Читать книгу Bounce: The Myth of Talent and the Power of Practice - Matthew Syed, Matthew Syed - Страница 16

At 3.00 p.m. on 10 February 1996, Garry Kasparov strode into a small room in the Pennsylvania Convention Center to contest one of the most anticipated chess matches in history. He was smartly dressed in a dark suit and white shirt and wore a look of intense concentration. As he sat down at the match table, he glanced across the board to the man on the other side: Dr FengHsuing Hsu, a bespectacled Taiwanese-American with a quizzical expression.

In the room, besides Kasparov and Hsu, were three cameramen, one match official, three members of Kasparov’s entourage, and a technical adviser. A strict silence was enforced, with the five hundred spectators packed into a nearby lecture hall to witness the event on screens fed from three TV cameras and live commentary from grandmaster Yasser Seirawan. The atmosphere was, by common consent, quite unlike that of any other chess match in living memory.

Kasparov is almost universally considered to be the greatest player in the history of the sport. His ELO rating – an official score measuring relative skill – remains the highest ever recorded: 71 points higher than that of Russian grandmaster Anatoly Karpov, and 66 higher than that of the great American player Bobby Fischer. Kasparov, at the time of the contest, had been the world number one for ten straight years, and his mere presence before a chessboard was enough to intimidate some of the world’s most revered grandmasters.

But his opponent on this day was susceptible neither to intimidation nor the other mind games for which Kasparov was famous. His opponent was oblivious to Kasparov’s status and reputation for guile and audacity. Indeed, his opponent was not even in the room, but many miles away in a large, dimly lit building in Yorktown Heights, New York. His opponent was a computer. Its name was Deep Blue.

The media, rather predictably, hyped the match as an historic showdown between man and machine. ‘The future of humanity is on the line,’ declared one newscaster. ‘The match goes further than mere chess, presenting a challenge to mankind’s sovereignty,’ intoned USA Today. Even Kasparov seemed to be seduced by the apocalyptic tenor of the pre-match hype, saying, ‘This is a mission to defend human dignity... It is species-defining.’

Kasparov’s opening move, pawn to C5, was typed into a computer adjacent to the match table by Mr Hsu (the brains behind the development of Deep Blue, on behalf of electronics giant IBM) and then transmitted across to the IBM Center in New York by a relatively new technology called the Internet.

At this point Deep Blue sprang into action. Powered by 256 specially developed chess processors operating in parallel, 32 concentrated on each eight-square section of the board, it was able to compute more than 100 million positions per second. A few moments later, Deep Blue’s response came winging its way across the ether, and Mr Hsu dutifully executed the instruction: pawn to C3.

For six games over eight days, the thrust and counterthrust between man and machine was beamed to a captivated world. Kasparov, an eccentric and hot-tempered Azerbaijani, was famous for his histrionics, often growling and shaking his head vigorously. Many had criticized Kasparov’s antics, accusing him of deliberately trying to disturb adversaries. But Kasparov was no less animated against his machine opponent, often rising from his chair to pace the room.

Just before the fortieth move in the final game on 17 February, Kasparov took his watch from the table and put it on his wrist. This was a familiar sign that the world champion believed the match was nearing its conclusion. The audience in the lecture hall held its breath. Three moves later Dr Hsu rose slowly to his feet and offered his hand to his opponent. The audience burst into wild applause.

Kasparov had triumphed.

The question is: How? How could a man unable to search more than three moves per second (this represents the current limit of human capacity) defeat a machine whose computing speed was measured in the tens of millions? The answer, as we shall see, will help us to unlock some of the deepest mysteries of expert performance, both within sport and in the wider world.

In the 1990s Gary Klein, a New York psychologist, embarked on a major study funded by the US military to examine decision-making in the real world. He was looking to test the theory that expert decision-makers wield logical methods, examining the various alternatives before selecting the optimal choice. Klein’s problem was that the longer the study went on, the less the theory bore any relation to the way decisions are made in practice.

The curious thing was not that top decision-makers – medical professionals, firefighters, military commanders, and so on – were making choices based on unexpected factors; it was that they did not seem to be making choices at all. They were contemplating the situation for a few moments and then just deciding, without considering the alternatives. Some were unable even to explain how they happened upon the course of action they actually took.

Here is an example of a fire lieutenant making a life-saving decision, as recounted in Klein’s book Sources of Power: How People Make Decisions:

There is a simple house fire in a one-storey house in a residential neighbourhood. The fire is in the back, in the kitchen area. The lieutenant leads his hose crew into the building, to the back, to spray water on the fire, but the fire just roars back at them.

‘Odd,’ he thinks. The water should have more of an impact. They try dousing it again, and get the same results. They retreat a few steps to re-group.

Then the lieutenant starts to feel as if something is not right. He doesn’t have any clues; he just doesn’t feel right about being in that house, so he orders his men out of the building – a perfectly standard building with nothing out of the ordinary.

As soon as his men leave the building, the floor where they had been standing collapses. Had they still been inside, they would have plunged into the fire below.

Later, when Klein asked the commander how he knew something was about to go terribly wrong, the commander put it down to ‘extrasensory perception’. That was the only thing he could come up with to explain a life-saving decision, and others like it, that seemed to emerge from nowhere. Klein was too much of a rationalist to accept the idea of ESP, but by now he had begun to notice equally perplexing abilities among other expert decision-makers. They seemed to know what to do, often without knowing why.

One of Klein’s co-workers, who had spent many weeks studying the neonatal unit of a large hospital, had found that experienced nurses were able to diagnose an infection in babies even when, to outsiders, there seemed to be no visible clues. This was not merely remarkable, but often life-saving: infants at an early stage of life can quickly succumb to infections if they are not detected early.

Perhaps the most curious thing of all was that the hospital would perform tests to check the accuracy of the nurse’s diagnosis, and occasionally these would come back negative. But sure enough, by the next day, the tests would come back positive – the nurse had been right all along. To the researcher this seemed almost magical, and even the nurses were baffled by it, attributing it to ‘intuition’ or a ‘special sense’.

What was going on? Can the insights gleaned from sport help to unlock the mystery?

Think back to Desmond Douglas, the Speedy Gonzales of English table tennis, who could anticipate the movement of a table tennis ball by chunking the pattern of his opponent’s movement before the ball was even hit. Think, also, of how other top performers in sport seem to know what to do in advance of everyone else, creating the so-called time paradox where they are able to play in an unhurried way even under severe time constraints.

Klein came to realize that expert firefighters are relying on precisely the same mental processes. They are able to confront a burning building and almost instantly place it within the context of a rich, detailed, and elaborate conceptual scheme derived from years of experience. They can chunk the visual properties of the scene and comprehend its complex dynamics, often without understanding how. The fire commander called it ‘extrasensory perception’; Douglas, you will remember, cited his ‘sixth sense’.

We can get an idea of what is going on by digging down into the mind of the fire commander who pulled his men out moments before the floor caved in. He did not suspect that the seat of the fire was in the basement, because he did not even know the house had a basement. But he was already curious, based upon his extensive experience, as to why the fire was not reacting as expected. The living room was hotter than it should have been for such a small fire, and it was altogether too quiet. His expectations were breached, but in ways so subtle he was not consciously aware of why.

Only with hindsight – and after hours of conversation with Klein – was it possible to piece together the sequence of events. The reason the fire was not quenched by his crew’s attack was because its base was underneath them, and not in the kitchen; the reason it was hotter than expected was because it was rising from many feet below; the reason it was quiet is because the floor was muffling the noise. All this – and many more interconnecting variables of indescribable complexity – was responsible for the fire commander taking the life-saving decision to pull his men.

As Klein explains, ‘The commander’s experience had provided him with a firm set of patterns. He was accustomed to sizing up the situation by having it match one of these patterns. He may not have been able to articulate the patterns or describe their features, but he was relying on the pattern-matching process to let him feel comfortable that he had the situation scoped out.’

A set of painstaking interviews with the nurses in the neonatal unit provided the same insights. In essence, the nurses were relying on their deep knowledge of perceptual cues, each one subtle, but which together signalled an infant in distress. The same mental process is used by pilots, military generals, detectives – you name it. It is also true, as we have seen, of top sportsmen. What they all have in common is long experience and deep knowledge.

For years, knowledge was considered relatively unimportant in decision-making. In experiments, researchers would choose participants with no prior experience of the area under examination in order to study the ‘cognitive processes of learning, reasoning, and problem solving in their purest forms’. The idea was that talent – superb general reasoning abilities and logical prowess – rather than knowledge makes for good decision-makers.

This was the presumption of top business schools and many leading companies, too. They believed they could churn out excellent managers who could be parachuted into virtually any organization and transform it through superior reasoning.

Experience was irrelevant, it was said, so long as you possessed a brilliant mind and the ability to wield the power of logic to solve problems. This approach was seriously misguided. When Jeff Immelt became the chief executive of General Electric in 2001, he commissioned a study of the best-performing companies in the world. What did they have in common? According to Geoff Colvin in Talent Is Overrated, ‘These companies valued “domain expertise” in managers – extensive knowledge of the company’s field. Immelt has now specified “deep domain expertise” as a trait required for getting ahead at GE.’

These insights have not just become central to modern business strategy; they also form the basis of artificial intelligence. In 1957 two computer experts created a programme they called the General Problem Solver, which they billed as a universal problem-solving machine. It did not have any specific knowledge, but possessed a ‘generic solver engine’ (essentially, a set of abstract inference procedures) that could, it was believed, tackle just about any problem.

But it was soon realized that knowledge-free computing – however sophisticated – is impotent. As Bruce Buchanan, Randall Davis, and Edward Feigenbaum, three leading researchers in artificial intelligence, put it: ‘The most important ingredient in any expert system is knowledge. Programmes that are rich in general inference methods – some of which may even have some of the power of mathematical logic – but poor in domain-specific knowledge can behave expertly on almost no tasks.’

Think back to the firefighters. Many young men and women are drawn to the profession because they think they’re good at making decisions under pressure, but they quickly discover they just can’t cut it. When they look at a raging fire, they are drawn to the colour and height of the flames and other perceptually salient features, just like the rest of us. Only after a decade or more of on-the-job training can they place what they are seeing within the context of an interwoven understanding of the patterns of fires.

The essential problem regarding the attainment of excellence is that expert knowledge simply cannot be taught in the classroom over the course of a rainy afternoon, or indeed a thousand rainy afternoons (the firefighters studied by Klein had an average of twenty-three years experience). Sure, you can offer pointers of what to look for and what to avoid, and these can be helpful. But relating the entirety of the information is impossible because the cues being processed by experts – in sport or elsewhere – are so subtle and relate to each other in such complex ways that it would take forever to codify them in their mind-boggling totality. This is known as combinatorial explosion, a concept that will help to nail down many of the insights of this chapter.

The best way to get a sense of the strange power of combinatorial explosion is to imagine folding a piece of paper in two, making the paper twice as thick. Now repeat the process a hundred times. How thick is the paper now? Most people tend to guess in the range of a few inches to a few yards. In fact the thickness would stretch eight hundred thousand billion times the distance from Earth to the sun.

It is the rapid escalation in the number of variables in many real-life situations – including sport – that makes it impossible to sift the evidence before making a decision: it would take too long. Good decision-making is about compressing the informational load by decoding the meaning of patterns derived from experience. This cannot be taught in a classroom; it is not something you are born with; it must be lived and learned. To put it another way, it emerges through practice.

As Paul Feltovich, a researcher at the Institute for Human and Machine Cognition at the University of West Florida, has explained: ‘Although it is tempting to believe that upon knowing how the expert does something, one might be able to teach this to novices directly, this has not been the case. Expertise is a long-term developmental process, resulting from rich instrumental experiences in the world and extensive practice. These cannot simply be handed to someone.’

All of which hints at the decisive advantage held by Kasparov over his machine opponent. Deep Blue had all the ‘talent’: the ability to search moves at a rate measured in tens of millions per second. But Kasparov, although limited to a derisory three moves per second, had the knowledge. A deep, fertile, and endlessly elaborate knowledge of chess: the configurations of real games, how they can be translated into successful outcomes, the structure of defensive and offensive positions, and the overall construction of competitive chess. Kasparov could look at the board and see what to do in the same way an experienced firefighter can confront a blazing building and see what to do. Deep Blue can’t.

It is worth noting something else here. You’ll remember that SF, the person who performed so well on the digit span task, was able to remember more than eighty numbers by relating them to his experiences as a competitive runner. The numbers 9 4 6 2, for example, became 9 minutes, 46.2 seconds – a very good time for running two miles. SF’s retrieval structure was, in effect, an ad hoc device derived from his life beyond the test.

Kasparov’s memory of chess positions, on the other hand, is embedded in the living, breathing reality of playing chess. When he sees a chessboard, he does not chunk the pattern by relating it to an altogether different experience but by perceiving it immediately as the Sicilian Defence or the Latvian Gambit. His retrieval structure is rooted within the fabric of the game. This is the most powerful type of knowledge, and is precisely the kind possessed by firefighters, top sportsmen, and other experts.

By now it should be obvious why Deep Blue’s gigantic advantage in processing speed was not sufficient to win – combinatorial explosion. Even in a game as simple as chess, the variables rapidly escalate beyond the capacity of any machine to compute. There are around thirty ways to move towards the beginning of a game, and thirty ways in which to respond. That amounts to around 800,000 possible positions after two moves each. A few moves after that, and the number of positions are measured in trillions. Eventually, there are more possible positions than there are atoms in the known universe.

To be successful, a player must cut down on the computational load by ignoring moves unlikely to result in a favourable outcome and concentrating on those with greater promise. Kasparov is able to do this by understanding the meaning of game situations. Deep Blue is not.

As Kasparov put it after winning game two of the six-game match: ‘Had I been playing the same game against a very strong human I would have had to settle for a draw. But I simply understood the essence of the end game in a way the computer did not. Its computational power was not enough to overcome my experience and intuitive appreciation of where the pieces should go.’

Gary Klein, the psychologist who studied the firefighters, wanted to double-check whether chess players really do make rapid decisions based on the perceptual chunking of patterns (as opposed to conducting brute-force searches, like computers).

He reasoned that if the chunking theory is correct, top chess players would make similar decisions even if the available time was dramatically reduced. So he tested chess masters under ‘blitz’ conditions, where each player has only five minutes on the clock, with around six seconds per move (in standard conditions there are forty moves in a ninety-minute period, allowing around two minutes, fifteen seconds per move).

Klein found that, for chess experts, the move quality hardly changed at all in blitz conditions, even though there was barely enough time to take the piece, move it, release it, and hit the timer.

Klein then tested the pattern-recognition theory of decision-making directly. He asked chess experts to think aloud as they studied mid-game positions. He asked them to tell him everything they were thinking, every move considered, including the poor ones, and especially the very first move considered. He found that the first move considered was not only playable but also in many cases the best possible move from all the alternatives.

This obliterates the presumption that chess is exclusively about computational force and processing speed. Like firefighters and tennis players, chess masters generate usable options as the first ones they think of. This looks magical when you first see it (particularly when chess masters are playing lots of games simultaneously), but that is because we have not seen the ten thousand hours of practice that have made it possible.

It is a bit like learning a language. At the beginning, the task of remembering thousands of words and fitting them together using abstract rules of grammar seems impossible. But after many years of experience, we can look at a random sentence and instantly comprehend its meaning. It is estimated that most English language users have a vocabulary of around 20,000 words. American psychologist Herbert Simon has estimated that chess masters command a comparable vocabulary of patterns, or chunks.

Now consider the scope of combinatorial explosion in games like rugby, football, tennis, ice hockey, American football, and the like. Even when scientists have invented simplified representations of these sports, they have quickly been overwhelmed by complexity. In robot football, for example, positions on the pitch are represented by 1,680 by 1,088 pixels. When you consider that a chessboard has eight by eight squares and that the pieces move in well-defined ways – unlike a football, which can fly anywhere at any time – you get some idea of the fiendish difficulty of designing a machine to compete without falling victim to information overload.

Now, here’s a description of Wayne Gretzky, arguably the greatest player in the history of ice hockey, taken from an article in the New York Times magazine in 1997:

Gretzky doesn’t look like a hockey player .. . Gretzky’s gift, his genius even, is for seeing ... To most fans, and sometimes even to the players on the ice, ice hockey frequently looks like chaos: sticks flailing, bodies falling, the puck ricocheting just out of reach.

But amid the mayhem, Gretzky can discern the game’s underlying pattern and flow, and anticipate what’s going to happen faster and in more detail than anyone else in the building. Several times during a game you’ll see him making what seem to be aimless circles on the other side of the rink from the traffic, and then, as if answering a signal, he’ll dart ahead to a spot where, an instant later, the puck turns up.

This is a perfect example of expert decision-making in practice: circumventing combinatorial explosion via advanced pattern recognition. It is precisely the same skill wielded by Kasparov, but on an ice hockey pitch rather than a chessboard. How was Gretzky able to do this? Let’s hear from the man himself: ‘I wasn’t naturally gifted in terms of size and speed; everything I did in hockey I worked for.’ And later: ‘The highest compliment that you can pay me is to say that I worked hard every day…That’s how I came to know where the puck was going before it even got there.’

All of which helps to explain a qualification that was made earlier in the chapter: you will remember that the ten-thousand-hour rule was said to apply to any complex task. What is meant by complexity? In effect, it describes those tasks characterized by combinatorial explosion; tasks where success is determined, first and foremost, by superiority in software (pattern recognition and sophisticated motor programmes) rather than hardware (simple speed or strength).

Most sports are characterized by combinatorial explosion: tennis, table tennis, football, ice hockey, and so on. Just try to imagine, for a moment, designing a robot capable of solving the real-time spatial, motor, and perceptual challenges necessary to defeat Roger Federer on a tennis court. The complexities are almost impossible to define, let alone solve. It is only in sports like running and lifting – simple activities testing a single dimension such as speed or strength – that the design possibilities become manageable.

Of course, not all expert decision-making is rapid and intuitive. In some situations, chess players are required to conduct deep searches of possible moves, and firefighters are required to think logically about the consequences of actions. So are top sportsmen and military commanders.

But even in the most abstract decisions, experience and knowledge play a central role. In an experiment carried out by David Rumelhart, a psychologist at Stanford University, five times as many participants were able to figure out the implications of a logical expression when it was stated in a real setting (‘Every purchase over thirty dollars must be approved by the manager’) than when stated in a less meaningful way (‘Every card with a vowel on the front must have an integer on the back’).

Earlier in this chapter we saw that the talent myth is disempowering because it causes individuals to give up if they fail to make rapid early progress. But we can now see that it is also damaging to institutions that insist on placing inexperienced individuals – albeit with strong reasoning skills – in positions of power.

Think, for example, of the damage done to the governance of Britain by the tradition of moving ministers – the most powerful men and women in the country – from department to department without giving them the opportunity to develop an adequate knowledge base in any of them. It is estimated that the average tenure of a ministerial post in recent years in Britain has been 1.7 years. John Reid, the long-serving member of Tony Blair’s government, was moved from department to department no less than seven times in seven years. This is no less absurd than rotating Tiger Woods from golf to football to ice hockey to baseball and expecting him to perform expertly in every arena.

What we decide about the relative importance of practice and knowledge on the one hand and talent on the other has major implications not just for ourselves and our families, but for corporations, sports, governments, and, indeed, the future of artificial intelligence.^*

On 3 May 1997, Kasparov and Deep Blue went head-to-head for a second time. The hype was no less intense and the stakes no less high. IBM put up over a million dollars in prize money, and the world’s media descended upon the venue – this time the thirty-fifth floor of the Equitable Center on Seventh Avenue in New York – in even greater numbers (IBM would later estimate that the company gained more than $500 million in free publicity).

But this time, Deep Blue was triumphant, defeating the world champion by two games to one, with three draws. It was a crushing blow for Kasparov, who stormed out of the venue. He would later allege that IBM had created playing conditions advantageous to Deep Blue and that they had refused to provide computer printouts which would have helped his preparation. He also made entirely unsubstantiated claims that IBM had cheated. He was not a good loser.

What had happened over the course of the preceding fifteen months? How had Deep Blue managed to convert defeat into a famous victory? Firstly, the machine had been provided with double the processing power (it was now able to compute more than 200 million moves per second). But its victory would have been impossible without another key innovation.

As the American Physical Society put it, ‘Deep Blue’s general knowledge of chess was significantly enhanced through the efforts of IBM consultant and international grandmaster Joel Benjamin, so that it could draw on vast resources of stored information, such as a database of opening games played by grandmasters over the last 100 years.’

Deep Blue’s programmers – like Gary Klein, Jim Immelt, and Wayne Gretzky – had realized that knowledge is power.