10 mai 2006

Livre : The wisdom of crowds - J.Surowiecki


In 1841, the Scottish journalist Charles Mackay published an essay entitled “Extraordinary popular delusions and the madness of crowds”. It is in fact a popular belief that people usually get dumb when they gather into crowds. James Surowiecki’s book discusses several situations where this is absolutely wrong, and collective perception, behaviour, or problem-solving can in fact outperform any given individual. He further stresses that, for such a “wisdom of crowds” to happen, certain conditions must be met.

Although it could be more scientific in its approach, and it mixes various subjects which more or less fit its title, this book is entertaining to read. There are plenty of opportunities one can think of where collective wisdom could apply, and it helps us identify those situations, and predict how they might eventually turn out.

As a first example, Surowiecki quotes an anecdote by the British scientist Francis Galton in 1906. Galton – a firm believer in the superiority of individual experts compared to the stupid crowd – goes to a livestock market where a contest is taking place. Betters must try to guess the weight of an ox after it has been slaughtered and “dressed”. Galton takes note of several hundreds of bets, and finds out that their average is 1197oz. when the actual weight of the ox is 1198oz!

Humans are imperfect decision-makers, and often display limited rationality when trying to solve a given problem. Nobel Prize economist Herbert Simon theorized this in what he coined “bounded rationality”. Instead of being perfectly rational being, we seem instead to make fairly good decisions given the incomplete information which is available to us and the limited amount of time we can spend in the process. Surowiecki gives a variety of examples of problem-solving situations which he classifies in three categories:

- cognition problems, such as guessing the weight of the ox;

- coordination problems, such as picking the best time to go to work in order to avoid heavy traffic, knowing that every other driver faces the same question;

- cooperation problems, such as getting independent, self-interested teams to work together to improve the global results of the company, although this might go against their personal interest.

He then discusses what he considers to be the recipe for success in crowd intelligence:

- diversity, to have people of various backgrounds with original ideas;

- independence, to allow these diverse opinions to occur rather than get people aligned all in the same direction;

- decentralization, to let individual judgements add up rather than have a top-level executive pick the ideas that he likes best.

Part I – Principles

  1. Cognition problems

Galton’s anecdote takes another form in the well-known “jelly-beans-in-the-jar experiment” which has been reproduced over and over. Fill a jar with beans and ask students to guess how many beans are in the jar. The average of the answers is usually quite close to the actual number, and outperforms almost any student in the class. A real-life example of such a cognition problem is seen in sports and games where gambling takes place. The goal can be to estimate how good a given hors is, and what the probability is that it will win the race. If the gambling market says this horse is three to one, this means it should win 25% of the time. This kind of guess turns out to be quite accurate in games where the gambling market is sufficiently large and liquid so that it cannot be easily manipulated and its equilibrium really reflects the average perception of all the players.

In a way, Google can be seen as such a kind of system, since it ranks web pages according to the number of other pages that point towards them, taking into account the individual weight of each page. Google therefore reflects the collective “opinion” of the web on a given page, and is able through this process to find those pages that are most likely to meet your interests.

Similarly, the University of Iowa’s College of Business founded in 1988 the IEM – Iowa Electronic Markets project. This electronic system is designed to predict the outcome of upcoming elections. People can buy and sell future contracts based on how they think a candidate will do in an election. For instance if the price for Bush futures is 48 cents, it means that Bush is expected by the market to score 48% at a given election. Between 1988 and 2000, a total number of 596 polls were compared with the IEM’s forecast price on the day of the poll. As it turns out, IEM was more accurate than the polls three times out of four! What the IEM also suggests is that people focus better on a decision when there is a reward attached to it. Usually, a financial incentive works very well, but status and reputation can provide other forms of incentives to make good decisions. Although such markets could be effectively used to make better political and corporate decisions, there has been much reluctance to do so until now.

  1. Diversity

The engineer Ransom E. Olds started building steam-powered cars in the 1880’s. By the 1900’s there were literally hundreds of car manufacturers in the United States, each with a different design, technology, such as steam, gas, and electric power. By 1910, it was clear that gas was the best energy source, only a few companies had survived, and Ford was among the winners thanks to its invention of the assembly line. In this kind of Darwinian process, it is key to have diversity among the entrepreneurs who come up with the ideas. In this regard, a government usually can’t – and therefore shouldn’t – “pick the winners”. Bringing in a few people who actually know less, are not experts, but have different skills, usually improves the group’s performance.

On the contrary, homogeneity increases the probability that group members will all think of the same bad solution. This is made worse by the well-known phenomenon by which group members tend to follow the opinion of the majority, especially when all have the same kind of background and education.

  1. Independence

If you stand at the corner of a street and stare at the empty sky for a minute, chances are that a few passers-by will look up to see what is there. If a group of five persons does the same staring exercise, the percentage of passers-by looking up will increase considerably. This is known as “social proof”, which is the tendency to assume that if lots of people are doing or believing something, there must be a good reason why. Keynes expressed this in his General theory of employment, interest and money by stating “Worldly wisdom teaches that it is better for reputation to fail conventionally than to succeed unconventionally”.

This kind of behaviour does make sense, since it can often help us solve problems effortlessly. After a certain point it becomes rational to stop paying attention to our own knowledge and to start looking at the actions of others. For instance this allows people to specialize in certain skills, and the benefits from their investment can be spread widely when other imitate them. But this can lead to an information cascade where the information deriving from collective behaviour becomes overwhelming and obliterates any individual judgement.

One way to avoid this kind of pitfall, is to have a sufficient people of overconfident people who will tend to ignore public information and will go on their gut. Fortunately, there is some evidence that information cascades are less likely to fail when important decisions must be taken, because the early adopters or trend-setters then spend enough time and energy making the right decision. However, the best way to avoid them is by making sure that individual decisions are taken all at once rather than sequentially, and getting people to pay less attention to what everyone else is saying.

  1. Decentralization

In 1947, the Congress passed the National Security Act by which the Central Intelligence Agency was created. The division of intelligence responsibilities during WWII was considered to be the cause of the failure to anticipate the attack on Pearl Harbour. Therefore, the goal was to implement a unity of intelligence and unity of command to avoid such a situation in the future. However, the 1993 World Trade Center bombing, several other terrorist attacks and finally September 11, 2001 came as a cruel reminder of the flaws of such an organization. Intelligence agencies had allowed vital information to pass by unnoticed and failed to uncover and prevent the attacks, once again…

The economist Friedrich Hayek called “tacit knowledge” the sum of what a worker knows which cannot easily by summarized and conveyed to others because it is intimately linked with his local environment and everyday situations. Figuring out how to take advantage of individual’s tacit knowledge is a central challenge for any group or organization. The strength of decentralization is therefore to encourage independence and specialization, but it can produce intelligent results only if there is a means of aggregating the information of everyone in the group.

Centralization and top-down organizations are not the answer because they tend to give people the wrong incentive, and make them worry more about their selfish interest than about the corporate goals. In the case of a free market, the aggregating mechanism is the price, and this could be implemented in corporate situations. In principle, an internal market would place the premium not on making one’s boss happy, but on making the most accurate forecast. In 2003, DARPA proposed to fund the development of a market aiming at predicting the probability of events in the Middle East in order to guide the decisions of the Department of Justice. This project received heavy criticism from the Senate and was rejected as “harebrained” and “morally wrong” because it would have allowed market members to make bets on things like terrorist attacks.

  1. Coordination problems

What defines a coordination problem, such as driving to work at the best time to avoid traffic, is that in order to solve it one has to think not only about the answer he believes to be the best, but also about what others might think the best answer is. When coordination does not work, it can lead to paradoxical situations best described by the following quote about a nightclub: “No one goes there anymore, it’s too crowded”.

Social studies suggest that we rely on “focal points” on which people’s expectations are likely to converge. If students are asked to chose a meeting point to meet another student in New York on a given day, but they have no way to communicate with their partner, a majority of students chose Grand Central Station, and when asked at what time they should be there to meet their friend, most answer “at noon”.

Many coordination problems are solved by conventional rules. For instance, seat allocation in the subway is solved by “first-come, first-served”. When students are asked to go in the subway and ask a seated passenger to give them their seat without providing any justification, a majority get a positive answer. Yet, it is difficult for the students to ask the question in the first place because they know that they are breaking a convention.

A free market is a mechanism which can solve one of the most challenging coordination problems: getting the right resources to be invested in the right places at the right cost. In the Arrow-Debreu, every buyer and seller needs full information for this goal to be achieved optimally. However, experiments can be made where each student is given a private price which he does not share with other buyers and sellers, according to which he should decide to buy or sell a stock. Such experiments have been replicated hundreds of times in various settings and the result is invariably the same: the market reaches a near-optimal equilibrium in spite of this imperfect information.

  1. Altruism and rationality

One of the best-known experiments in behavioural economics is called the ultimatum game. In the most common setting, two players are given $100 to divide between them. The first player, called the “proposer”, decides on his own what the split should be. The second player, called the “responder”, is then presented with this split as an ultimatum: he can either accept the split as it is, or no one gets anything and the game is over. A perfectly rational responder should always accept the split, no matter how small the amount he gets is: a small something is better than nothing. In practice, however, most responders reject offers where they get less than $20. Interestingly, proposer seem to anticipate this irrational behaviour and usually propose rather “fair” splits in order to avoid the risk of a refusal.

Some monkeys have been made to play slightly different games, and a pattern seems to emerge according to which we (and monkeys) want there to be a reasonable relationship between merit and reward. When the proposer “gains” his title through a selection process, responders tend to accept more unfair splits, and the proposer proposes more unfair splits accordingly! We display a tendency to reward good behaviour and punish bad behaviour even when we get no immediate benefit from doing so, or even undergo some moderate cost. This kind of behaviour is called “strong reciprocity” and is generally useful to the group’s common good.

Similarly, the group benefits largely when its members are trusting and trustworthy, because it can spare the energy that would be needed to supervise and control everyone’s actions. In fact, over centuries, capitalism has evolved in the direction of a more transparent and trusting society and not towards the self-interested red-blooded capitalism we would have imagined. This is because in time a successful market tends to teach its members to recognize those benefits. Individuals can spend less time protecting themselves from being robbed in economic transactions and contracts are less needed. In the late 1990’s, a study conducted in fifteen small-scale economies such as tribes with various levels or development confirmed that the more market-oriented a society was, the greater its level of prosociality was.

We can easily think of situations in which everyone wants to be a free rider and waits for others to do their duty. However, a somewhat irrational trend in our individual behaviour exists so that groups do cooperate nonetheless. Without undue generalization, we could state that individual irrationality helps reach collectively rational results, and that the failure to solve coordination problems often results from individual rationality.

Paying taxes is an example of coordination situations where everyone would rather be a free rider. Evidence shows that tax evasion depends very much on what we believe other people are doing. The more we believe most people pay their taxes and there’s a good chance those who don’t will be caught, the more willing we are to pay our share. People with such a behaviour are called “contingent consenters” because they do cooperate, but only under certain conditions. They represent the largest share of the population. A smaller group is made of selfish – that is perfectly rational – people. Finally the altruists represent a small minority. Note that, despite American’s vehement anti-tax critics, they actually evade taxes much less than Europeans do…

Part II – Examples

  1. Failure to coordinate in traffic

For a few years now, drivers have been charge £5 to enter the center of London during daytime. Traffic has fallen by 20 percent, and on average cars are able to drive 40 percent faster. This type of regulation is made necessary by our failure to coordinate, which often leads to traffic congestion.

  1. Collaboration, competition and cooperation in science

Collaboration between scientists is part of a kind of division of labour, each one being specialized in a narrow field. Because scientific knowledge is cumulative, you can make discoveries more efficiently if your predecessors have shared their findings with the community. Finally, the incentive of scientists is not so much getting cash as getting scientific recognition from their peers. Reversely, showing the flaws in other peoples’ work can be a way to make a name for yourself. Scientific truth is established when a vast majority of scientists accept it without question. Yet science is not a democracy, as anybody’s voice is obviously not worth as much as Einstein’s when it comes to questions about relativity.

Research displays an interesting situation where collaboration works and benefits the group, although individuals are rewarded by their main competitors! Therefore, there is some fear that, as an increasing share of research is being financed by private corporations, there might be a tendency to protect information and keep it private.

  1. Committees

As the saying goes: “A camel is a horse produced by a committee”.

Decision makers often tend to unconsciously seek those bits of information which confirm their underlying intuitions. This is called the “confirmation bias”. Small groups tend to emphasize consensus over dissent. The information which tends to be talked about the most is the information that everyone already knows. Therefore, it is known that group deliberations are more successful when they have a clear agenda and when leaders make sure everyone gets a chance to speak.

In corporate environments, employees tend to be hired immediately after their graduation so that they have very similar culture and background and lack diversity. In addition, a phenomenon known as group polarization causes group deliberations to often radicalize rather than moderate their members’ opinions. Provided, however, that groups are given a methodology of properly aggregating the opinions of their members, they do make intelligent decisions quickly and better than their smartest members. Under such conditions, groups should be given more power and allowed to make decisions.

  1. Bosses

Zara has been quoted by an LVMH executive as “possibly the most innovative and devastating retailer in the world”. What is their secret? Zara manufactures its products itself in highly automated factories instead of giving this activity away to subcontractors in Asia. Sales of each retail shop are collected in real-time by a computer network, so that new trends in local fashion can be detected and turned into a large number of new products produced at record speed in small quantities. By being built from bottom-up, Zara has solved the problem of unsold seasonal inventories, which compensates the extra cost of production in Spain.

If coordination of small entities by market-like processes produces such results, why do we need large corporations at all? Because outsourcing everything to small specialized entities raises the cost of monitoring and contracting. For many activities, corporations are better off hiring permanent employees to work on assembly lines or in similar organizations. In The godfather, mafia business is run like a traditional top-down fashion, allowing Corleone to make decisions and get them carried out quickly. The downside is that the boss has a hard time managing his distant lieutenants and getting them to disclose the information they have. In Reservoir dogs, on the contrary, a group of individuals comes together to pull off a job and then disperses. The drawback then is that they don’t trust each other and spend a great deal of time securing transactions.

Too often, corporations are organization in a way such that bosses are unlikely to get the right information. Managers tend to be dishonest about their actual performances and problems, and such “unauthentic behaviour” is actually the norm. Internal rivalries defeat the purpose of having a formal organization in the first place, because they diminish economies of scales and increase the costs of monitoring peoples’ behaviour. You should be able to trust your fellow workers more than your competitors, but this is not always the case. Since they give everyone an incentive to work in the group’s best interest, stock options grants to a large number of employees have been shown to improve productivity and profits.

  1. Markets

After the 1929 crash, some observers were tempted to put the blame on speculators who had been short-selling stocks. Shorting has been a traditional scapegoat for all sorts of market diseases, and until 1997 regulations remained in place in the United States to make short selling more difficult or event totally forbidden for mutual funds. Of course, as it turns out, most of the real villains in market bubbles are on the long rather than short side, pushing the prices ever higher with rumours…

Short sellers are in fact useful because they are a small minority of people who tend to make their own judgement on companies rather than to turn to other investors to decide when to sell or buy back. This is in fact a difficult activity, requiring some unusual mental gymnastics to perform consistently well in an environment of negative reinforcement.

The problem with stock markets is that there is no point in time when you can be proved right or wrong like in presidential elections for instance. If everybody knew there was a point in the future when you would be rewarded according to the actual value of the company, it would be harder to speculate. A bubble usually starts as a rational attempt to cash in on a powerful business trend, and the trouble is that speculation can go on forever and the longer it lasts, the more it seems to be “normal”.

As Keynes famously wrote: “Professional investment is like a contest in which contesters have to pick the six prettiest faces from a hundred photographs, the prize being awarded to the competitor whose choice most nearly corresponds to the average preference”. Keynes recognized that investors are sometimes more concerned with what other investors think, or even what they think other investors think, etc. than with the companies’ actual perspectives. Of course, not all investors act like this, but bubbles occur when such panurgic behaviour – relayed by modern media – becomes overwhelming, and the minority of investors with strong individual opinions no longer make the difference. In this regard, the way the Federal Reserve announces its interest-rate decision with sober comments seems wise.

  1. Democracy

Public-choice theorists explain quite accurately that, due to the selfish interests of various interest groups, long-term problems tend to be deferred in favour of short-term political considerations. But they also seem to miss something when they say that voters think only of their personal condition and not of bigger social and economical considerations.

American voters seem to know remarkably little about the questions they are supposed to vote on. On average, for instance, they believe that the United States spends 24 percent of its budget on foreign aid (vs. a few tenths of a percent). So we hardly see how they could make a good decision on such an issue. Instead, in representative democracy, they are asked to pick among several candidates the one which is the most apt to make the right decisions. Even so, political decisions differ from most others because there is no objective standard to decide when a decision is “good” or “bad”. Choosing candidates and policies in a democracy are not, therefore, a pure cognition problem. Instead, it has to do with the most fundamental coordination problem of all: How should we live together to our mutual benefit?

1 commentaire:

Anonyme a dit…

Amazing!! thanks very muck for your work

New York, U.S