Can Game Theory Predict When Iran Will Get the Bomb?
Is Iran going to build a bomb?
Many people wonder, but Bruce Bueno de Mesquita claims to have the answer.
Bueno de Mesquita is one of the world’s most prominent applied game theorists. A professor at New York University and a senior fellow at the Hoover Institution at Stanford, he is well known academically for his work on “political survival,” or how leaders build coalitions to stay in power. But among national-security types and corporate decision makers, he is even better known for his prognostications. For 29 years, Bueno de Mesquita has been developing and honing a computer model that predicts the outcome of any situation in which parties can be described as trying to persuade or coerce one another. Since the early 1980s, C.I.A. officials have hired him to perform more than a thousand predictions; a study by the C.I.A., now declassified, found that Bueno de Mesquita’s predictions “hit the bull’s-eye” twice as often as its own analysts did.
Last year, Bueno de Mesquita decided to forecast whether Iran would build a nuclear bomb. With the help of his undergraduate class at N.Y.U., he researched the primary power brokers inside and outside the country — anyone with a stake in Iran’s nuclear future. Once he had the information he needed, he fed it into his computer model and had an answer in a few minutes.
In June, I visited Bueno de Mesquita at his San Francisco home to see the results. A tall man with a slab of gray hair, Bueno de Mesquita, who is 62, welcomed me with painstakingly prepared cups of espresso. Then he pulled out his beat-up I.B.M. laptop — so old that the lettering on the A, S, D and E keys was worn off — and showed me a spreadsheet that summarized Iran’s future.
The spreadsheet included almost 90 players. Some were people, like the Iranian president, Mahmoud Ahmadinejad, and Supreme Leader Ali Khamenei; others were groups, like the U.N. Security Council and Iran’s “religious radicals.” Next to each player, a number represented one variable in Bueno de Mesquita’s model: the extent to which a player wanted Iran to have the ability to make nuclear weapons. The scale went from 0 to 200, with 0 being “no nuclear capacity at all” and 200 representing a test of a nuclear missile.
At the beginning of the simulation, the positions were what you would expect. The United States and Israel and most of Europe wanted Iran to have virtually no nuclear capacity, so their preferred outcomes were close to zero. In contrast, the Iranian hard-liners were aggressive. “This is not only ‘Build a bomb,’ ” Bueno de Mesquita said, characterizing their position. “It’s probably: ‘We should test a bomb.’ ”
But as the computer model ran forward in time, through 2009 and into 2010, positions shifted. American and Israeli national-security players grudgingly accepted that they could tolerate Iran having some civilian nuclear-energy capacity. Ahmadinejad, Khamenei and the religious radicals wavered; then, as the model reached our present day, their power — another variable in Bueno de Mesquita’s model — sagged significantly.
Amid the thousands of rows on the spreadsheet, there’s one called Forecast. It consists of a single number that represents the most likely consensus of all the players. It begins at 160 — bomb-making territory — but by next year settles at 118, where it doesn’t move much. “That’s the outcome,” Bueno de Mesquita said confidently, tapping the screen.
What does 118 mean? It means that Iran won’t make a nuclear bomb. By early 2010, according to the forecast, Iran will be at the brink of developing one, but then it will stop and go no further. If this computer model is right, all the dire portents we’ve seen in recent months — the brutal crackdown on protesters, the dubious confessions, Khamenei’s accusations of American subterfuge — are masking a tectonic shift. The moderates are winning, even if we cannot see that yet.
Could this possibly be what will happen? Certainly Bueno de Mesquita has his critics, who argue that the proprietary software he uses can’t be trusted and may cast doubt on the larger enterprise of making predictions. But he has published a large number of startlingly precise predictions that turned out to be accurate, many of them in peer-reviewed academic journals. For example, five years before Ayatollah Khomeini died in 1989, Bueno de Mesquita predicted in the journal PS that Khomeini would be succeeded by Ali Khamenei (which he was), who himself would be succeeded by a then-less-well-known cleric named Akbar Hashemi Rafsanjani (which he may well be). Last year, he forecast when President Pervez Musharraf of Pakistan would be forced out of office and was accurate to within a month. In “The Predictioneer’s Game,” a book coming out next month that was written for a popular audience, Bueno de Mesquita offers dozens more stories of his forecasts. And as for Iran’s bomb?
In a year, he said with a wide grin, we’ll know if he’s right.
“I’m not an Iran expert,” Bueno de Mesquita told me cheerfully as we walked down his tree-lined street on our way to grab some Burmese food. Indeed, his career has been built on a peculiar concept: If you want to predict political events, wisdom and expertise, deep knowledge of a country’s culture and history, aren’t enough. To forecast the future, you need to be an expert not in statecraft but in the way individual people make decisions. You need “rational actor” game theory.
Bueno de Mesquita began studying political science in the 1960s. While working on his dissertation at the University of Michigan on parliamentary politics in India, a professor assigned him William H. Riker’s book “The Theory of Political Coalitions,” one of the first works to apply game theory to politics. Game theory is a branch of mathematics that studies the way people will behave in strategic situations — that is to say, when they’re making decisions based on how they think other people will make decisions. Generally, game theory assumes that people are always rational and selfish; they’re always angling to get what’s best for them, which means their behavior can often be predicted. One famous application of rational-choice theory that particularly intrigued Bueno de Mesquita was Duncan Black’s analysis of “committee voting,” which argues that if two rival candidates are trying to get elected on a single issue — say, taxes — they’ll inevitably shift their positions toward the median voter.
Bueno de Mesquita was enthralled by the idea of rendering the messy business of politics and history into precise, logical equations. He began his signature academic work on “the selectorate,” or the group of actors who run a country. In Bueno de Mesquita’s worldview, there is no such thing as a “national interest” (or “state”). There are just leaders trying desperately to stay in power by building coalitions within their selectorate — buying off cronies in the case of a dictatorship, for example, or producing enough good works to keep hoi polloi happy in a democracy.
When Bueno de Mesquita spotted a logical error in one of Riker’s books, he wrote the author a letter; Riker offered Bueno de Mesquita a job in 1972 at the University of Rochester, where a new generation of political scientists was starting to apply formal mathematical models to political analysis.
That’s where Bueno de Mesquita began programming his computer model. It is based loosely on Black’s voter theory, and it works like this: To predict how leaders will behave in a conflict, Bueno de Mesquita starts with a specific prediction he wants to make, then interviews four or five experts who know the situation well. He identifies the stakeholders who will exert pressure on the outcome (typically 20 or 30 players) and gets the experts to assign values to the stakeholders in four categories: What outcome do the players want? How hard will they work to get it? How much clout can they exert on others? How firm is their resolve? Each value is expressed as a number on its own arbitrary scale, like 0 to 200. (Sometimes Bueno de Mesquita skips the experts, simply reads newspaper and journal articles and generates his own list of players and numbers.) For example, in the case of Iran’s bomb, Bueno de Mesquita set Ahmadinejad’s preferred outcome at 180 and, on a scale of 0 to 100, his desire to get it at 90, his power at 5 and his resolve at 90.
Then the math begins, some of which is surprisingly simple. If you merely sort the players according to how badly they want a bomb and how much support they have among others, you will end up with a reasonably good prediction. But the other variables enable the computer model to perform much more complicated assessments. In essence, it looks for possible groupings of players who would be willing to shift their positions toward one another if they thought that doing so would be to their advantage. The model begins by working out the average position of all the players — the “middle ground” that exerts a gravitational force on the whole negotiation. Then it compares each player with every other player, estimating whether one will be able to persuade or coerce the others to move toward its position, based on the power, resolve and positioning of everyone else. (Power isn’t everything. If the most powerful player is on the fringe of an issue, and a cluster of less-powerful players are closer to the middle, they might exert greater influence.) After estimating how much or how little each player might budge, the software recalculates the middle ground, which shifts as the players move. A “round” is over; the software repeats the process, round after round. The game ends when players no longer move very much from round to round — this indicates they have compromised as much as they ever will. At that point, assuming no player with veto power had refused to compromise, the final average middle-ground position of all the players is the result — the official prediction of how the issue will resolve itself. (Bueno de Mesquita does not express his forecasts in probabilistic terms; he says an event will transpire or it won’t.)
The computer model, in short, predicts coalitions. And computers are much better at doing this than humans, because with more than a few players the number of possible coalitions quickly multiplies. With 40 players, the typical size of one of Bueno de Mesquita’s forecasts, there are 1,560 possible pairs to consider just for starters. This is why, he says, his model often produces surprising results. It’s not that it is smarter than humans. But it methodically works through not only the obvious coalitions we know about and expect but also the invisible ones that we don’t.
For Bueno de Mesquita, the first prominent use of the model came in 1979, when the State Department was canvassing academics with expertise on India, including Bueno de Mesquita, to see how some parliamentary maneuverings would unfold. Bueno de Mesquita decided to use his first version of the software (which was, as he puts it, “barely working”) and his own knowledge of India to determine the power players and each of their numbers. Then the university’s mainframe computer worked on the data all night.
In the morning, Bueno de Mesquita said, he was astonished: the predicted victor was a seemingly minor figure, someone discounted by the experts. Bueno de Mesquita shared their opinion, he told me, but he accepted the computer’s verdict anyway. “So I called the person back at the State Department, and told him what I had concluded,” Bueno de Mesquita went on. “And there was a long, quiet period and some laughing. He said: ‘How did you arrive at that? Nobody’s saying that.’ So I told him I had a little computer model. He just guffawed. He said, ‘I wouldn’t repeat that if I were you.’ ”
Three months later, according to Bueno de Mesquita, his prediction turned out to be right.
The son of Jewish immigrants who arrived from Brussels during World War II, Bueno de Mesquita grew up in Manhattan, where his father ran a small publishing company and his mother managed a women’s clothing shop. He went to Queens College when he was 16 — “way too young,” he says — and read history and literature voraciously. (Bueno de Mesquita spent years researching and writing a short novel that defends Ebenezer Scrooge as a kindhearted man.) “He is one the most remarkably intelligent human beings I’ve met in my life, and Bruce does not hesitate to tell you that,” Kevin Gaynor, an environmental lawyer who has twice hired Bueno de Mesquita to advise his corporate clients on “extremely sensitive” government negotiations, told me half-jokingly. “He’s not self-effacing. But he’s not self-effacing in a charming way.” Bueno de Mesquita’s voluminous academic work — he has published 16 books and more than 100 papers — is credited with helping to move game theory and mathematical modeling into the mainstream of political science; according to one count, by 1999 fully 40 percent of papers in the American Political Science Review used modeling. (The figure was so high it prompted deep consternation among non-game-theory political scientists.) While few perform the consulting work he does, other game theorists have produced models very similar to Bueno de Mesquita’s, and he actively promotes his technique, including training N.Y.U. undergraduates to do similar predictions.) He spends half the year at N.Y.U., where he recently finished a four-year stint as the chairman of the political-science department, and half the year at the Hoover Institution at Stanford. Under the terms of his academic contracts, he is permitted to spend one day per week during the academic year doing outside consulting.
It is this consulting, more than his academic work, that has made Bueno de Mesquita both well off and controversial. He began offering predictions to the private sector in 1982, when A.F.K. Organski, a former professor of his, suggested they go into business using Bueno de Mesquita’s model. Business negotiations, they reasoned, were like international relations in that they involved players trying to wheedle and coerce one another. Soon Bueno de Mesquita and Organski (who died in 1998) acquired clients ranging from Arthur Andersen to Union Carbide, which tapped them for advice on placating the Indian government after the Bhopal chemical spill. Today Bueno de Mesquita’s firm essentially consists of himself and Harry Roundell, a former banker at J. P. Morgan who met Bueno de Mesquita when Roundell hired him in 1995 to help the bank figure out how to push for new, favorable regulations in the U.S. They charge $50,000 and up to do a prediction and offer negotiating tips, and they take on 18 to 20 of these assignments a year. Beyond saying it was “a reasonable amount of money,” Bueno de Mesquita would not describe his income from the company.
To produce a corporate prediction, Roundell and Bueno de Mesquita determine the numerical values of the players in a negotiation by interviewing a firm’s executives. This can take anywhere from a few hours on the phone to two days of face-to-face conversations. Both men conduct the interviews, and Bueno de Mesquita enters the information into a spreadsheet.
The real value of Bueno de Mesquita’s work, several clients told me, is not only in his predicting how a corporate event might unfold. It is also in figuring out how to influence that event. Because Bueno de Mesquita’s model forecasts the future by calculating the impact every player has on every other player, round by round, Bueno de Mesquita can go back and see when some players suddenly become more flexible midway through a negotiation. He can thus perform “what if” experiments: What if that person could be persuaded to change his mind? He’ll enter new values into the model, manually changing that player’s position, then run it again to see if this change recasts the future to his client’s advantage. If it does, Bueno de Mesquita now has a piece of advice: focus on that player in real life, and try to influence him. If there are dozens of players and dozens of rounds, the number of possible “what if” scenarios becomes enormous: it can take Bueno de Mesquita days of peering at his spreadsheets to identify useful pressure points.
One of Bueno de Mesquita’s most prominent public consultations occurred in 1999, when Richard Lapthorne, then the vice chairman of British Aerospace, asked him to help engineer a $10 billion acquisition. The British government wanted British Aerospace to form a pan-European firm by merging with the German firm DASA and the French giant Aérospatiale; British Aerospace, however, was more interested in trying to buy the British electronics giant Marconi Electronic Systems. To persuade the British government to approve the Marconi deal, Lapthorne asked Bueno de Mesquita to predict the viability of mergers between the German and French firms. The model forecast that the three firms would never be able to agree on terms, and that the Marconi deal was the better option; when Bueno de Mesquita showed his analysis to the government heads, they agreed to permit the Marconi acquisition. “There’s nothing shimmy shammy or flip-flop about it,” Lapthorne says of the logical nature of Bueno de Mesquita’s prediction. “It’s very clear where the information came from. It has intellectual rigor.” Lapthorne is now chairman of the U.K. telecommunications company Cable and Wireless; he has used Bueno de Mesquita for seven predictions since, though he would not disclose the subjects.
It is difficult to verify how accurately Bueno de Mesquita’s model performs in corporate settings because most firms are loath to discuss his work for them. For most of the cases we discussed, Bueno de Mesquita would disclose details of the negotiation but wouldn’t name the firms in question. In other cases, clients would talk to me and praise Bueno de Mesquita’s work for them, but they would not disclose verifiable details of specific negotiations. There were a few exceptions: Robert F. Kelley, a retired former partner of Arthur Andersen, described using Bueno de Mesquita for “60 or 70” cases, ranging from internal firing decisions to figuring out how to persuade the U.S. to support China’s entry into the World Trade Organization. (Bueno de Mesquita also offered to use his software to predict which of Arthur Andersen’s clients — including, at the time, Enron — were likely to engage in financial fraud. But the firm’s lawyers, Bueno de Mesquita says, didn’t want to use the tool for fear it would put them in awkward legal positions. “Had I been able to convince the firm” to use the model, Kelley says, “I think that Andersen would be alive today.”)
Bueno de Mesquita’s most regular client by far has been the C.I.A. He says he has performed more than 1,200 predictions for the agency, tackling questions like “How fully will France participate in the Strategic Defense Initiative?” and “What policy will Beijing adopt toward Taiwan’s role in the Asian Development Bank?” In 1987, Stanley Feder, a research political scientist for the C.I.A., published a report analyzing forecasts that Bueno de Mesquita’s firm did of political events in 27 countries; he found that the success rate of its predictions was the same as that of the C.I.A.’s own analysts, only more precise. (He “got the bull’s-eye twice as often,” Feder wrote in his report, which was declassified in 1993. No other reports have been declassified since.) Feder noted, for example, that Bueno de Mesquita’s model predicted in a forecast done of Italy’s budget one year a specific figure that turned out to be off by only 1 percent; the C.I.A. method would predict just a deficit.
Those who have watched Bueno de Mesquita in action call him an extremely astute observer of people. He needs to be: when conducting his fact-gathering interviews, he must detect when the experts know what they’re talking about and when they don’t. The computer’s advantage over humans is its ability to spy unseen coalitions, but this works only when the relative positions of each player are described accurately in the first place. “Garbage in, garbage out,” Bueno de Mesquita notes. Bueno de Mesquita begins each interview by sitting quietly — “in a slightly closed-up manner,” as Lapthorne told me — but as soon as an interviewee expresses doubt or contradicts himself, Bueno de Mesquita instantly asks for clarification.
“His ability to pick up on body language, to pick up on vocal intonation, to remember what people said and challenge them in nonthreatening ways — he’s a master at it,” says Rose McDermott, a political-science professor at Brown who has watched Bueno de Mesquita conduct interviews. She says she thinks his emotional intelligence, along with his ability to listen, is his true gift, not his mathematical smarts. “The thing is, he doesn’t think that’s his gift,” McDermott says. “He thinks it’s the model. I think the model is, I’m sure, brilliant. But lots of other people are good at math. His gift is in interviewing. I’ve said that flat out to him, and he’s said, ‘Well, anyone can do interviews.’ But they can’t.”
You might expect Bueno de Mesquita to be the toast of both Washington and Wall Street, constantly in demand for prognostications. Yet he and Roundell have found that it is not so easy to attract clients. This is partly because most of their clients — especially the C.I.A. — swear them to secrecy. (And perhaps also because, as Roundell says, “Bruce and I are . . . terrible salespeople.”) But they have also faced a barrier that’s almost existential, a skepticism that computer models can truly predict the outcome of negotiations. The C.I.A., for example, built its own replica of Bueno de Mesquita’s original forecast model, but as Feder noted in his report, “the vast majority of analysts” didn’t use it because it seemed too rigid. They thought of analysis as reading and pondering until they had an aha! moment — not feeding data points into a computer model and waiting to see what comes up.
When we spoke, Bueno de Mesquita often seemed irritated by resistance to his work. For all his gifts of intuition, he has a Spocklike disdain for gut instinct. When he occasionally hires colleagues to help him with a complex bit of corporate work, he sternly warns them that they must refrain from expressing any personal opinions and describe only what they see in the spreadsheets. Bueno de Mesquita habitually and hissingly disparages traditional political analysis. He is savvy enough to know that nobody likes a scold, yet he can’t help himself; he sheepishly admits to becoming “confrontational” when people think mathematical reasoning can’t be used. At the C.I.A., Feder told me, “there were some people who found him arrogant, which was maybe a reasonable reaction.”
Donald Green, a political scientist at Yale, questions whether Bueno de Mesquita is serving the discipline well. “When I see clips of Bruce at the TED conference,” he says, referring to the annual conference promoting ideas in technology, entertainment and design, “I watch the video and I think, Wow, this is so far from the typical way in which political scientists of any stripe behave.” Some political scientists are openly dubious about the accuracy of Bueno de Mesquita’s model. Stephen Walt, a Harvard professor of international affairs, says that Bueno de Mesquita’s nonprediction work — like his theory of the “political survival” of heads of state — make him a “respected scholar, deservedly so.” It’s the predictions that Walt doesn’t trust, because Bueno de Mesquita does not publish the actual computer code of his model. (Bueno de Mesquita cannot do so because his former firm owns the actual code, but he counters that he has outlined the math behind his model in enough academic papers and books for anyone to replicate something close to his work.) While Bueno de Mesquita has published many predictions in academic journals, the vast majority of his forecasts have been done in secret for corporate or government clients, where no independent academics can verify them. “We have no idea if he’s right 9 times out of 10, or 9 times out of a hundred, or 9 times out of a thousand,” Walt says. Walt also isn’t impressed by Stanley Feder’s C.I.A. study showing Bueno de Mesquita’s 90 percent hit rate. “It’s one midlevel C.I.A. bureaucrat saying, ‘This was a useful tool,’ ” Walt says. “It’s not like he’s got Brent Scowcroft saying, ‘Back in the Bush administration, we didn’t make a decision without consulting Bueno de Mesquita.’ ” Other academics point out that rational-actor theory has come under increasing criticism in recent years, as more evidence accumulates that people make many decisions irrationally.
And it’s true that there have been cases when Bueno de Mesquita’s model has gone awry. In his 1996 book, “Red Flag Over Hong Kong,” he predicted that the press in Hong Kong “will become largely a tool of the state” — a highly debatable claim today. (In 2006, Reporters Without Borders noted concerns about self-censorship but said that “journalists remain free in Hong Kong.”) In early 1993, a corporate client asked him to forecast whether the Clinton administration’s health care plan would pass, and he said it would.
What’s more, with corporate clients in particular, there’s always the potential problem of reflexivity, of the prediction itself influencing events and making it hard to evaluate the prediction’s value. Suppose a firm is told a merger will fail, for example, and abandons its merger efforts. Was the prediction accurate or a self-fulfilling prophecy?
Spending time with Bueno de Mesquita is alternately alarming and reassuring, because he has such confidence in his own predictions about our global fate. Like many, he believes the future of Pakistan is “incredibly distressing” right now, but he has reached this conclusion in his own way: when he and his students modeled its future last year, the power of Al Qaeda and the Taliban in that region grew quickly throughout 2009, far outstripping that of the government and military. Global warming is another area where politics are doomed to fail. World governments are set to meet this December in Copenhagen to commit to firm CO2-reduction levels — but when Bueno de Mesquita modeled the future of these targets, most countries renege on them. No democratic government will seriously limit CO2 if it will hurt its citizens economically.
“When people are asked to make personal sacrifices for the greater good in the longer term, they seem to find 1,001 reasons why their particular behavior is so virtuous that this one particular deviation is really O.K.,” Bueno de Mesquita told me recently as we talked in his home office. “ ‘I have to drive an S.U.V. because I want to protect my little children from a car accident!’ ”
Yet Bueno de Mesquita remains cheerful, almost unnervingly so. Years of peering at his model have shown him that conflicts almost always have hidden solutions — places where the computer illuminates the sort of leverage that could be employed to create a sudden, useful countercoalition. For example, with Pakistan, his model showed that if the U.S. merely doubled its annual aid from $700 million to $1.5 billion, America’s influence in the country would significantly jump, while the militants’ would drop drastically. Why? Because with that sort of financial flow, corrupt rural officials would suddenly profit more from helping the U.S. than from helping the Taliban.
In the short term, though, Bueno de Mesquita’s reputation will be colored by Iran. The last time we met, it was two weeks after the Iranian election, and the opposition protests had been quashed. The hard-liners, I noted, seemed to be winning — did this mean that the prediction was wrong? “The street movement is running out of steam,” Bueno de Mesquita agreed. “Shooting people does act as an effective deterrent.” But he still maintained that his model was likely to prevail, and that domestic coalitions we might not detect from abroad are gathering to overwhelm the religious conservatives.
He spent that morning looking over his Iranian data, and he generated a new chart predicting how the dissidents’ power would grow over the next few months. In terms of power, one category — students — would surpass Ahmadinejad during the summer, and by September or October their clout would rival that of Khamenei, the supreme leader. “And that’s huge!” Bueno de Mesquita said excitedly. “If that’s right, it’s huge!” He said he believed that Iran’s domestic politics would remain quiet over the summer, then he thought they’d “really perk up again” by the fall.
Bueno de Mesquita also approved of Obama’s hands-off approach. Bueno de Mesquita ran an experimental version of his Iranian model without the U.S. in it as a player at all, and the coalitions that oppose Ahmadinejad and the bomb emerge a few months more quickly. In other words, American meddling is indeed counterproductive; the less America tries to influence Iran, the more quickly Iran will abandon nuclear weapons, if the logic of the computer is correct.
It’s a fascinating analysis, but, I wonder, has he given it to anyone in the State Department? He laughed. “I’m working on access.”
No comments:
Post a Comment