1장
When Machines Outthink Champions
In 1997, a watershed moment in human history occurred when IBM's Deep Blue supercomputer defeated world chess champion Garry Kasparov. This wasn't just any chess match - it represented the culmination of a decades-long quest to build a machine that could outthink the greatest human minds in a domain long considered the pinnacle of intellectual achievement. The match captivated global attention, making front-page headlines and sparking profound questions about the future of human cognition in an increasingly computerized world. Kasparov, who had dominated chess for over a decade with an almost supernatural intuition for the game, found himself facing something entirely new: an opponent with no psychology to exploit, no fatigue to leverage, and the ability to calculate 200 million positions per second. What began as a scientific experiment evolved into a deeply personal and philosophical journey for Kasparov, who would spend the next twenty years exploring the implications of that defeat and pioneering new forms of human-machine collaboration. This exploration reveals not just the evolution of artificial intelligence, but offers a window into how we might navigate a future where machines increasingly outperform humans at cognitive tasks once thought to be exclusively human.
2장
The Ancient Game Meets Modern Technology
Chess has captivated human minds for over 1,500 years, spreading from its origins in India's chaturanga to become a global phenomenon. While Americans might consider chess a niche intellectual pursuit, in many countries it's a national passion. In Armenia, a country of just 3 million, chess is mandatory in schools and produces a disproportionate number of champions. In Russia and other former Soviet states, chess was elevated to an art form and symbol of intellectual superiority during the Cold War.
What makes chess so compelling as a testbed for artificial intelligence is its perfect balance of clarity and complexity. The rules are simple enough to be programmed, yet the game's depth makes mastery elusive. With more possible chess positions than atoms in the observable universe, brute calculation alone cannot solve chess. Human grandmasters rely on pattern recognition, intuition, and strategic understanding rather than calculating every possible move.
This distinction between human and machine approaches to chess reveals what scientists call Moravec's paradox: machines excel at tasks humans find difficult (like rapid calculation) but struggle with skills humans find effortless (like recognizing patterns or making analogies). A computer can calculate millions of positions per second but might miss strategic concepts a human instantly grasps.
The quest to create a chess-playing machine predates modern computers. In the late 18th century, the "Mechanical Turk" chess automaton toured Europe, defeating luminaries including Napoleon Bonaparte and Benjamin Franklin. It was later revealed to be an elaborate hoax concealing a human chess master inside its cabinet. The first genuine chess algorithm was created by Alan Turing in 1952, computed by hand on paper since no computer powerful enough existed yet.
Early AI pioneers believed chess would be the perfect proving ground for machine intelligence. Claude Shannon's groundbreaking 1949 paper "Programming a Computer for Playing Chess" outlined two potential approaches: Type A ("brute force") examining every possible move, and Type B ("intelligent search") focusing only on promising moves. Shannon believed Type B would be necessary since the computational demands of brute force seemed insurmountable. Ironically, it was ultimately the brute force approach, refined through clever pruning algorithms, that conquered chess.
The evolution of chess machines provides a perfect case study in how technology progresses: from laughably weak to interesting but flawed, then to useful but limited, and finally to transcendent and superior. This pattern repeats across countless technological domains, and understanding it helps us anticipate how artificial intelligence will transform other fields in the coming decades.
3장
The Human Mind vs. The Machine Mind
The fundamental difference between human and machine thinking becomes starkly apparent at the chessboard. Grandmasters don't systematically evaluate every possible move with numerical precision. Instead, they rely on pattern recognition and intuition to identify candidate moves before calculation begins. This process is remarkably efficient but inherently imprecise.
During competition, the human mind wanders. As world champion Mikhail Tal once famously described, while contemplating a knight sacrifice, his thoughts drifted to how one might extract a hippopotamus from a marsh. These mental digressions sometimes lead to blunders, but they can also produce brilliant, paradoxical moves that wouldn't emerge from systematic analysis. Under time pressure, even grandmasters experience visualization errors and calculation mistakes.
Machines suffer from none of these psychological limitations. They don't get nervous, tired, or overconfident. They don't fear their opponents or become demoralized after mistakes. This psychological immunity is as much a competitive advantage as their raw calculating power. When I faced Deep Blue, I wasn't just competing against silicon and code - I was battling my own emotions and doubts while the machine remained perfectly stable.
The machine's approach to chess is fundamentally different. Early chess programs evaluated positions primarily by counting material values - pawns worth one point, knights and bishops worth three, and so forth. This material-focused approach resembles how novice humans play, especially children. But the similarity is misleading. While both beginners and early chess programs focus on material, they do so for entirely different reasons. Humans lack calculation ability, while computers lack pattern recognition and positional understanding.
As chess programs evolved, programmers added positional factors to their evaluation functions - pawn structure, king safety, piece mobility - attempting to encode the strategic knowledge of grandmasters. But these heuristics were crude approximations of human understanding. The real breakthrough came with the "alpha-beta" algorithm, which allowed programs to rapidly prune weak moves from consideration, enabling them to search much deeper in promising variations.
By the 1980s, chess machines had reached expert level - the top 5% of tournament players. Ken Thompson's special-purpose machine Belle searched 180,000 positions per second and played at master level. By 1988, Hans Berliner's HiTech reached grandmaster strength, and Deep Thought became the first machine to defeat a grandmaster in tournament play. When Deep Thought's creators joined IBM in 1989, the project evolved into Deep Blue, beginning the final chapter of machine chess's ascent.
The question "When will a chess-playing machine beat the world champion?" followed every chess programmer for decades. Early predictions were wildly optimistic, but technology rarely follows the linear progress we expect. Instead, it follows an S-curve - years of struggle, then breakthrough, then maturation. By the mid-1990s, that breakthrough was imminent.
4장
The Psychology of Competition
Chess is fundamentally a competitive sport, despite debates about its classification. The psychological and physiological exertion in chess matches the intensity found in physical sports. Players experience extreme nervous tension that fluctuates with every move for hours, creating an emotional roller coaster that leaves even the most composed players exhausted.
Recovery from defeat is particularly challenging in chess because there are no external factors to blame - no referees, teammates, or luck elements. A loss means you failed, period. Competitive players must balance putting losses behind them while objectively analyzing failures to avoid repeating them.
This psychological dimension is entirely one-sided when humans face computers. Machines don't experience confidence, dejection, fatigue, nervousness, hunger, or distraction. This asymmetry makes it even harder for humans to manage their own emotions against silicon opponents.
Motivation critically impacts chess performance, particularly in maintaining intense concentration over extended periods. What psychologists call "chess talent" remains somewhat mysterious - some players are simply much better than others, beyond what experience or training can explain. While Malcolm Gladwell's "ten thousand hours" theory suggests practice determines exceptional achievement in complex activities like chess, this underestimates talent's potency, especially in early development.
My experience with the Kasparov Chess Foundation confirms this. We identify talent early by recognizing rating outliers - children performing two or three years ahead of peers. A nine-year-old with an expert rating of 2100 shows more promise than a twelve-year-old at the same level. These exceptional children rarely regress unless faced with choosing between professional chess and other life paths.
The availability of grandmaster-strength computers has profoundly affected how humans play against each other. Young players who've trained with super-strong engines develop differently than those from earlier eras. These machines, free from prejudice and dogma, have created a generation of players who approach chess more mechanically - a move is good simply if it works, not because of established theory.
The problem emerges when databases and engines transform from coach to oracle. When I ask students why they made a move, they often answer "Because that's the main line" or "It's the best move." But they struggle to explain why it's best. Without understanding the position's fundamentals, players become lost without their digital crutch.
For true innovation in chess, you must look earlier than where the database ends. You need to question established moves that everyone assumes are best because they've been played countless times. If we only use machines to become good imitators, we'll never become creative innovators.
5장
The Deep Blue Challenge
My rise to world champion coincided with Gorbachev's glasnost and perestroika, allowing me to challenge Soviet restrictions on athletes. My five consecutive world championship matches against Anatoly Karpov from 1984-1990 elevated chess nearly to the level of the famous Fischer-Spassky match. As the Soviet system weakened, our matches moved progressively westward - from Moscow to London/Leningrad to Seville to New York/Lyon - mirroring the opening of a new world of opportunity for chess.
Around the time Deep Thought became the first real threat to grandmasters in the late 1980s, artificial intelligence was experiencing a resurgence after the "AI winter" of disillusionment. The field had shifted from ambitious goals of understanding human cognition to practical, narrow applications - "Don't make it think, just make it work."
My match against Deep Thought took place with the machine hundreds of miles away and an operator making its moves. While I was known as chess's most thoroughly prepared player, computers can simply load new opening books before each game, negating my advantage. Opening books allow computers to skip the strategic early phase where they're weakest, jumping straight to the tactical middlegame where they excel.
I won both games convincingly. With black in the first game, I slowly built a dominating position and eventually broke through after 52 moves. In the second game with white, I offered a "poisoned pawn" that the computer eagerly took, falling into difficulties as my pieces swarmed the board. The machine's team later claimed there was a "castling bug" they discovered weeks later - a theme that would recur in future matches.
In May 1994, I finally lost to a machine - Fritz 3 at a blitz tournament in Munich sponsored by Intel Europe. Despite starting the tournament with eight straight wins, I met Fritz 3 at the top of the standings. I gained a crushing position early but made one lazy move, allowing a counterattack. Fritz 3 won, marking the first victory over a world chess champion in a serious game by a machine.
By 1995, discussions about a match with Deep Blue finally began. IBM wanted it and so did I - the question was whether the machine would be ready. With a $500,000 prize fund split 4-1 for the winner, I was confident but concerned about the lack of information on this new version's capabilities.
This new Deep Blue, with 216 chess chips connected to an IBM supercomputer, could search one hundred million positions per second - twenty times faster than its predecessor. The IBM team had also hired Grandmaster Joel Benjamin to prepare the opening book and tune the evaluation function. Even the fastest chess machine needed human chess knowledge.
6장
The Historic Confrontation
Twenty years later, I still struggle to describe what it's like for a world champion to play against a world-champion-level chess machine. It's not like a video game or a metaphorical competition in the job market. It's direct competition against a computer at the highest level of a human discipline.
The media coverage echoed romantic notions about chess and intelligence, with headlines like "The Brain's Last Stand" and "Kasparov Defends Humanity." Even late-night comedians made nervous jokes about machines coming for human jobs. The organizers and participants indulged these flattering narratives about chess as "the pinnacle of human intellectual activity" and me as a "living Mount Everest."
In game one of our 1996 match, Deep Blue played several strong moves to create real threats. This thing was strong. This was different. I resigned on move thirty-seven, making history as the first world chess champion defeated by a computer in a classical game. In mild shock, I reflexively asked Feng-hsiung Hsu "Where did I go wrong?" but he couldn't answer, creating an awkward moment. That evening, I wondered to my friend Frederic, "What if this thing is invincible?"
I quickly discovered Deep Blue wasn't invincible. In game two, I employed a slow, maneuvering opening to deny the machine clear targets, knowing it couldn't formulate strategic plans like humans. My strategy worked perfectly. Deep Blue struggled with the long-term structural weakness I created, confirming my theory that positions where general principles outweigh calculations would be its downfall. After hours of careful maneuvering, I won.
Now I knew this "new intelligence" was merely a faster version of programs I understood well. It had clear deficiencies I could exploit by targeting its weaknesses while avoiding its strengths. By the end of our six-game match, I had won 4-2, exactly the score I had predicted.
Though starting with minimal publicity, the first Deep Blue match became the largest Internet event in history at that time, requiring IBM to assign a supercomputer just to handle website traffic - and this was in 1996 when most people used dial-up connections. While the Deep Blue team was disappointed with the match result, IBM was ecstatic. The publicity value far exceeded my winner's check, dramatically boosting IBM's stock price and modernizing the company's image. According to Monty Newborn's book, IBM's stock rose by $3.31 billion in just over a week, during a period when the Dow Jones was declining significantly.
The PR bonanza virtually guaranteed a rematch, though timing depended on how quickly the Deep Blue team could make substantial improvements. As negotiations progressed, one thing became abundantly clear: any rematch wouldn't be driven by the Deep Blue team's desire to improve or my interest in another paycheck - it would happen because IBM wanted to win.
7장
The Rematch That Changed Everything
Our contacts with IBM before the 1997 rematch revealed a complete change in attitude. The friendly, open approach from Philadelphia was replaced with obstruction and hostility. Project manager C.J. Tan had bluntly told the New York Times, "We're not conducting a scientific experiment anymore. This time, we're just going to play chess."
Had I recognized this shift from Chopin waltz to Tchaikovsky march earlier, I could have adjusted my demeanor accordingly. This was particularly difficult since IBM was simultaneously my opponent, host, organizer, and sponsor.
IBM had taken over several floors of Manhattan's Equitable Center for the match. Deep Blue's main system operated on-site in a room with extraordinary security, with backup systems in Yorktown Heights and in the building that could take over seamlessly. The new Deep Blue ran on a supercomputer twice as fast as before, containing 480 of Hsu's improved chess chips and reaching 200 million positions per second.
Game one of the rematch might have been the most anticipated chess game since Fischer-Spassky in 1972, with magazine covers, bus-stop ads, and TV talk shows creating enormous pressure. I opened with my knight to f3, part of my "anti-computer" strategy to avoid sharp positions where Deep Blue might excel. My strategy initially paid dividends as Deep Blue made strange, shuffling moves that revealed its imperfect understanding.
Eventually, Deep Blue launched a fierce counterattack, but its downfall came when it overestimated its material advantage and exchanged queens, failing to see it would have no way to improve its position while I could. After several inaccuracies, Campbell resigned. I'd beaten the computer at its own calculating game.
In game two, Deep Blue's play demonstrated remarkable positional understanding. I played passively, missing my last chance for active defense. Deep Blue built pressure with Karpov-like patience, and after nearly four hours in a passive position I despised, fatalistic depression set in. When Deep Blue attacked my queen with its rook on move forty-five, I resigned in disgust and stormed away.
The next day came the devastating revelation: the final position was actually drawn. My team discovered that perpetual check with Queen to e3 would have saved the game. For the first time in my life, I had resigned in a drawn position. This was crushing - losing the game twice. Against a computer checking 200 million positions per second, I'd given the machine the benefit of the doubt, assuming it wouldn't allow a drawing variation.
Game six would set several unwanted records: my shortest career loss, my first classical match defeat, and the first time a machine defeated a world champion in a serious match. With the score tied 2.5-2.5, I abandoned my strategy and played the Caro-Kann, a solid positional opening I'd used extensively in my youth. Deep Blue followed a main line I knew well from playing it with white.
On move seven, I played my h-pawn one square instead of the normal bishop move. Deep Blue instantly sacrificed its knight, exposing my king with overwhelming threats. I knew immediately the game was over but went through the motions for another dozen moves before resigning in less than an hour.
Deep Blue never played another game. IBM had achieved what it wanted - a giant PR boost and an $11.4 billion increase in stock value in just over a week. Despite my phone call with IBM CEO Lou Gerstner, there would be no rubber match. The chess community was outraged at IBM's decision to dismantle Deep Blue. As Frederic Friedel told the New York Times, "Deep Blue's victory was a milestone in artificial intelligence, but it's a crime that IBM didn't let it play again. It's like going to the moon and returning home without looking around."
8장
Human Plus Machine: The Future of Intelligence
As humanity quickly recovered from my defeat to Deep Blue, I found myself returning to chess while the Deep Blue team had engineered their own obsolescence. Deep Blue's victory proved what AI advocates had warned about: there was little to learn beyond confirming that faster machines would inevitably defeat the human champion around the year 2000.
As human versus machine competition ended, human plus machine collaboration took center stage. Our technology now supplements fundamental cognitive functions like memory. Tech writer Cory Doctorow coined the term "outboard brain" in 2002 to describe how digital repositories increase "the volume and quality of the yield" of our information processing.
Concerns about cognitive outsourcing emerged when young people adopted these technologies in ways parents didn't understand. Some worry about becoming "mentally crippled" when offline or that machine memory shuts down other ways of understanding the world. The acquisition of knowledge must serve more than immediate tasks if we seek wisdom. Your phone making you an instant expert doesn't make us dumber any more than encyclopedias did. The real danger is substituting superficial knowledge for the understanding required to create new things.
In 1998, I created an experiment called "Advanced Chess" in Leon, Spain, where humans and machines played as partners rather than opponents. Each player had a computer running chess software during the game, aiming to create the highest level of chess ever played - a synthesis of human and machine strengths.
The experiment continued in 2005 when Playchess hosted a "freestyle" tournament where anyone could compete with computer assistance. Surprisingly, the winners weren't grandmasters with powerful computers, but two amateur American players using three computers simultaneously. Their superior process for coordinating with machines beat both stronger computers alone and stronger humans with machines using inferior processes. This demonstrated that weak human + machine + better process was superior to strong human + machine + inferior process - a conclusion that became known as "Kasparov's law" and garnered significant interest from technology companies.
After decades of trying to replace human intelligence with algorithms, many companies and researchers now focus on bringing the human mind back into data analysis processes. As with chess programs that went from knowledge to brute force and back toward knowledge as returns diminished, the key is the process - something only humans can design.
9장
Embracing Our Technological Future
The concept of the singularity - a tipping point beyond which human history becomes incomprehensible due to superintelligent machines - has diverse interpretations. Vernor Vinge warned in 1993 that within thirty years "we will have the technological means to create superhuman intelligence" after which "the human era will be ended." Nick Bostrom has become an evangelist about these dangers, explaining in detail how super-intelligent machines might not care to keep humans around.
Meanwhile, Ray Kurzweil presents an almost utopian vision where the singularity combines genetics and nanotechnology to augment minds and bodies toward advanced cognition and extended lifespans.
MIT's Andrew McAfee identifies the biggest misconception about AI: "The hope that the singularity - or the fear that super-intelligence - is right around the corner." His pragmatic investigations match my outlook. Machine learning expert Andrew Ng similarly notes that worrying about super-intelligent AI today is like worrying about "overcrowding on Mars."
While I appreciate people like Bostrom worrying about long-term AI risks, I prefer focusing on immediate issues. New technologies bring growing pains that often prove less consequential than initially feared. Pessimism is more detrimental to civilization's development than optimism.
We cannot predict all technological changes, but I trust young people growing up with technology to find surprising new ways to use it, just as every generation has done before.
Rather than winding down, I want to stir things up. This debate isn't academic - it's happening now. The more people believe in a positive technological future, the greater chance we have of creating one.
Nothing is decided. We're all on the board in this game. The only way to win is to think bigger and deeper. This isn't about utopia versus dystopia or humans versus machines. We need our ambition to stay ahead of our technology. As we excel at teaching machines our tasks, we must keep creating new challenges even we don't know how to solve.
Our technology removes difficulty from our lives, so we must seek ever more difficult challenges. Technology can make us more human by freeing our creativity, but being human goes beyond creativity. We have purpose while machines have instructions. Machines cannot dream. We need our intelligent machines to turn our grandest dreams into reality - if we stop dreaming big, we may as well be machines ourselves.
