Read The Invisible Gorilla: And Other Ways Our Intuitions Deceive Us Online
Authors: Christopher Chabris,Daniel Simons
I
N JUNE 2000,
U.S. president Bill Clinton and British prime minister Tony Blair jointly announced the completion of the initial phase of the Human Genome Project, the celebrated international effort to decode the DNA sequence of all twenty-three human chromosomes. The project ultimately spent about $2.5 billion over ten years to produce a “first draft” of the sequence, and over $1 billion more to fill in the gaps and polish the results.
1
One of the most intriguing questions that biologists hoped the project would answer seemed to be a simple one: How many genes are there in the human genome?
2
Before the sequence was completed, prevailing opinion held that the complexity of human biology and behavior must be the product of a large number of genes, probably between 80,000 and 100,000. In September 1999, a high-flying biotech company called Incyte Genomics proclaimed that there were 140,000 genes in the human genome. In May 2000, top genelicists from around the world converged at the
“Genome Sequencing and Biology” conference at the Cold Spring Harbor Laboratory in New York, and a lively debate about the true count ensued. Yet no consensus estimate emerged; some agreed with counts as high as those claimed by Incyte, and others argued that the number might be lower than 50,000.
With so many different opinions on offer, Ewan Birney, a geneticist at the European Bioinformatics Institute, started a betting pool for his fellow researchers to predict the final count. Each participant put in a dollar, and the winner would receive the total amount collected, plus a signed, leather-bound copy of Nobel Prize–winner James Watson’s memoir,
The Double Helix
. Incyte’s Sam LaBrie came in with the highest initial estimate: 153,478 genes. The average of the first 338 predictions entered was 66,050. Birney raised the entry fee to five dollars in 2001, and then to twenty dollars in 2002—it wouldn’t really be fair to let later bettors in for the same amount as earlier ones, since the late bettors could use the earlier estimates as well as their own research findings to guide their guesses. The 115 later entries averaged 44,375, and the pot grew to $1,200. Over the full two-year betting period, the lowest entry was 25,747, submitted by Lee Rowen from the Institute for Systems Biology in Seattle.
The terms of the competition, set in 2000, required Birney to declare a winner in 2003. However, to Birney’s surprise, there was still no consensus “final count” at that point. Based on evidence available at the time, Birney estimated the total count to be about 24,500. He decided to award portions of the pool to the three entrants who bet on the lowest numbers, with Rowen getting the largest prize. The final number is still in dispute, but the most accepted value has dropped to 20,500, squarely in the range between the roundworm called
C. elegans
(19,500) and the mustard plant called
Arabidopsis
(27,000).
The bettors all were leaders in the field of genetics, and they were sure that the number was higher than it actually was; the range of their 453 predictions, from the highest to the lowest estimate, did not even include the correct count. Francis Collins of the National Institutes of Health and Eric Lander of the Massachusetts Institute of Technology, leaders of the Genome Project, were each off by more than 100 percent, no better
than the average guess. The collective also had a pretty poor idea of how quickly the gene-count question would be resolved (predicted: 2003, actual: 2007 or later). Collins reacted stoically: “Oh well, live and learn.”
This is far from the only example of scientists overestimating their knowledge in their own fields of expertise. In 1957, two of the pioneers of computer science and artificial intelligence, Herbert Simon and Allen Newell, publicly predicted that within ten years a computer would be able to defeat the world chess champion in a match.
3
By 1968 no one had come close to creating a machine capable of that feat. David Levy, a Scottish computer programmer and chess player who would later achieve the title of international master (one level below grandmaster), met with four other computer scientists and bet them £500 of his own money—an amount equal to about one-half of his annual income at the time—that no computer would be able to beat him in a match within the
next
ten years. In 1978, with the pot sweetened to £1250 by further wagers, Levy in fact defeated the best computer program by a score of 3½—1½. Together with
Omni
magazine, he then offered a new prize of $5,000 to anyone whose computer could beat him, with no time limit on the bet. Finally, in 1989, Levy lost to Deep Thought, a predecessor of IBM’s Deep Blue computer. Only in 1997 did Deep Blue, with its multiple processors and custom-designed chess chips, defeat world champion Garry Kasparov 3½—2½ and fulfill the Simon-Newell prophecy—thirty years behind schedule.
4
In 1980, ecologist Paul Ehrlich, a professor at Stanford University, and his colleagues John Harte and John Holdren of the University of California at Berkeley, were convinced that global overpopulation would lead to drastic increases in the prices of food and other commodities that were in finite supply. Indeed, Ehrlich had been convinced that this threat was dire for some time, having written in 1968, “In the 1970s the world will undergo famines—hundreds of millions of people are going to starve to death.”
5
He and Holdren predicted the imminent “exhaustion of mineral resources.”
6
Julian Simon, an economist at the University of Maryland, had the opposite view. He published an article in the journal
Science
titled “Resources, Population, Environment: An Oversupply of False Bad News.”
7
Simon, whose previous claim to fame was inventing the system under which airlines reward passengers for giving up their seats on overbooked flights, proceeded to challenge the doomsayers to put their money where their mouths were: Pick five commodities and bet that their prices would increase over the next ten years, as one would expect if demand were always increasing and supply were constant or decreasing. Ehrlich was outraged by the apostasy displayed by Simon (whom he referred to as the leader of a “space-age cargo cult”), so he got Harte and Holdren to join him in accepting the wager proposed by the economist. They selected five metals—chrome, copper, nickel, tin, and tungsten—and calculated the amount of each that could be purchased for $200 in 1980. If these metals’ prices were higher ten years later, Simon would pay Ehrlich, Harte, and Holdren the difference; if the prices were lower, they would pay him. By 1990, all five commodities had gone down in price. In fact, they had collectively dropped more than 50 percent. Simon received an envelope containing a check in the amount of his winnings. There was no cover note.
8
You might object that we’ve cherry-picked examples in which experts made their most horribly errant predictions. We agree that these examples are atypical, and we’re not arguing that experts know nothing and are always wrong. Especially in scientific domains, they know a lot more and are right much more often than the average person. But these stories show that even scientific experts can dramatically overestimate what they know. Every single geneticist guessed high on the gene count, and some were off by a factor of five; the computer scientists were off by a factor of four; and the ecological doomsayers were wrong about every one of the metals they selected. If expert judgments can be so misguided, the rest of us must also be capable of overestimating what we know. Whenever people think they know more than they do, they are under the influence of our next everyday illusion: the
illusion of knowledge
.
Spend a moment now and try to form an image in your mind of a bicycle. Even better, if you have a piece of paper, draw a sketch of a bicycle.
Don’t worry about making a great piece of art—just focus on getting all the major parts in the right place. Sketch out the frame, the handlebars, the wheels, the pedals, and so on. For simplicity, just make it a singlespeed bicycle. Got it? If you had to rate your understanding of how a bicycle works on a 1 to 7 scale, where 1 means “no understanding” and 7 means “complete understanding,” what score would you give yourself?
If you are like most of the people who participated in a clever study by British psychologist Rebecca Lawson, you thought you had a pretty good understanding of bicycles; her subjects rated the level of their knowledge at 4.5 out of 7, on average.
9
Now either look at your drawing or refresh your mental image and then answer the following questions: Does your bicycle have a chain? If so, does the chain run between the two wheels? Does the frame of your bicycle connect the front and back wheels? Are the pedals connected to the inside of the chain? If you drew a chain connecting the two wheels of your bicycle, think about how the bicycle would turn—the chain would have to stretch whenever the front wheel rotated, but chains aren’t stretchy. Similarly, if a rigid frame connected both wheels, the bicycle could only go straight. Some people draw pedals outside the loop of the chain, making it impossible to turn the chain by pedaling. Errors like these were common in Lawson’s study, and they are not trivial details of the functioning of a bicycle—the pedals turn the chain, which causes the back wheel to rotate, and the front wheel must be free to turn or the bicycle cannot change direction. People are much better at making sense of a bicycle’s workings when the thing is sitting right in front of them than they are at explaining (or drawing) a bicycle purely from memory.
This example illustrates a critical aspect of the illusion of knowledge. Because of our extensive experience and familiarity with ordinary machines and tools, we often think we have a deep understanding of how they work. Think about each of the following objects and then judge your knowledge of it on the same 1 to 7 scale: a car speedometer, a zipper, a piano key, a toilet, a cylinder lock, a helicopter, and a sewing machine. Now try one more task: Pick the object that you gave the highest rating, the one you feel you best understand, and try to explain how it works. Give
the kind of explanation you would give to a persistently inquisitive child—try to generate a detailed step-by-step description of how it works, and explain why it works. That is, try to come up with the causal connections between each step (in the case of the bicycle, you would have to say something about
why
pedaling makes the wheels turn, not just
that
pedaling makes the wheels turn). If you aren’t sure how two steps are causally connected, you’ve uncovered a gap in your knowledge.
This test is similar to a series of ingenious experiments that Leon Rozenblit conducted as part of his doctoral research at Yale University with Professor Frank Keil (who, incidentally, was also Dan’s graduate school adviser).
10
For his first study, Rozenblit approached students in the hallways of the psychology building and asked them if they knew why the sky is blue or how a cylinder lock works. If they answered yes, he then played what he calls the “why boy” game, which he describes as follows: “I ask you a question and you give me an answer, and I say ‘why is that?’ Channeling the spirit of a curious five-year-old, I then just keep following each explanation with another ‘why is that?’ until the other person gets really annoyed.”
11
The unexpected result of this informal experiment was that people gave up really quickly—they answered no more than one or two “why” questions before they reached a gap in their understanding. Even more striking were their reactions when they discovered that they really had no understanding. “It was clearly counterintuitive to them. People were surprised and chagrined and a little embarrassed.” After all, they had just claimed to know the answer.
Rozenblit pursued this illusion of knowledge in more than a dozen experiments over the next few years, testing people from all walks of life (from undergraduates at Yale to members of the New Haven community), and the results were remarkably consistent. No matter whom you talk to, you will eventually reach a point where they can no longer answer the why question. For most of us, our depth of understanding is sufficiently shallow that we may exhaust our knowledge after just the first question. We know that there is an answer, and we feel that we know it, but until asked to produce it we seem blissfully unaware of the shortcomings in our own knowledge.
Before you tried this little test, you might have thought intuitively that you understood how a toilet works, but all you really understand is how to work a toilet—and maybe how to unclog one. You likely understand how the various visible parts interact and move together. And, if you were looking inside a toilet and playing with the mechanism a bit, you might be able to figure out how it works. But when you aren’t looking at a toilet, your impression of understanding is illusory: You mistake your knowledge of
what
happens for an understanding of
why
it happens, and you mistake your feeling of familiarity for genuine knowledge.
We sometimes encounter students who come to our offices and ask how they could have worked so hard but still failed our tests. They usually tell us that they read and reread the textbook and their class notes, and that they thought they understood everything well by the time of the exam. And they probably did internalize some bits and pieces of the material, but the illusion of knowledge led them to confuse the familiarity they had gained from repeated exposure to the concepts in the course with an actual understanding of them. As a rule, reading text over and over again yields diminishing returns in actual knowledge, but it increases familiarity and fosters a false sense of understanding. Only by testing ourselves can we actually determine whether or not we really understand. That is one reason why teachers give tests, and why the best tests probe knowledge at a deep level. Asking whether a lock has cylinders tests whether people can memorize the parts of a lock. Asking how to pick a lock tests whether people understand
why
locks have cylinders and what functional role they play in the operation of the lock.