, state space bounded, the goal defined — machines will prevail, not by being more intelligent but by bringing down the number of mistakes. Compare today's engine ratings, around 3400 to 3500, with Carlsen in the 2800s, and the gap by chess standards is insane, like the gap between Carlsen and an ordinary player in an open international tournament. It is not that the engine understands chess better. It has a steady hand. Ferrari versus a family car, Kasparov says. We thought we could compete with computers in intellectual fields. That idea, he argues, was wrong from the beginning.

1997 was not even the first machine victory over grandmasters. He played decent engines from the late 1980s, including Deep Thought, the Deep Blue prototype, in two rapid games in New York in 1989, and won both handily. He lost some games and a 1994 rapid match. He lost one game in the 1996 Deep Blue match and won the match, a fact people tend to forget. The psychological mistake was thinking they lost blitz and rapid only because they did not have enough time, and that a longer match would erase the errors. Nonsense. Humans got more time; the machine got more time too, and the machine is always steadier. Today nobody talks about humans playing machines as equals. Engines can give handicaps to top players and still be favorite.

What he recognized in 1998, after a year licking his wounds, is that it is no longer humans versus machines. It is humans working with machines. He played two more matches in 2003, against Deep Fritz and Deep Junior; both ended as ties, and those machines were at least as strong as Deep Blue. A chess app on a phone today is probably stronger than 1997 Deep Blue. Ask Stockfish, Houdini, or Komodo to look at the games trumpeted as brilliant in that match and within thirty seconds they will show how many mistakes both Kasparov and Deep Blue made. The century's contest, under a modern engine, is full of holes.

Fridman asks what remains out of reach, and whether language and conversation will fall the way chess did, with the same learning approaches, as machines grow. Kasparov says it is not about growing in size. It is about closed versus open-ended systems. In an open world, machines are not lost, but their effectiveness is different, because a machine does not understand when it has reached diminishing returns. It can ask questions. It will never tell you which questions are relevant. Human territory may shrink to the last few decimal points. He offers an image: a very powerful gun, and all you can do is alter the direction of the bullet by 0.1 of a degree. A mile away, that is ten meters. Unique human qualities, in the foreseeable future, will not be reproduced. The danger is interfering with the machine in the 95 percent where it already knows how to do better. In radiology he would sometimes rather have an experienced nurse than a professor, because she will not try to override the model's reading.

He splits what gets called AI into two kinds. Most of it, including Deep Blue and a great deal of modern deep learning, is a variation of what Claude Shannon characterized as Type A, brute force: optimization, improving the processing of human-generated data. AlphaZero is a first step toward machine-produced knowledge. The irony, he says, is that IBM got there first, in backgammon. Gerald Tesauro's TD-Gammon in the early 1990s was an AlphaZero-type program, discovering strategy by playing itself. Deep Blue's fame put that project on hold. Kasparov has looked at AlphaZero's chess with excitement: broken material balances, sacrifices, staying a move or two ahead of other engines not by calculating harder but by patterns distilled from some sixty million self-play games. It feels almost like intuition.

Then the weakness. He has visited the lab and spoken with Demis Hassabis's team. Suppose AlphaZero faces another powerful computer accompanied by a human who has found a blind spot. How many games to correct it? Hundreds of thousands, even while it keeps losing, because the learning is statistical. A human can make a tweak from a single hint. Humans are still more flexible. The collaboration works only if we are honest about which part we are for.

Morality and machines, Fridman says, is the thing they certainly do not know how to do yet. Kasparov treats the panic about biased models as a phantom being fed to the public. How do we avoid machines having prejudice? You cannot, he says, because it is like looking in the mirror and complaining. If the society has a bias, the machine will follow it. You can break the mirror, distort it, or change yourself. You cannot expect machines to cure the ills of the society. They will amplify them. People are more comfortable with other people doing injustice than with a machine having the same bias. That is the interesting standard we place on automated systems.

About forty thousand people a year die in car accidents in the United States, almost all from human error, and those deaths are statistics. One accident with an autonomous vehicle is the front page, even when the fault is the human's. No machine will ever be 100 percent perfect. That is another story being fed to the public: if it is not perfect, it is not safe. All you can ask of a computer, whether it is playing chess, doing market calculations, or driving, is to make fewer mistakes. The double standard has a cost. Treat every autonomous-vehicle crash as intolerable and you get fewer miles of testing and, he implies, more of the deaths we already find ordinary. Fridman asks whether Kasparov treated Deep Blue differently from a human opponent. At the time, Kasparov says, it was a match, and he was angry because he believed it was not fairly organized. The displeasure was aimed at the humans behind IBM. He knew he was objectively stronger then, which added to the anger. He knew he could beat the machine. He was not well prepared. He underestimated the publicity. He made too many mistakes.

Fridman, also born in the Soviet Union, asks what lessons Kasparov draws from its rise and fall. The lesson, Kasparov says, is clear. Undemocratic systems, totalitarian regimes that control every aspect of citizens' lives, that offer no room for private initiative, central planning — they cannot be a driving force for innovation. They can distort the idea of progress and even call themselves progressive. The damage they cost humanity is yet to be measured. At the end of the day they fail. The end of the Cold War was a triumph of free will, which is not the same as saying the free world is perfect.

One of his favorite books is *The Lord of the Rings*. He reads from it a political claim: there is no absolute good, but there is absolute evil. Good comes in many forms, and you can always find a spot on the sun. Human history is a never-ending quest for perfection, and we know the other pole when we see it. Nobody argues about Hitler being absolute evil. He thinks it is telling that Stalin is not discussed with the same standard. Communism, he says, caused more damage than any other ideology in the twentieth century. There was no Nuremberg for communism, and that is why Stalin's successors still feel comfortable.

He had just given a speech in Toronto on the Molotov-Ribbentrop Pact, something he believes should be taught in schools. World War II, in this telling, was started by two dictators signing a criminal collusion in August 1939. The fact that Stalin later had no choice but to join the Allies, because Hitler attacked him, does not erase that he helped Hitler start the war, or that he was an early beneficiary, annexing parts of Eastern Europe, and after the war almost the entire region. For many Eastern European nations, the end of the war was the beginning of communist occupation.

Kasparov has been a vocal opponent of Vladimir Putin. Fridman quotes Oleg Kalugin, a retired KGB general, from 2007: people who knew the details and talked are dead; Kasparov is vocal, and he may be next on the list. How does fear enter the mind when the work can get you killed? That, Kasparov says, is the reason he lives in New York. It was not his first choice. He knew he had to leave Russia at some point, and among the options New York was the safest. Is it safe? He knows what has happened to many of Putin's enemies. What can he do? He can try to change what he can influence. He cannot stop doing what he has been doing for a long time. It is the right thing to do. He grew up with a dissident's wisdom: do what you must, then accept the consequences.

He can be cautious. He turns down invitations to countries that are no longer destinations he can afford to travel. His mother still lives in Moscow; he sees her a few times a year. She was devastated when he left. His father died in 1971, when she was thirty-three, and she dedicated her life to her only son. In a year or so she recognized that leaving was the only chance for him to continue a normal life, to be relatively safe, and to do what she had taught him to do: make a difference. Will he return? Sooner than many people think, he says. Putin's regime is facing insurmountable difficulties. He has read enough history to know how dictatorships end. On Sunday the dictator feels comfortable and believes he is popular. On Monday morning, bust. The bad news is that Kasparov does not know when or how Putin will go. The good news is that Putin does not know either.

Did Russia interfere in the 2016 U.S. election, and will it in 2020? Kasparov jokes that the question is insulting his intellect after an intelligent conversation. Of course they interfered. They did everything they could to elect Donald Trump. He has said so many times. He has met enough KGB colonels in his life to tell you that the way Putin looks at Trump is the way KGB officers looked at their assets. He does not need to hear the words or read a congressional investigation. For 2020, of course they will try to help Trump survive. The damage Trump's relations could cause to America and the free world is, he says, beyond imagination. If Trump is re-elected he will run further in an authoritarian direction; the re-election hurdle is still limiting him. Kasparov's deeper worry is not Trump the person. Trump is a symptom. He does not see, on the American political horizon, politicians who will take him on for the damage to the free world. The Democratic campaign, in his view, is fixed on important but secondary issues. When the framework of American political life is at risk, you cannot talk only about health care. There is already evidence of Russian interference in Brexit and in elections across Europe. Thinking Moscow would be shy of 2020, with Trump in office, is not a serious thought.

Last question: if he could relive a single moment, which one? He will not pick. There are moments he thinks about doing differently. He made many mistakes. He also believes in the butterfly effect. If in 1989 or 1993 he had not done the stupid thing, would he have all the other accomplishments? He does not know. He cites Forrest Gump: life is a box of chocolates, you go one by one. He is happy with who he is and where he is. He is proud not only of the chess but of the transition — a reputation built after he left the game, with some influence on chess and not derived directly from it. He married in 2005, his third marriage; he is close with two children from earlier ones, and has two more born in New York. Sometimes he wishes he could limit the commitments that still take time and energy. Life is exciting. As long as he can feel that he has energy, strength, and passion to make a difference, he is happy.