HUMAN COMPETITION with machines has been part of the conversation about technology since the first machines were invented. We continue to update the terminology, but the basic narrative remains the same. People are being replaced, or are losing a race, or being made redundant, because technology is doing what humans used to do. This “human versus machine” narrative framework arose to prominence during the industrial revolution, when the steam engine and mechanized automation in agriculture and manufacturing began to appear on a large scale.
The competition story line grew more ominous and more pervasive during the robotics revolution of the 1960s and 1970s, when more precise and more intelligent machines began to encroach on the jobs of people with more powerful social and political representation, like unions. The information revolution came next, culling millions of jobs from the service and support industries.
Now we have reached the next chapter in the human versus machine employment story, when the machines “threaten” the class of people who write articles about it. We read headlines every day about how the machines are coming for the lawyers, bankers, doctors, and other white-collar professionals. And make no mistake, they are. Every profession will eventually feel this pressure, and it must, or else it will mean humanity has ceased to make progress. We can either see these changes as a robotic hand closing around our necks or one that can lift us up higher than we can reach on our own, as has always been the case.
Romanticizing the loss of jobs to technology is little better than complaining that antibiotics put too many grave diggers out of work. The transfer of labor from humans to our inventions is nothing less than the history of civilization. It is inseparable from centuries of rising living standards and improvements in human rights. What a luxury to sit in a climate-controlled room with access to the sum of human knowledge on a device in your pocket and lament how we don’t work with our hands anymore! There are still plenty of places in the world where people work with their hands all day, and also live without clean water and modern medicine. They are literally dying from a lack of technology.
It’s not just college-educated professionals who are under pressure today. Call center employees in India are losing their jobs to artificially intelligent agents. Electronics assembly-line workers in China are being replaced by robots at a rate that would shock even Detroit. There is an entire generation of workers in the developing world who were often the first in their families to escape farming and other subsistence labor. Will they have to return to the fields? Some may, but for the vast majority this isn’t an option. It’s like asking if all the lawyers and doctors will have to “return to the factories” that don’t exist anymore. There is no back, only forward.
We don’t get to pick and choose when technological progress stops, or where. Companies are globalized and labor is becoming nearly as fluid as capital. People whose jobs are on the chopping block of automation are afraid that the current wave of tech will impoverish them, but they also depend on the next wave of technology to generate the economic growth that is the only way to create sustainable new jobs. Even if it were possible to mandate slowing down the development and implementation of intelligent machines (how?), it would only ease the pain for a few for a little while and make the situations worse for everyone in the long run.
Unfortunately, there is a long tradition of politicians and CEOs sacrificing the long term and the greater good in order to satisfy a small constituency at the moment. Educating and retraining a workforce to adapt to change is far more effective than trying to preserve that workforce in some sort of Luddite bubble. But that takes planning and sacrifice, words more associated with a game of chess than with today’s leaders.
Donald Trump won the US presidency in 2016 with promises of “bringing jobs back” from Mexico and China, as if American workers can or should be competing for manufacturing jobs with countries where salaries are a fraction of those in the United States. Putting high tariffs on foreign-made products would make nearly every consumer good far more expensive for those who can least afford such an impact. If Apple offered a red, white, and blue iPhone made in the United States that cost twice as much as the same model made in China, how many would they sell? You can’t discard the downsides of globalization while keeping the benefits.
It’s a privilege to be able to focus on the negative potential of world-changing breakthroughs like artificial intelligence. As real as these issues may be, we will not solve them unless we keep innovating even more ambitiously, creating solutions and new problems, and yet more solutions, as we always have. The United States needs to replace the jobs being lost to automation, but it needs new jobs to build the future instead of trying to bring back jobs from the past. It can be done and it has been done before. Here I’m not referring to the 30 percent of Americans who lived on farms in 1920, down below 2 percent nearly a century later, but to a much more recent retooling.
The launch of the tiny Sputnik device by Sergey Korolyov on October 7, 1957, turned the space race into a sprint that lasted for decades. President Eisenhower immediately ordered all American projects to move up their timetables, which likely helped contribute to the failed launch of the first American satellite, Vanguard, in December 1957. The media dubbed the failure, seen live on television, “Flopnik,” and the embarrassment drove the administration to push even harder for results.
The phrase “Sputnik moment” subsequently entered the national lexicon to represent any foreign accomplishment that serves to remind America that it is not without rival. For example, the OPEC oil embargo of the 1970s was supposed to be a Sputnik moment that would goad the United States into developing renewable energy. Then came Japanese manufacturing technology in the 1980s, the expanded European Union in the 1990s, and the rise of Asia in the last decade.
A more recent Sputnikian wakeup call to rouse the American giant was supposed to be the 2010 revelation that kids in Shanghai scored far better on standardized math, science, and reading tests than their peers in other nations. An October 13, 2016, Washington Post headline warned that “China has now eclipsed us in AI research.” Perhaps this fact is not unrelated to those 2010 test scores. Yet another Sputnik moment? As you can see, the track record of Americans picking up any of these gauntlets is quite poor, except, of course, the original.
Inevitably, all these repetitions have trivialized the impact of Sputnik, which combined many of the day’s real and imagined fears into a twenty-three-inch-diameter metal sphere. American editorial pages of the day were filled with wonder and dread at this shocking combination of Communist ideology and unmatched technology. Sputnik stoked American fires in the most primeval ways: creating fear and anger, and denting America’s national ego and pride.
The United States responded. In 1958, three years before President John F. Kennedy boldly promised to put a man on the moon by the end of the decade, then–Senator Kennedy supported legislation called the National Defense Education Act, which directly funded science education across the country. The future engineers, technicians, and scientists produced by the program would form the generation that designed and built much of the digital world we live in today.
It is still an open question whether a national revitalization effort can be summoned, like Aladdin’s genie, on demand. It is depressing to consider the thought that war and fear are necessary requirements to inspire united action since we are obviously better off in a world with as little as possible of both. But existential threats do focus the mind wonderfully, as Samuel Johnson said about an impending hanging. Any transformative effort on a national scale requires the focused minds of politicians, business leaders, and a plurality of citizens to support it.
In the 1970s, superior Japanese cars were bought by American consumers in the millions. Chinese graduates are enthusiastically welcomed into every American university and firm. In today’s globalized world, technological competition has given way to the sense that we all benefit from someone, somewhere, doing things right, or at least doing them better. While this is no doubt better than no one doing it right anywhere, we cannot abandon the quest for scientific excellence in the United States. America still possesses the unique potential to innovate on a scale that can push the entire world economy forward. A world in which America is content with mediocrity is, literally, a much poorer world.
When questioned by Congress about the Soviet success, President Eisenhower’s special assistant for science and technology, Dr. James Killian, also president of MIT, gave a cultural answer to a technical question: “There is no doubt that the Soviets have generated a respect and enthusiasm for science and engineering that has operated to give them a large supply of trained professionals in these fields.” He was quoted in the December 1957 issue of the Bulletin of the Atomic Scientists, whose editors were even more critical of the American mindset that had allowed the Soviets to pull ahead in space, as this editorial comment in the same article made clear: “We have catered to desires for undisturbed comfort rather than focusing on larger goals and developing our potentialities.”
This was a polite and professorial way of saying that Americans had become lazy, short-sighted, and unwilling to take the risks required to stay on the cutting edge of technology. I’m worried that this is where the United States is finding itself once again. Silicon Valley is still the greatest hub of innovation in the world and America possesses more of the conditions necessary for success than anywhere else. But when is the last time you heard about a government regulation that promoted innovation instead of trying to limit it?
I’m a firm a believer in the power of free enterprise to move the world forward. All that Soviet respect for science was no match for the American innovation machine once unleashed. The problem comes when the government is inhibiting innovation with overregulation and short-sighted policy. Trade wars and restrictive immigration regulations will limit America’s ability to attract the best and brightest minds, minds needed for this and every forthcoming Sputnik moment.
FIGHTING TO THWART the impact of machine intelligence is like lobbying against electricity or rockets. Our machines will continue to make us healthier and richer if we use them wisely. They will also make us smarter. It’s an interesting diversion to consider what the first invention was that directly increased human knowledge and our understanding of the world. Starting in the thirteenth century, grinding glass led to glasses, and eventually to the telescope and the microscope, tools of human enhancement that dramatically improved our ability to control our environment via improved navigation and medical research. Perhaps only the compass is an earlier invention that provided us with information otherwise difficult or impossible to obtain. The abacus, from the third millennium BC, is as much a method as a machine, but it is likely the first device to augment human intelligence. The alphabet, paper, and the printing press didn’t exactly create knowledge, but performed the essential corollary task of preserving and distributing it, much as the Internet does.
My own experiences battling computers across a game board are the exception that proves the rule. We aren’t competing against our machines, no matter how many human jobs they can do. We are competing with ourselves to create new challenges and to extend our capabilities and to improve our lives. In turn, these challenges will require even more capable machines and people to build them and train them and maintain them—until we can make machines that do those things too, and the cycle continues. If we feel like we are being surpassed by our own technology it’s because we aren’t pushing ourselves hard enough, aren’t being ambitious enough in our goals and dreams. Instead of worrying about what machines can do, we should worry more about what they still cannot do.
I will say again that I am not unsympathetic to those whose lives and livelihoods have been negatively impacted by disruptive new technology. Few people in the world know better than I do what it’s like to have your life’s work threatened by a machine. No one was sure what would happen if and when a chess machine beat the world champion. Would there still be professional chess tournaments? Would there be sponsorship and media coverage of my world championship matches if people thought the best chess player in the world was a machine? Would people still play chess at all?
The answer to all of these questions turned out to be yes, thankfully, but these doomsday scenarios were one reason some in the chess community criticized my eagerness to participate in human versus machine events at all. I suppose I could have delayed the inevitable a little by declining, and forcing the programmers to challenge other top players. If a machine had beaten Anand or Karpov, the next players on the rating list after me at the time of my rematch with Deep Blue in May 1997, the story would have been “Nice, but would it beat Kasparov?” But that would only last until I was no longer the world champion, which happened in 2000, or until I was no longer ranked number one and I retired from chess, which happened in 2005. I was never one to duck a challenge, and being remembered as the first world champion to lose a match to a computer cannot be worse than being remembered as the first world champion to run away from a computer.
And I didn’t want to run away. I was thrilled by these new trials, by the scientific pursuit, by the new avenues to promote chess, and, frankly, by the attention and the money that sometimes came with it all. Why should someone else be the first, for better or for worse? Why should I exchange a unique and historic role as a participant to become just another spectator?
Nor did I believe the apocalyptic predictions about what might happen if I lost a match to a machine. I was always optimistic about the future of chess in the digital age, and not because of the trite and imprecise “people still run footraces even though cars are faster” justifications that many were making at the time. John Henry aside, automobiles didn’t make walking obsolete or put pedestrians out of work. Many things on Earth are faster than Usain Bolt’s top speed of thirty miles per hour, from coyotes (40 m.p.h.) to kangaroos (44 m.p.h.). So what?
Chess is a very different matter from physical sports, as strong chess machines can directly and indirectly influence human play. You can think of them as more analogous to steroids and other forms of doping in physical sports, as an external augmentation with the potential to boost performance or to damage the sport if abused. Chess is concrete; a move or strategy employed by a computer can be exactly duplicated by a human. What if machines showed us that some of the most popular chess openings were bad, and how to beat them? Would we human players become the automatons ourselves, regurgitating the moves and ideas shown us by our machines? Would the winner be the player with the strongest computer at home? Would there be an epidemic of computer-assisted cheating? These were realistic and serious questions, and they still are, but these are not the same as the dismal fantasies about computers solving chess for good or making human versus human play obsolete.
As with nearly every new technology, for every potential downside there were many upsides to the increasing strength and availability of strong chess machines. I admit, however, that I was late in recognizing this. The first few generations of PC chess software, powered by what are called “chess engines” in our vernacular, were too weak to be very useful to professional players. The most popular programs were directed toward casual consumers and focused more on pretty 3-D boards or animated pieces than on the strength of the engine. Even as they got much stronger and became dangerous opponents in the early 1990s, the chess they played was ugly and inhuman, not very useful for serious training.
Instead, my early interest was in developing computer tools to help with my preparation and that of other serious players. Instead of digging through dozens of reference books and my stacks of notebooks full of analysis, a database of thousands of games could be searched in a few seconds and could also be easily updated. In 1985, I started discussing the creation of such an app with the German tech writer Frederic Friedel, who was a serious aficionado of computer chess. He and a programmer acquaintance, Matthias Wüllenweber, founded ChessBase in Hamburg and released the ground-breaking program of the same name in January 1987. And with that, an ancient board game was pulled into the information age, at least if you had an Atari ST. The ability to collect, organize, analyze, compare, and review games with just a few clicks was, as I put it at the time in 1987, as revolutionary for the study of chess as the printing press.
As for chess engines, by the early 1990s I had lost a few blitz games to the top PC programs and it was clear they were only going to keep getting stronger. Before that, back when home computers were still uncommon in most of the world, machine capabilities were often wildly over- and underestimated. There had been a few early theories, optimistic ones from my perspective, that the exponential branching factor of chess analysis would create a barrier at some point, but programming techniques and ever-faster CPUs kept the machines’ ratings rising steadily.
I gradually understood that the proliferation of strong programs could greatly democratize the sport worldwide. My success in chess was as much a case of geography being destiny as it was my having natural talent and a determined mother. In the Soviet Union, I had easy access to chess books, magazines, coaches, and a ready supply of strong opponents. Nowhere else in the world could offer these advantages, except perhaps the former Yugoslavia. Other national chess powers also counted on longstanding chess traditions that provided the resources necessary for talent to develop.
The existence of a Grandmaster-level chess program available on an inexpensive personal computer upended that hierarchy. While not as good as an experienced human coach, it was far better than nothing. Combined with the Internet’s ability to bring the game to every corner of the world, a shift was under way. The key factor in producing elite chess talent is finding it early, and thanks to strong computers this is now very easy to do just about anywhere. It’s no coincidence that the current list of elite chess players contains many representatives of countries with little or no old chess traditions. Computers tend to have this impact in many ways, reducing the influence of dogma. Chess in China and India has been boosted by government support and local stars, but the ability to train with Grandmaster machines helped make their rise into the elite ranks startlingly quick. Previously, it was necessary to import Soviet coaches and host expensive international tournaments or send local players abroad to find strong competition. China currently has six players among the top fifty in the world. Russia still has the most, eleven, but their average age is thirty-two, while that of the Chinese players is twenty-five.
The current world champion, Magnus Carlsen, is from Norway and was born in 1990. He has never known a world in which computer chess programs weren’t stronger than he is. Ironically, he is very much a “human style” player, whose intuitive positional chess does not directly reflect much silicon influence. This is not the case for many of his contemporaries, however, something we will examine more closely later on.
BEFORE MOVING to my own experiences facing chess machines, it’s worth taking a look at the history of this long-running rivalry. Despite my personal investment in such competitions during my career, looking back I can say that the sporting aspect is less interesting than how much we can learn about artificial intelligence and human cognition from the history of computer chess, and especially the competitions between computers and strong humans.
This is not because of how our silicon creations inevitably surpassed us over the board, as much as a holy grail as it was. Nor are many of the games themselves particularly fascinating for nonexperts. The most interesting games are those that represent advances in computer play in some way, because they reflect scientific progress. It’s unavoidable that the results will get most of the attention, but it’s important to look beyond the wins and losses. In order to use chess as a way to better understand what computers and humans are good at and what they struggle with and why, the moves matter more than the results.
Thanks to the international rating system we use in chess to rank players, a simple chart can show us quite clearly that chess machines have gotten stronger on a steady linear path from the first mainframes to specialized hardware machines to the top programs today. They went from novice level in the 1960s to strong play in the seventies to Grandmaster level in the late eighties and world champion level in the late nineties. There were no giant leaps, just a slow and steady evolution as the global community of developers learned from each other and competed with each other while Moore’s law worked its inexorable magic on their hardware.
This growth of machines from chess beginners to Grandmasters is also a progression that is being repeated by countless AI projects around the world. AI products tend to evolve from laughably weak to interesting but feeble, then to artificial but useful, and finally to transcendent and superior to human.
We see this path with speech recognition and speech synthesis, with self-driving cars and trucks, and with virtual assistants like Apple’s Siri. There is always a tipping point at which they go from amusing diversions to essential tools. Then there comes another shift, when a tool becomes something more, something more powerful than even its creators had in mind. Often this is the result of a combining of technologies over time, as in the case of the Internet, which is really a half-dozen different layers of technology working together.
It’s remarkable how quickly we go from being skeptics to taking a new technology for granted. Despite the rapid pace of technological change that has been the norm for our entire lives, we are briefly amazed, or horrified, or both, by anything new, only to get used to it in just a few years. It’s important to keep our heads on straight during that exciting cusp period between shock and acceptance so that we may look ahead clearly and prepare the best we can.
NINE DAYS before I was born in Baku, twenty-two years before I faced thirty-two computers at the same time in Hamburg and thirty-four years before my fateful rematch with Deep Blue, the first recorded game between a chess machine and a human Grandmaster (GM) took place in Moscow. The encounter has largely been forgotten, and it certainly isn’t worth being remembered for its chess merits, but it was a landmark nonetheless.
Soviet GM David Bronstein, who passed away in 2006, was a kindred spirit to me in many ways. He was always one of the most curious and experimental minds in chess both on and off the board, and he occasionally ran afoul of the Soviet authorities for his candid nature. Bronstein proposed many innovative ideas for promoting chess and even new variations of the game itself. He was interested in chess computers and artificial intelligence right from the start, and was always eager to play against the newest generation of programs. Bronstein also saw the potential of computer chess to provide insight into how humans think and he wrote many articles on computer chess as his professional playing career wound down.
In 1963, Bronstein was still one of the strongest players in the world, a dozen years removed from drawing a world championship match with the mighty Botvinnik. On April 4, 1963, at the Moscow Institute of Mathematics, he played a full game against a Soviet program running on a Soviet M-20 mainframe computer. I would like to have been able to ask Bronstein what he felt as the first moves were made. He couldn’t have been completely sure that the machine played like a beginner. It was a step into the unknown, with no way to prepare for this unique opponent.
It quickly turned out that, to adapt Samuel Johnson’s famous quip, the surprise wasn’t that the computer played chess well, but that it did it at all. Bronstein played aggressively and toyed with the feeble machine. He allowed the computer to win some material while he moved his pieces into attacking position and flushed out the black king. He finished with a pretty mate in ten, ending the game in just twenty-three moves.
Bronstein’s win against M-20 was an urtext of the first generation of (strong) human versus machine chess: the computer gets greedy and is punished. Early programs’ evaluation functions were heavily weighted toward material value. That is, which side has more pieces and pawns. It’s the easiest factor to evaluate and to program; assign a value to everything on the board and count—and computers are very good at counting. The basic set of values was established two centuries ago: pawns are worth one; knights and bishops are worth three; rooks are worth five; the queen is worth nine.
The king is trickier because, while it’s not so powerful in terms of mobility, it must be protected at all costs. The king cannot be captured and if it cannot escape inevitable capture, the game is over: checkmate. One trick is to assign the king a value of one million so the program knows not to put it in danger. Checkmate is an unambiguous and terminal event, another thing computers understand very well. If there is a way to force checkmate in four moves, a computer that looks four moves deep will find it no matter how complicated the position would look to the human eye.
Focusing only on material is also how novice humans play, especially kids. They care only about gobbling up their opponent’s pieces and ignore other factors in the position, such as piece activity and whose king is safer. Eventually they learn from experience that, while material is important, it doesn’t matter how many of your opponent’s pieces you’ve captured if your king is getting checkmated.
Even the scale of material values is full of exceptions based on the type of position on the board. For example, a well-placed knight can be worth as much or more than a rook with limited scope. During the middlegame—the dynamic, tactical phase of the game—a bishop is likely to be more valuable than three pawns, while the tables can turn in the endgame. Adjusting the various values during a game is possible, but that also adds even more knowledge to the algorithm, slowing down its search.
Early chess machines couldn’t learn from experience the way people can. Those greedy kids are learning each time they get checkmated. Even when they lose horribly, they are accumulating useful patterns in their memory. Computers, meanwhile, would make the same mistake over and over, something their human opponents understood and exploited quite well. Even well into the 1980s, if you timed it just right you could replay an entire game against a computer, beating it the same way move for move.
Timing matters because from one microsecond to the next as its search expands, the computer may switch to a different move. A human spending sixty seconds on each move is very unlikely to play much differently than if he spent fifty-five seconds per move, but this isn’t true for computers since every sliver of time is put directly into deeper search, with a linear payoff in higher-quality moves.
The apparent similarity between early chess programs and human beginners is a trap, part of the familiar fallacy of expecting computers to think like humans. As Moravec’s paradox dictates, computers are very good at chess calculation, which is the part humans have the most trouble with. Computers are poor at recognizing patterns and making analogical evaluations, a human strength. Other than checkmate, nearly every factor that goes into evaluating a chess position is conditional on many other factors. This, along with the slow speed of computers at the time, is why early experts thought it would be impossible to make a strong Type A (brute force) program.
They were wrong, although it would take a while to figure that out. Many of the first programs were attempts at Type B, which sought to intelligently reduce the size of the algorithm’s search tree early on the way humans do. Other research groups saw the advantage of tackling the relatively concrete task of improving the machine’s search speed and therefore depth, which always improved strength in a predictable way.
The first program that played competent chess was developed at MIT in the late 1950s, a few years ahead of the Soviet program beaten by Bronstein. The Kotok-McCarthy program ran on an IBM 7090 and included some of the techniques that would become the basis for every strong algorithm that followed, including alpha-beta pruning to speed up the search.
The leading Soviet team at the time took a Type A approach, which is interesting considering that they were surrounded by strong chess players, unlike the Americans. Alan Kotok and John McCarthy were both very weak players and had a romantic view of how the game was played. To me, the Soviet embrace of brute force search is not ironic at all, but, to the contrary, reflects a superior understanding of how good chess is played and won. Chess is a very precise game when played well. The advantage of a single pawn is usually more than enough for victory between strong players. Weak players see chess through the lens of their own limitations and frequent mistakes. A novice or nonplayer sees the game as a roller-coaster of cut and thrust, full of blunders on both sides that swing the game this way and that.
If you are designing a chess machine with that romantic vision of the game in mind, scientific precision is less important than moments of inspiration. Occasional blunders aren’t so bad if you are counting on your opponent to return the favor, which means there is an element of self-fulfilling prophecy. Type B thinking assumes that the entire system is chaotic and noisy to begin with and just tries to make the best of it by selecting the moves to focus on very early on. Instead of looking at the best twenty moves, or ten, and going from there, the Kotok-McCarthy program started out very narrowly, with just four moves. That is, looking ahead one ply it picked the four best moves and then figured out the three best replies. Then it looked at the two best replies to those moves, etc., getting deeper and narrower.
By design, this is superficially similar to how a strong human player’s analysis works, but it ignores that a master’s mind can do it effectively only because its assessment of thousands of patterns and the immense parallel processing power of the human brain are choosing that initial menu of three or four candidate moves with formidable accuracy. Expecting a machine to select the right few moves to focus on via calculation, without the benefit of all that experience, is closer to blindfold darts than to blindfold chess.
One of the many handy aspects for chess as an AI laboratory is that we have a good way to measure progress and to test competing theories: over the chessboard! The Soviets started later than the Americans, but their program ITEP had been in development more recently by the time they played a match over telegraph in 1966–67. The ITEP machine, named for the Institute for Theoretical and Experimental Physics in Moscow, was Type A and turned out to be too accurate for the outdated Kotok-McCarthy program and won the match with a score of 3–1.
Around this time, American programmer Richard Greenblatt built on the Kotok-McCarthy concepts with his much better chess understanding, widening the search dramatically. His program Mac Hack VI started with a search width of 15, 15, 9, 9, compared to the Kotok-McCarthy 4, 3, 2, 2. This had the effect of reducing the level of “noise” and making the program far more accurate and stronger. Mac Hack VI also added a database of thousands of opening moves and would become the first computer program to play in a human chess tournament and to receive a chess rating. But despite these improvements and successes, the days of Type B programs were numbered, even more so than those of humans. Brute force was coming.
I WAS INTRODUCED to computers in 1983, although I didn’t play chess with them at the time. The British computer company Acorn, the “British Apple,” sponsored my match against Viktor Korchnoi in London that year, and of course their products were on display. Businesses, hobbyists, and other early adopters across Europe were paying large sums for the first few generations of home computers and Acorn was doing very well. I won the match, putting me a step away from my first world championship contest with Anatoly Karpov the next year, and was also given an Acorn home computer to take back to Baku. I flew on Aeroflot sitting next to the Soviet ambassador, and my fragile new trophy had its own VIP seat and blanket.
To me, coming from the USSR, owning a computer seemed a little like science fiction. First, I had dedicated my life to climbing up the chess Olympus and this left very little time for other interests. Second, the USSR was still a computing desert outside of research institutions. A Soviet clone of the 1977 Apple II, the AGAT, came out around 1983 and slowly started to appear in schools across the country, but it was far out of reach for most private citizens, costing around twenty times the average monthly Soviet salary. And like most Soviet knock-off tech, it wasn’t even a very good clone of a computer that was already six years old. America’s BYTE magazine wrote in 1984 that “the AGAT wouldn’t stand a chance in today’s international market, even if they gave it away.”
This was far from just a little Cold War jab. The PC revolution was already well under way in America by this time. They were still expensive for what you got, but easily available to the middle class. The hugely popular Commodore 64 was released in August 1982. The standard-setting IBM PC XT came out in early 1983. By late 1984, over 8 percent of American households owned a computer. For comparison, the number of personal computers in Baku, Azerbaijan, a capital city of over a million people, probably went from zero to one when the plane landed with me and my Acorn.
I would like to say this first encounter with a computer was a transformative moment, but as I said, I was a little busy at the time. My cousins and friends mostly used my eight-bit Acorn, a BBC Micro model, I believe, to play video games. One in particular would come to alter my perception of computers and my life in an important way, but it wasn’t a chess game. It involved moving a little green frog across traffic.
One day early in 1985, I received a package from a stranger named Frederic Friedel, a chess fan and science writer based in Hamburg, Germany. He sent me a nice note and a floppy disk containing several computer games, including my new favorite, called Hopper. I admit I spent much of my free time over the next few weeks playing Hopper and setting ever-higher record scores.
A few months later, I traveled to Hamburg for several events, including the computer simul, and I also visited Mr. Friedel at his suburban home. I met his wife and two young sons, Martin, age ten, and Tommy, age three. They made me feel quite at home and Frederic was eager to show me the latest developments on his own computer. I managed to work into the conversation that I had completely mastered one of the little games he had sent me.
“You know, I’m the best Hopper player in Baku,” I said, omitting any mention of the total lack of competition. I told him that I had scored sixteen thousand points and was a little surprised that this extraordinary number failed to elicit at least a raised eyebrow.
“Very impressive,” Frederic said, “but that’s not such a big score in this house.”
“What? You can beat it?” I asked.
“No, not me.”
“Ah, okay, Martin must be the video game whiz.”
“No, not Martin.”
It was with a sinking feeling that I realized the smile on Frederic’s face meant that the household Hopper champion was the three-year-old. I was incredulous. “You can’t mean Tommy!” My fears were confirmed when Frederic led his little boy over to the computer and sat him down next to us as the game loaded. Since I was the guest they let me go first and I rose to the occasion with a personal best of nineteen thousand points.
My success was short-lived, however, as Tommy took his turn. His little fingers were a blur and it wasn’t long before the score read twenty thousand, then thirty thousand. I conceded defeat to avoid having to sit watching through dinnertime.
Losing to a little kid at Hopper was easier on my ego than any loss to Karpov, but it still gave me food for thought. How was my country going to compete with a generation of little computer geniuses being raised in the West? Here I was, one of the few people in a major Soviet city with a computer, and I had been handily outperformed by a German toddler.
And so, when I signed a sponsorship deal with the computer company Atari in 1986, I took as payment over fifty of their newest machines to bring back to form a youth computer club in Moscow, the first of its kind in the Soviet Union. I continued to supply the club with hardware and software acquired on my travels and it became a hub for many talented scientists and hobbyists.
They would often give me lists of equipment they wanted for their projects, leading to some amusing scenes at the airport when I would return from my travels like Father Christmas delivering presents. Mixed in with the chess fans welcoming me home, there would be computer experts hoping I’d managed to find the items on their wish lists. I even recall being met by a shout that would get quite a lot of attention from security at any airport today: “Garry! Did you bring the Winchester?!” It was a much-coveted type of hard drive.
Frederic and I also had the chance to talk about the potential implications computers had for professional chess. Businesses were rapidly adopting PCs for spreadsheets, word processing, and databases, so why couldn’t this sort of Hoppering be done for chess games? This would be a powerful weapon, one that I couldn’t afford to be the last to have.
As described above, our conversations led to the creation of the first version of ChessBase, a name that soon became synonymous with professional chess software. In January 1987, I tried out an early version of the program to prepare for a special simultaneous exhibition against a strong team. I had narrowly lost a similar event in 1985, playing against eight members of a professional German league team at the same time. I had come in tired and overconfident, especially since I didn’t know much about most of my opponents and had no way to quickly prepare for them.
For the rematch, I discovered how much ChessBase was going to change professional chess and my life. Using an Atari ST and a ChessBase diskette labeled “00001” that I was given by Frederic and Matthias, I was able to bring up and review my opponents’ previous games in hours, a process that would have taken weeks without a computer. With just two days of preparation I felt comfortable going into the match and won in crushing fashion, 7–1. That was when I knew I was going to be spending a lot of time in front of a computer for the rest of my career. I just didn’t realize yet how much of that time would be spent playing against them.
HOW QUICKLY and completely computers came to dominate chess preparation was illustrated a few years later when an interviewer and photographer came to where I was staying. The photographer wanted some pictures of me at a chessboard to accompany the story. The only problem? I didn’t have a chessboard with me! All my preparation was done on my laptop, a Compaq that really stretched the definition of “portable.” It must have weighed close to twelve pounds. Even so, it was far lighter and more efficient than traveling with my paper notebooks and a stack of opening encyclopedias. The advantages would accumulate when the Internet made it possible to download the latest games nearly as soon as they had been played, instead of having to wait weeks or months for them to be published in a magazine.
Soon nearly every Grandmaster traveled to every tournament with a laptop, although there was a jagged generational break in this regard. Many older players found them too complicated, too alien, especially after having decades of success with their traditional training and preparation methods. Laptops were also still very expensive, and few players had my advantages of sponsorship deals and world championship prize purses.
How professional chess changed when computers and databases arrived is a useful metaphor for how new technology is adopted across industries and societies in general. It’s a well-established phenomenon, but I feel that the motivations are underanalyzed. Being young and less set in our ways definitely makes us more open to trying new things. But simply being older isn’t the only factor that works against this openness—there is also being successful. When you have had success, when the status quo favors you, it becomes very hard to voluntarily change your ways.
In my lectures to business audiences I call this the “gravity of past success,” and often give a painful example from my own career: the loss of my world championship title to Vladimir Kramnik in 2000. I was at the height of success at the time, in the midst of an unprecedented winning streak at top-level tournaments and raising my rating to its highest peak ever. I felt great and had prepared deeply for our October match in London, scheduled for sixteen games. Kramnik was my most dangerous opponent, twelve years younger and with years of strong performances against me. But it was his first world championship match and my seventh. I had experience, better results, and felt good. How could I lose?
The answer was “by playing into my opponent’s strength and refusing to adapt.” Kramnik had prepared very cleverly, using his turns with the black pieces to draw me into tedious positions I disliked. This was entirely to his credit, and it was up to me to find a strategic response for the rest of the match. But instead of avoiding these positions entirely and playing to my strengths, I continued to charge straight ahead like a bull at a red cape. I eventually lost the match with two losses, thirteen draws, and without winning a single game.
I was thirty-seven at the time, not exactly ancient. And I was never afraid of pushing myself to stay on the cutting edge, including my embrace of technology. My weakness was a refusal to admit that Kramnik had outprepared me—preparation was supposed to be my strong suit. Each one of my successes up to that moment was like being dipped in bronze over and over, each success, each layer, making me more rigid and unable to change, and, more importantly, unable to see the need to change.
This metaphorical gravity isn’t only a problem for individuals, or only a matter of ego. Fighting against disruption and change is also a standard business practice, one that is usually employed by a market leader trying to protect that lead. There are countless examples of this from the real world, but I’ll take one ad absurdum case from science fiction, the 1951 movie The Man in the White Suit, starring Alec Guinness. Guinness, the protagonist, is a rogue research chemist who invents a miracle fiber that never wears out and never gets dirty. Instead of the fame, riches, and Nobel Prize you might expect, he ends up being chased through the streets by angry mobs once various interest groups realize what his invention will mean. No more demand for new cloth, so the textile industry will be wiped out along with thousands of union jobs. No more need for laundry soap or laundry workers, who join in the pursuit.
Far-fetched? Certainly, but I don’t think you have to have my suspicious mind to wonder if lightbulb companies would sell an indestructible and everlasting bulb if they could make one. But resisting change and delaying it to squeeze a few more dollars out of an existing business model usually just makes the inevitable fall all the worse. I once made a television commercial for the search engine company AltaVista in 1999, but that didn’t mean I wanted to follow it to oblivion when the chess equivalent of Google came along.
I was in my twenties when the digital information wave rolled over the chess world, and it was a fairly gradual one, not a tsunami. Flicking through games on a screen was far more efficient than on printed materials, a real competitive advantage but not a nuclear bomb. The impact of the Internet a few years later was just as great, dramatically accelerating the information warfare that Grandmasters wage against each other over the board. A brilliant new opening idea played in a game in Moscow on Tuesday could be imitated by a dozen players around the world on Wednesday. It shortened the lifespan of these secret weapons, what we call opening novelties, from weeks or months to hours. No more could you hope to ensnare more than a single opponent with a clever trap.
Of course, that was only true if your opponents were also online and up to date, which wasn’t the case for a while. Asking a fifty-year-old Grandmaster to ditch his beloved leather-bound notebooks of analysis, printed tournament bulletins, and other preparation habits was like asking a successful writer to switch to a word processor or an artist to start drawing on a screen instead of a canvas. But in chess, it was a matter of adapting to survive. Those who quickly mastered the new methods thrived; the few who didn’t mostly dropped down the rating lists.
There’s no way to prove causation, but I’m certain that the rapid decline of many veteran players in the 1989–95 span, when ChessBase became standard, had much to do with their inability to adjust to the new technology. The 1990 rating list included over twenty active players born before 1950 among the top one hundred in the world. By 1995, there were just seven, and only one among the elite: the ageless Viktor Korchnoi, born in 1931, who was my opponent in that 1983 London candidates match sponsored by Acorn. Another exception was my great rival Karpov, born in 1951, who stayed near the top into his fifties despite his personal reluctance to embrace computers and the Internet. But along with his tremendous talent and experience, as a former world champion with considerable resources he also counted on the assistance of colleagues for his research, an advantage few others had. Reducing the advantage of being able to afford assistants, or “seconds,” as they are called in chess, in a tribute to the age of duels, was one of the many democratizing impacts technology had on the chess world.
While they may have shortened the careers of a few older players, computers also enabled younger players to rise more quickly. Not just the playing engines, but because of how PC database programs allowed elastic young brains to be plugged into the fire hose of information that was suddenly available. Even I am startled to watch kids zipping from one game to the next, one branch of analysis to another, in the blink of an eye. Computer-centric training also has drawbacks, and I’ll get to those later, but there is no doubt it tipped the playing field, or the chessboard, even further toward youth. As my professional career progressed, not only would I be facing the challenge of every champion to fend off the next generation of players, but it would be a generation that had grown up with sophisticated tools that hadn’t existed when I was a kid.
I was born just in time to ride this wave instead of being swept away by it. But this timing also put me on the front lines against a new enemy that was growing stronger by the day. The chess machines were finally coming for the world champion, and, as of November 9, 1985, that was me.
WHEN WILL a chess-playing machine be able to beat the world champion?” This question was put to every chess programmer in history dozens of times. As you could expect, the earliest predictions, from the days of the digital computer’s infancy, were wildly off the mark. At least the Carnegie Mellon group’s daring 1957 promise of 1967 was in some ways avenged since it was a group from the same school whose Deep Blue eventually did the job—if forty years later instead of ten.
At the twelfth annual North American Computer Chess Championship, held in Los Angeles in 1982, the world’s best chess machines battled each other for supremacy. Ken Thompson’s special-purpose hardware machine Belle continued to show its superiority over the rest, and to show the potential for a hardware architecture and customized chess chips that would later be realized by Deep Blue. Thompson, with Belle codeveloper Joe Condon, worked at the famous Bell Laboratories and, among many other accomplishments, was one of the creators of the Unix operating system.
As far as results go, Belle was the definitive answer to the dilemma Claude Shannon presented in 1950 between “fast but dumb” Type A brute force and “smart but slow” Type B artificial intelligence programs. It was now clear the brute force, with a fast-enough search, was enough to play very strong chess. Despite a relative lack of knowledge and other evaluation limitations, Belle’s raw speed, up to 160,000 positions per second, produced results that were leaving smarter microprocessor machines and even Cray supercomputers in the dust. Interviews with various computer chess luminaries at the 1982 event about when a machine would defeat the world champion (then Karpov) revealed cautious optimism.
Monty Newborn, long one of the motive forces behind computer chess, especially as a promoter and organizer, was remarkably optimistic with his answer of five years. Another expert, Mike Valvo, who was also an International Master, said ten years. The creators of the popular PC program Sargon nailed it exactly with fifteen. Thompson thought it was still twenty years off, putting him on the pessimistic side of the vast majority that said it would happen around the year 2000. A few even said it would never happen, reflecting some of the problems even the faster machines were having with the law of diminishing returns that arose when adding chess knowledge to their creations. But this was the last time that the question was “when or whether” instead of just “when.”
By the late 1980s, after another decade of steady progress, the computer chess community was well aware that time was on their side in the human versus machine contest and they could narrow the range of their prognostications effectively. In 1989, at the World Computer Chess Championship in Edmonton, Canada, a survey of the forty-three experts present reflected the recent achievements in human-machine play. A computer had just beaten a Grandmaster in tournament play for the first time the year before and the road map of further improvement was coming into sharp focus: a little more knowledge and a lot more speed. Still, only one correctly picked 1997 as the year of destiny, while other guesses ranged from within a decade of that. Notable was a member of the Deep Blue team, Murray Campbell, guessing 1995, and Claude Shannon himself saying 1999.
It’s a little unfair to highlight the early erroneous forecasts and spurious rationales that came out of the computer chess community over the years. After all, human calculation may be weak, but our hindsight is always perfect. But there is a point to it, since in many cases their sins, both their overoptimism or resigned pessimism, are a distant mirror for today’s flood of predictions about artificial intelligence.
Overestimating the potential upside of every new sign of tech progress is as common as downplaying the downsides. It’s easy to let our imaginations run wild with how any new development is going to change everything practically overnight. The unforeseen technical roadblocks that inevitably spring up are only one reason for this consistent miscalculation. Human nature is simply out of sync with the nature of technological development. We see progress as linear, a straight line of improvement. In reality, this is only true with mature technologies that have already been developed and deployed. For example, the way Moore’s law accurately described the advances in semiconductors, or the way solar cell efficiency is improving at a slow but steady pace.
Before that predictable progress phase, there are two previous phases: struggle and then breakthrough. This fits the axiom of Bill Gates, “We always overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten.” We expect linear progress, but what we get are years of setbacks and maturation. Then the right technologies combine or a critical mass is reached and boom, it takes off vertically for a while, surprising us again, until it reaches the mature phase and levels off. Our minds see tech progress as a straight diagonal line, but it’s usually more of an S-shape.
The chess machines of the fifties and sixties were still in the struggle phase. Researchers were doing a lot of experimentation with few resources, still trying to figure out if Type A or Type B was the most promising, all while using primitive coding tools on hardware that was incredibly slow. Was chess knowledge the key? Was speed the most important factor? With so many of the basic concepts still up in the air, each new breakthrough felt like it could be the big one.
One strong chess player decided he could put the scientists’ optimism to good personal advantage. Long before I took my turn as the computer chess world’s “most wanted,” a Scottish International Master named David Levy turned beating them into a profitable sideline. In 1968, after hearing two eminent AI experts predict that a machine would beat the world champion in a decade, Levy made a famous bet that no computer would be able to defeat him in a match in that span. If you looked at the modest progress chess machines made in the two decades after Claude Shannon created the road map in 1949, you could see his point.
(To quickly clarify some terminology, at around 2400 in rating, an International Master is above master (2200) and a rank below Grandmaster (2500 and up). Twenty-seven hundred is considered elite today, with around forty players in the world surpassing that mark, on up to Magnus Carlsen’s record of 2882. My peak was 2851 in 1999 and my rating was 2795 when I played my second match with Deep Blue. It should be noted that ratings are getting higher over time: Bobby Fischer’s 1972 peak of 2785 was like Mount Everest in its day, but quite a few players have surpassed that number, while I cannot say they have surpassed Fischer. We say “match” for a series of games between two opponents, as opposed to a tournament with many players.)
Levy was much stronger than the programs of the early 1970s; no program would approach master level until the bet came due. What’s more, Levy was also very savvy about the strengths and weaknesses of computer chess players. He understood that while they were getting quite dangerous in tactical complications thanks to their increasingly deep search abilities, they were clueless about strategic plans and the subtleties of endgame play. He would maneuver patiently, employing an anti-computer strategy of “doing nothing, but doing it well” until the machine would overextend and create weaknesses in its own position. Then Levy would clean up on the board—and in his bets.
It looked like smooth sailing for Levy until the appearance of a program from Northwestern University, called simply “Chess.” The program by Larry Atkin and David Slate was the first chess machine to play the strong, consistent chess needed to beat experts without a serious human blunder. By 1976, version 4.5 of Chess was good enough to win a weak human tournament. The next year, 4.6 finished first in an open tournament in Minnesota, reaching a rating performance near expert level, if not quite master.
The struggle phase of development was over and the rapid growth phase had begun. The combination of faster hardware and twenty years’ worth of programming improvements had crested. After decades of disappointments after overvaluing potential advances, real progress came faster than anyone expected. When the time came for Levy to meet the computer world champion in 1978, Chess 4.7 was far stronger than he imagined any machine would be by that time. It wasn’t quite strong enough, however, although it did score a draw and a win against him in their six-game match.
Levy went on to be an important force in the machine chess world and has written countless books and articles on the subject. He’s the president of the International Computer Games Association (ICGA), the organization that oversaw my 2003 match against the program Deep Junior in New York City. In 1986, Levy wrote an article in the ICGA Journal titled “When Will Brute Force Programs Beat Kasparov?” I think he was quite glad to have the target on his back transferred to someone else.
Levy collected his winnings and threw down a new gauntlet, a bounty of $1,000 to the computer that could beat him. The American science magazine Omni sweetened the pot with another $4,000. It would be another ten years before someone collected that money, a group of grad students at Carnegie Mellon with a custom hardware–based chess machine called Deep Thought.