The Brain Across the Table: Garry Kasparov vs. Deep Blue, 1997
“The idea was to create the highest level of chess ever played, a synthesis of the best of man and machine.”
— Garry Kasparov, 2010
Part 1: The Man
Kasparov was confident.
His mind drifted back twelve years, to 1985, to Hamburg. This was before USB cables, before Super Mario 64, before there were more emails being sent in the U.S. than postal mail. Kasparov was standing in the center of the circle, that circle composed of thirty-two computers, each one specifically designed to play chess, each sitting on a table next to its respective human creator. Kasparov played them — the computers, not the creators — in a simultaneous round-robin format: one move to one computer, then another to another, and so on. It took him 5 hours, but Kasparov left that event with 32 wins.
He had never lost to a computer. He had never lost, period.
Then he contemplated his bout the previous year, when he had first squared off against his opponent today: a 3,000-pound supercomputer, named Deep Blue by its creators. The match had been a tough one — in the first of the tournament-regulated six games, Deep Blue actually scratched out a victory against humanity’s reigning world chess champion. It was the first time a computer had beaten a World Chess Champion in a single game under tournament conditions.
But chess matches are marathons, not sprints. Kasparov knew this: he centered himself, and won the next game resoundingly. After coming to a draw in the following two games (each side received half points for these ties), he finished up the match with a pair of wins, bringing the final tally of the overall match to 4–2, in favor of the human.
Kasparov and the chess community breathed a sigh of relief, and Kasparov went about his life.
But here he was again, now 1997, having received a second invitation from IBM to face down Deep Blue — IBM jokingly called it “Deeper Blue” — in another highly publicized match. Kasparov didn’t know much about what IBM had done to improve Deep Blue. All he knew was that this was a machine specifically designed to defeat him, and IBM wasn’t going to back down until it did.
But what did IBM know about chess? Kasparov had been seriously studying the game since the age of 10, when he had been sent to the academy of world champion Mikhail Bottvinnik. At 12, he was the USSR’s youngest Junior Champion in history. Four years later he won the World Junior Championship as well. On April 13th, 1980 — his seventeenth birthday — he was awarded grandmaster status. And at 22, he defeated Anatoly Karpov to become the 13th World Chess Champion. Chess wasn’t a series of 1s and 0s to him, like it was to Deep Blue. It was strategy, creativity, calculation, skill.
As with the previous year, this match was being framed as a production, a performance. The media covered the event as an almost apocalyptic showdown of man versus machine. One reporter called it “a game with serious implications.” Newsweek ran a story with the headline “The Brain’s Last Stand.” Maurice Ashley, a fellow grandmaster covering the event, jokingly gave the introduction, “This is international chess match, I’m Maurice Ashley. The future of humanity is on the line. Now the weather.” In the tradition of most famous chess matches, the playing area — a room in the Equitable Center in midtown Manhattan — more resembled a theater, with about a dozen raised rows of seats designed to fit over a hundred people. These seats faced a stage, dressed up to mimic a living room or home office, complete with a half-full bookcase (also housing three golden mallard decoys), a blue-thread carpet, and a small tree tucked into the corner. A poster hung above the table declared the event in no uncertain terms: “Kasparov vs. Deep Blue: The Rematch.” The feeling of competition was palpable — this was Sarah Connor versus the Terminator, Dr. Bowman versus HAL 9000.
The first of the six-game match was simple, decisive. Kasparov developed a strong position and trapped the computer’s king on the most-right column of the board. This was the new-and-improved Deep Blue? It was being defeated more easily than the previous year. The computer was continuing to fight, however, apparently looking to force a draw. Kasparov knew he wasn’t perfect — even players at his caliber usually make a tactical mistake once every seventy-five turns — so he kept playing at the same intensity, hoping none of his moves would result in an error.
But then, Deep Blue’s forty-fourth turn came around. It had the option of an obvious move. Kasparov had a pawn two rows away from the end of the board (and with it, the ability to swap that pawn for a queen), but had left his king open to being checked by one of the computer’s rooks. Put the king in check, and it puts pressure on the grandmaster, meaning he has to waste moves that would otherwise be applied to advancing that pawn.
But Deep Blue didn’t make the obvious move. It moved its rook straight past the row that would have cornered Kasparov’s king, all the way to the end of the board. There was no advantage to this strategy, if it could even be called that. It was a waste of a turn, and it baffled Kasparov. The question repeated again and again in his mind: why would Deep Blue commit suicide like that? Computers aren’t supposed to make tactical errors.
That night, back in his hotel, Kasparov should have been celebrating his victory by resting his powerful brain. But he stayed up. Deep Blue had essentially thrown the game, but why? Did it know it was going to lose, and was trying to conceal its strategy for future games? Could Deep Blue actually be intelligent? Intelligent enough to play chess, that is. Or — were the IBM programmers simply trying to throw him off, programming in a terrible move to make Kasparov doubt his own abilities? Well, it wasn’t going to work. Kasparov had hubris, sure — but for good reason. He was the best in the world at the world’s most dignified game. No matter why Deep Blue had done what it did, Kasparov knew the pressures that were on him. He was a proxy for all of humanity. He wasn’t going to lose.
Game two. Kasparov decided to challenge the computer. There was a trick that he had used against computers in the past. He would set up his position in such a way that it gave the computer a choice — either it could advance its queen forcefully down-board, or it could swap pawns with Kasparov and open up the left side of the board. The latter was more advantageous tactically, but it was also a subtler move, and Kasparov knew that computers aren’t exactly great at nuance. He set the trap, hoping to entice Deep Blue’s queen out of its position.
The computer “thought” for a long moment, whirring through hundreds of millions of configurations in its processor, and then… swapped pawns with Kasparov.
The grandmaster was visibly shaken. He buried his head in his hands. This couldn’t be possible. Perhaps there was a human, another grandmaster, in another room, telling Deep Blue what to play. IBM just wanted publicity — the publicity of having won the match — and they would do anything to get it. He wouldn’t put it above them to cheat. These thoughts spun his head round and round, until — quite abruptly to all watching — he resigned ten moves later.
That night, as he did every night, he studied the day’s match. His consultant pointed out to him that he easily could’ve worked Deep Blue into a draw if his mind was clear enough to see it. If there are sins in chess, resigning a game you could have drawn is one of them. Kasparov shook his head, staring out the window at Fifth Avenue. What in the world was going on? Garry Kasparov never beat Deep Blue again. He cranked out three consecutive draws with the computer, until a sound defeat in round six. The final score: 3 1⁄2 to 2 1⁄2, in favor of Deep Blue.
Apocalyptic before the match, the media reaction was grim afterwards. “We humans are trying to figure out our next move,” Dan Rather said on air. And then there was the clip of Kasparov himself, sweating on the microphones in his face:
“I have to apologize again. I am ashamed by what I did at the end of this match…”
Part 2: The Machine
Feng-hsiung Hsu knew IBM’s advantages were considerable.
Not only had exploding technologies meant that the company was able to literally double Deep Blue’s processing power and deepen its search algorithm over the past year, but losing the 1996 match against Kasparov meant that they had even more of the World Champion’s matches to load into the computer, and even more concrete examples of his style to overcome. And beyond all of that, Hsu and his team could and would adjust Deep Blue’s programming in-between each of the six matches — the machine would adapt to Kasparov faster than he could adapt to it. Then you have the things that all computers do better than all humans anyway — the brute processing speed, the ability to do more in less time.
This was a computer that took up a room. It was a machine designed to do one thing — beat Garry Kasparov in chess. And, by God, it was going to do that.
Hsu was 38, a Taiwanese immigrant in his first — and so far, only — project at IBM Research. He had never played chess competitively, but that didn’t stop him when he and two fellow doctoral candidates at Carnegie Mellon decided to build a chess-playing computer. This was 1985, the year Kasparov defeated 32 such computers simultaneously in Hamburg — and the year “Chiptest” was born by Hsu and his buddies. Five years later, the whole trio was hired by IBM to continue work on their computer, which eventually blossomed into Deep Blue.
Chess was a frontier for early computer scientists, Hsu knew that. Early artificial intelligence researchers — in the ’40s and the ’50s — had an impression that chess was something distinctly, innately human. To create a computer that could play chess was in some way, then, a milestone — proof that computers could be innately human, too. Famed British mathematician Alan Turing wrote a chess-playing program in 1948 — before there was a computer that could run it. But Hsu wanted to go one step further. He didn’t just want his computer to play chess. He didn’t just want his computer to beat any random novice. He knew that until his computer could defeat a grandmaster — until it was better at chess than the best human chess player in the world — it wouldn’t be taken seriously. And it had been a twelve-year journey, but here Hsu was, watching his creation go toe-to-toe with the world’s greatest chess player for a second time.
As someone both who had picked up quite the eye for chess and who understood Deep Blue’s inner workings better than anyone, Hsu knew that first match was going downhill. Kasparov’s position was strong, and everyone in the audience knew the computer was eventually going to lose.
Then, that forty-fourth turn.
Deep Blue, unlike its opponent, never resigned on its own. What Kasparov didn’t know was that the computer’s prerogative was always to prolong a game as long as possible. In this case, the move that would work towards that result was obvious to Hsu just as it was to Kasparov — put Kasparov’s king in check. When Deep Blue instead committed suicide, perhaps the only person more surprised than Kasparov was Hsu. But while Kasparov was thrown by the move as a shrewd and sly show of intelligence, Hsu was thrown for another reason — it was a mistake, a mistake Deep Blue wasn’t supposed to make. Hsu immediately came forward and resigned the game for the machine — Deep Blue obviously couldn’t make that decision for itself, and it was close to over anyway, Hsu figured. Mostly, he was caught up by the mistake — because computer scientists have another word for “mistake.”
That night, while Kasparov and his consultant evaluated the game, Hsu and his team did the same thing. They looked at the configuration the board was in at the time of the blunder, and poured over the code. And then they looked up at each other, the minute what they all suspected was confirmed.
It was a bug.
That forty-fourth move was not a tactical error, nor Deep Blue playing its cards close to its proverbial vest. It was a bug. While cycling through configurations, the machine was simply unable to find the best possible option. When it reached a certain point, an error in the code made the machine break its loop and pick a move at random. That was it. No strategy, no long-game. Just a few lines of code leading to a random selection. In fact, Hsu recognized it. A few months earlier, they had been training the machine on another chess player and the same bug led to other terrible, seemingly random moves. They thought they had fixed the code, but it looks like they had missed one of the edge-cases.
No big deal, though. Like Kasparov, Hsu knew chess matches were marathons, not sprints. They had lost one game, but they would be back in a few days with another match, and Deep Blue would be even stronger than it was before. The next day, while Kasparov had something akin to an existential crisis because of the machine, Hsu and his team patched up Deep Blue’s bug in a few hours.
A week later, it was game six. Hsu watched Kasparov, who was getting more and more flustered by the moment. Due to a failed gambit, Kasparov allowed Deep Blue to commit a knight sacrifice, and Kasparov’s defense didn’t stand a chance. After just 19 moves, Hsu looked on as Kasparov limply shook the hand of the computer’s operator, flung himself out of his chair, and stormed away, hands in the air. It was over. Hsu wished that Kasparov had even acknowledged the computer — the operator was just moving the pieces after all — but it seemed Kasparov couldn’t stand to look at it. After the victory, IBM quickly dismantled Deep Blue, lest Kasparov demand another rematch or accuse the company of cheating. In 1999, as the Deep Blue project officially wound down, Hsu left IBM — his work at the company now complete. Later, much of Hsu’s ideas and technology that powered Deep Blue went into IBM’s next major project: their famous Jeopardy-champion computer Watson.
Like everyone else who had been invested in the Kasparov match, Hsu saw the headlines that emerged at the time, portending the dark future of human inferiority to the machines. He read them, but was able to shrug them off easily. He knew one thing: that Deep Blue’s victory was a triumph, a triumph without precedent — and, for computer scientists, that whatever dark future the media was predicting was actually quite bright.
As for the second game, when Deep Blue refused to be enticed into Kasparov’s trap? Hsu knew the truth behind that as well:
Deep Blue was really, really good at playing chess.
Part 3: Advanced Chess
Kasparov stared at his computer.
Then at his opponent, Bulgarian grandmaster Veselin Topalov. Then back at his computer, before making his move.
He was proud to have his brainchild seeing the light of day. The months after the loss to Deep Blue were complicated, to say the least. He had a hunch that IBM had cheated on that second game, the one where the machine avoided his trap. No, it was more than a hunch — it was only a hunch because he couldn’t know for sure. In fact, he was so confident that he had publicly accused IBM of cheating, but they had already dismantled Deep Blue and were refusing to hand over any of the computer’s data. One thing was indisputable, however. During the match, IBM had access to all of Kasparov’s former games, but they declined to provide any information on Deep Blue’s gameplay to Kasparov. No one had acknowledged this.
Fine, Kasparov thought to himself through the resigned anger. He could lick his wounds, but he was still the number one chess player in the world. IBM had their moment and got their publicity. In the days following Kasparov’s loss, IBM stock had risen 3.6%, the highest it had been in ten years, and just short of the company’s all-time record.
That’s what IBM wanted. Not progress, not knowledge. Money. Kasparov saw that now.
He had gone off the map for the rest of 1997. That week in New York City had provided him with enough media exposure for a year. He had time again to think. He had either the benefit or the curse, depending which way you looked at it, of being the world chess champion during the period in history in which computers got incredibly good at chess. But it wasn’t just chess. Kasparov thought about how a young person entering adolescence during that year would not remember a world in which personal computers didn’t exist. He thought about how this would only continue to develop, until everyone, including Kasparov himself, would lose their memory of such a time. This technology was here to stay, and now it was a matter of how we thought about it. And slowly, Kasparov’s feelings about Deep Blue began to change.
Though he never heard anything from IBM or its creators as to why Deep Blue committed suicide in game one or avoided his trap (cheated, as he believed it) in game two, he remembered the nuance and skill with which the computer played. He thought about Moravec’s paradox, which he had heard about around the time he was first invited to battle computers. Moravec’s paradox states that it’s comparatively easy to get a computer to display high-level intelligence in difficult games like chess, but comparatively difficult to program things humans consider basic, like perception or mobility. The best computers are barely comparable with one-year-olds in such categories. In other words, the paradox encapsulates why it’s difficult for computers to truly replicate human intelligence: humans and computers have markedly different skillsets.
Kasparov knew — and IBM would readily admit — that Deep Blue is not generally intelligent, and that it didn’t play chess in the same way that Kasparov did. Deep Blue chugged through tens of millions of chess configurations through sheer brute force, and selected the one with the highest probability of leading to a victory — and the result was nuanced and skilled gameplay. In other words, it did what computers do best. But Deep Blue’s gameplay, at the end of the day, was still straightforward. It didn’t set any traps for Kasparov in the way he had set traps for it. There was very little indication of long-term strategy, or creativity. Those things, Kasparov figured, are relegated to the category of what humans do best.
At the end of the day, Deep Blue was no more intelligent than an alarm clock, doing predetermined computations to perform a specific goal. While Kasparov failed to find comfort in losing to a ten million dollar alarm clock, he did come away with an idea. So much fuss had been made about the match with Deep Blue as a competition. But, the grandmaster figured, what if the emphasis wasn’t on “humans versus machines” but instead on “humans with machines”? How would human-computer teams do in games of chess? It would be, at the very least, a fun experiment. Kasparov had dubbed it “Advanced Chess.”
So here he was, in Leon, Spain, a little over a year after his defeat to Deep Blue. He was sitting across the table from Topalov, each of them with a bulky PC at his side. They were still on a time limit, like they would be in any competition match, but, in advanced chess, they could consult their computer on each turn. There was not nearly as much media attention here as there was in New York the previous year, but, then again, a doomsday prophecy of the inferiority of the human brain wasn’t at stake this time around.
Kasparov, for one, was finding his experiment equal parts thrilling and off-putting. He knew that so much of chess was based on memory — remembering past games, past configurations, how you acted in those situations, and what response you might expect from your opponent this time around. But in advanced chess, the computer carried that information for you. And that outsourcing of mental burden — and the fact that Topalov had that same exact advantage — meant that the real advantage was the creativity of the human using the computer. Who could use the immense information at their disposal to come up with a better strategy, faster? Previously, Kasparov had defeated Topalov handily — 4–0 — without computer assistance. This time, they tied. Each won 3 games of a 6 game match. Whatever innate upper hand Kasparov had was wiped out by the presence of the computer.
After the game, Kasparov’s mind drifted again. He was twenty-two, having just become the youngest world champion in history. He was being questioned by reporters about his success. However, none of these questions were about strategy and skill, the only tangible measures of performance in a chess game. They were about his diet, his memory, his work ethic. And when Kasparov found the words to answer these — frankly, irrelevant — questions, none of his answers ever seemed to satisfy anyone. What they were looking for was a golden ticket, a DIY for greatness. This greatness — Kasparov’s greatness — was mysterious, pale, unknowable. Even he didn’t know how to tap into it. It just was.
But now Kasparov saw — perhaps there was more greatness to be had. Humans do certain things better than computers, and computers do certain things better than humans. That’s a fact. However, what these two entities can do alone was nothing, potentially, compared to what they can do together.
This could require further experimentation, Kasparov thought, as he looked towards the future.
Sources
Finley, Klint. “Did a Computer Bug Help Deep Blue Beat Kasparov?” Wired. N.p., 28 Sept. 2012. Web. <https://www.wired.com/2012/09/deep-blue-computer-bug/>.
Garry Kasparov: What Constitutes Intelligence. TechCrunch. N.p., 17 May 2017. Web. <https://techcrunch.com/video/garry-kasparov-what-constitutes-intelligence/591c6c699e45 10219e2ecd78/>.
IBM Research Scientist Dr. Murray Campbell on Deep Blue. IBM Research, n.d. Web. 10 May 2012. <https://www.youtube.com/watch?v=KFSVZlkHHmM>.
“I.B.M.’s Stock Surges by 3.6%.” The New York Times. The New York Times, 12 May 1997. Web. <http://www.nytimes.com/1997/05/13/business/ibm-s-stock-surges-by-3.6.html>.
Kasparov, Garry. “The Chess Master and the Computer.” The New York Review of Books. N.p., 11 Feb. 2010. Web <http://www.nybooks.com/articles/2010/02/11/the-chess-master-and-the-computer/>.
“Kasparov vs. Deep Blue.” NPR. NPR, 08 Aug. 2014. Web <http://www.npr.org/2014/08/08/338850323/kasparov-vs-deep-blue>.
“Rules of the Rematch.” IBM Research. IBM, 23 Feb. 2001. Web <https://www.research.ibm.com/deepblue/watch/html/c.8.html>.
Silver, Nate. “Rage Against The Machines.” FiveThirtyEight. FiveThirtyEight, 23 Oct. 2014. Web. <http://fivethirtyeight.com/features/rage-against-the-machines/>.
“The Terminator and the Washing Machine.” Retro Report. Retro Report, 06 Mar. 2016. Web. <https://www.retroreport.org/video/transcript/the-terminator-and-the-washing-machine/>.
“Timeline.” Garry Kasparov. N.p., 12 June 2014. Web.<http://www.kasparov.com/timeline/>. Weber, Bruce.
“Swift and Slashing, Computer Topples Kasparov.” The New York Times. The New York Times, 11 May 1997. Web <http://www.nytimes.com/1997/05/12/nyregion/swift-and-slashing-computer-topples-kasparov.html>.
Wright, Robert. “Can Machines Think?” Time. Time Inc., 25 Mar. 1996. Web.<http://content.time.com/time/subscriber/article/0,33009,984304-1,00.html>.