Book summary: chapter by chapter
Article completion date: July 2006

The Wisdom of Crowds

by James Surowiecki

PART I

PART II

About the author - James Surowiecki

An American journalist who has written a business column for the New Yorker since 2000 (The Wisdom of Crowds grew out of this column). Born in 1967 and currently living in Brooklyn, New York. He is a former columnist of New York magazine, a former contributing editor of Fortune, and his work has appeared in various publications. He has one previous publication - as the editor of Best Business Crime Stories, which is a collection of business articles chronicling the fall from grace of various CEOs.

Quick summary of The Wisdom of Crowds

Subtitled Why the Many Are Smarter Than the Few, the book explores the idea that crowds or groups (e.g. company employees, teams of scientists, stock market investors) are smarter and come to better decisions than any single individual (or elite few) in the crowd. The author claims that this is true when the crowd satisfy certain conditions (being smart crowds under these conditions). He then goes to great lengths to explain the conditions needed to make a smart crowd for certain defined types of problem. Along the way he gives plenty of innovative examples of good group action (e.g. Linux) and of bad group action (e.g. stock market bubbles); explaining why the crowd doesn't satisfy the conditions to be a smart crowd in the case of the bad actions. The obvious corollary is that we should make use of crowds much more in a variety of situations, and we should make sure we set up the conditions so that they are smart.

The Introduction immediately sets out the different types of problem that the book examines, the conditions to make smart crowds and the structure of the book. So if you want to quickly get to the heart of the book's thesis ... read the review of the Introduction.

The book was published in 2004 and runs to about 300 pages, including 20 pages of notes/references. The notes are well worth reading: they are not referenced in the main text and each note refers to a page (or range of pages); as well as giving a reference to an article or paper they often give a quick summary of that source. The book's website is at http://www.randomhouse.com/features/wisdomofcrowds/index.html - the Q & A session with the author is especially worth reading.

Introduction

Summary: The Introduction starts and finishes with an example of collective intelligence; it defines the types of problem that are examined in the book; it states (loosely) the conditions necessary for the crowd to be smart; it mentions several themes that run through the book; and it sets out the structure of the book.

Examples
The book kicks off with an example of a competition to guess the weight of an ox (after it had been slaughtered and dressed) at a Plymouth fair in 1906. Eight hundred people entered the competition, some of them butchers and farmers and some of them people with no expert information. The scientist Francis Galton took the entries and calculated the mean of all the bets as being 1197 pounds; the true value was 1198 pounds. Note that in this case the mean of all the bets is the group's aggregate decision or answer.

The second example is more complex, and concerns the attempt to find a US submarine that sank in 1968. The navy knew the last reported location, but had no idea how much further it had traveled - its initial search area was a circle 20 miles wide. A naval officer adopted a novel technique to narrow the search. He concocted a series of scenarios as to what might have happened to the submarine and invited a team to bet on what the likelihood of each scenario was. The team consisted of men with a wide range of experience - including mathematicians, salvage men and submarine experts. They were betting on things like why the submarine hit problems, the steepness of its descent, etc. The officer used statistics (Baye's theorem) to put all the answers together to come up with an estimated position - which ended up within 220 yards of the sub's position when it was eventually found. Note that the aggregation of the individual decisions is much more complex than the simple mean in the case of the ox.

Problem types
The author states that collective intelligence can be used on a wide variety of problems and that he will concentrate on three types of problem. He does not say that these are the only types of problem amenable to collective intelligence. When asked, on the book's website, about what types of problems crowds are not good at solving he comes up with 'problems of skill', e.g. surgery or flying a plane. The three types of problem he addresses are:

• Cognition: These are problems that have or will have a definitive answer (e.g. how heavy is this ox?, will the stock market go up this year?). This definition is also widened to include problems where there may not be a single correct answer, but where some answers are clearly better than others (e.g. where is the best place to build this swimming pool?).
• Coordination: How do members of a group coordinate their behaviour with each other, knowing that everyone else is trying to do the same (e.g. how do you drive safely in traffic?, how do buyers and sellers find each other in a market?) ? This is not always a set problem to solve, and can be something which just seems to happen, and sometimes very efficiently. However, as will be addressed later, it doesn't always happen as well as it could - we need the crowd conditions to be right.
• Cooperation: This concerns the challenge of getting people to work together, especially where this seems to be against narrow self-interest (e.g. why do they pay taxes?).

Smart crowd conditions
The following three conditions necessary to form smart crowds are quoted:

• Diversity of group members
• Independence of group members
• Decentralization (of a certain kind)

Themes
The following themes are introduced, and are reinforced throughout the book:

• Historically a number of scientists/philosophers/etc. have been firmly convinced of the stupidity of crowds.
• Many people believe valuable knowledge is concentrated within a few people and that we should try and 'chase the expert' to solve a problem - the book's thesis being that this is wrong; we should ask the crowd (under the right circumstances).
• Groups are given a broad definition (e.g. small teams, betters on a horse race, electorates) - and are not always aware of their identity as a group.
• For many examples of collective decision-making, it is not instantly obvious that collective action is at play.
• Good collective decisions come about through disagreement and competition rather than consensus and compromise (along with a method for aggregation of a collection decision). This is why independence and diversity are important to group structure.

Book structure
The book is separated into two parts of six chapters each. The author explains that the first part is theoretical (albeit with examples) - there is a chapter for each of the three different types of problems and a chapter for each of the three conditions for smart crowds. The second part consists of case studies of different systems (e.g. markets, traffic) and how and why collective action works or fails in these systems. My review/summary will concentrate more heavily on the first 6 chapters because this gives the theory and sets up the subject; but I am also very interested in Chapters 9 and 10, because these are relevant to any of us who work in teams or for a company.

PART I

1. Wisdom of Crowds

Summary: This chapter looks at cognition-type problems (without explicitly saying so). It mainly gives a series of examples; moving from simple to complicated problems, then to problems which require predictions of the future. It ends with some examples of decision markets that have been set up to exploit group intelligence - and wonders why corporate America makes so little use of them. It makes the point that decision markets are just one way of tapping into group wisdom - and it is not the actual method that counts, but that the conditions for crowd intelligence are met. Also buried in this chapter is a small paragraph hypothesising (or stating) why group intelligence works so well under the right conditions.

Simple example: jelly beans in a jar
A number of experiments into group intelligence were conducted between the 1920s and the 1950s. A classic example asks a class to estimate the number of beans in a jar and takes the mean of all their guesses as the group's aggregate solution. The results are summarised as follows:

• The group's aggregate was pretty close to the right answer; however a small percentage of individuals in the group got closer for each guess.
• If the experiment was repeated then the people giving better guesses than the group were different people each time. Over a number of guesses, the group's performance was better than any individual in the group.
• The author makes the point that the possibility of beating the group is good in applications where you are rewarded for this (e.g. the stock market) - since you know it is possible to win, you will keep participating [probably a bad thing for gambling addicts, though].
• The point is made that the group members were working independently in these experiments. They were not talking to each other, or working together.

Complicated example: Challenger
The Challenger space shuttle blew up in 1986. Within minutes stocks of the four major contractors involved in the launch were diving. Later that day three of them recovered and one (Morton Thiokol) kept falling. In effect, the collective decision of all the stock market investors was that Thiokol was responsible. There were no rumours circulating blaming Thiokol and no suspicious dumps of Thiokol stock by Thiokol executives or competitor company executives (which would imply insider information). It took an inquiry six months to establish that Thiokol O-ring seals were responsible. A report into the market reaction couldn't work out why Thiokol stock alone fell and assumed there must have been inside information somehow. The author disagrees and suggests that the investors met all the conditions to be a smart group, and that overall they 'knew' Thiokol was responsible.

Other examples are given - the illustration of the Google search engine is interesting.

Future predictions: sports betting
In sports betting the aim of the bookies is to balance the amount of money paid out to winners with the money bet on all the other losing outcomes (and if they get this pretty much right, there's a small profit built into the system). To do this the bookies set out initial odds and then adjust the odds as bets come in - for example, if a favourite is heavily backed they'll shorten the odds to make it less attractive; hence the final odds are determined by the bettors. So here we have a system that allows a diverse, independent group of people to give a final aggregate decision on the likely outcome of a contest. Note the subtlety - the bettors are not giving a decision on the outcome (if they always got this right, you could just bet on the favourite and make your fortune); they are giving a decision on the likely outcome. Thus horses whose odds end up at 3-1 are likely to win once out of four; if a spread bet finalises with one team tipped to win by 5 points it is likely to make the 5 point margin half the time; favourites should win most often, second favourites second most often; and so on. And ... this pretty much works out; even professional gamblers struggle to beat the odds. The author makes a few notes: 1. Gamblers are more likely to make money in more obscure sports where there are smaller numbers of bettors and a few bets can significantly change the odds; 2. There are a few quirks in sports betting - e.g. horse bettors bet on longshots slightly more than they should; the author puts this down as psychological (risk-seeking or going for the big return); 3. If the bookie consistently gets the initial odds wrong the bookie is likely to take bad bets (until the bets coming in stabilise the odds) and lose money.

Decision markets
Well, since sports betting shows how good groups are at predictions, why not set up similar systems to predict other things ? Examples could be: what percentage of votes is a candidate going to get in an election; our company is thinking of developing certain products - how well will they sell; which films are going to win Oscars this year; what will be the next technological developments; pretty much anything you can think off. The author describes several such decision markets that have been set up, including Iowa Electronic Markets (IEM) used to predict elections and the Hollywood Stock Exchange (HSX) used to predict box office returns and Oscar winners. Again, he demonstrates impressive performance; for example IEM (with 800 participants) generally outperform polls. One point to make about these decision markets are that they are set up so there is an incentive for people to get it right, although it doesn't need to be a massive incentive - in the case of IEM a small amount of money is wagered; in some systems it is simply status and reputation that is at stake (they are games and use pretend money or points). It seems that larger financial rewards will lead to a slightly better performance, though. A couple of other points are made - the participants are diverse and independent, but they are not necessarily representative (e.g. IEM participants are mainly men from Iowa); and people are not betting on their own behaviour - they are betting on what they think other people will do (as opposed to polls, which asks people what they will do).

Why are crowds wise ?
Why are groups that satisfy the right conditions wise ? The book doesn't say too much about this - the book is more about demonstration by multiple examples (with analysis as to whether or why the smart conditions are met) than a theoretical analysis of why this might be so. However, hiding in this chapter is the following: when you ask a large enough group of diverse independent people for an estimate or prediction, each person's answers (or what they based their answer on) will contain both information and error - and when you put them all together the errors cancel, leaving an intelligent collective answer/decision. [Comment: This is entirely unproven here, but probably not far off.] The author makes three additional comments - 1. that there needs to be some decent information in the information part of 'information plus error' (so, e.g. don't ask a group of 5 year olds about pensions); 2. that there really is a lot of good information in a collective decision; and 3. in most endeavours the average is mediocre (try averaging running times or exam scores) and much worse than the best.

2. The Difference Difference Makes: Waggle Dances, the Bay of Pigs, and the Value of Diversity

Summary: The chapter starts with a description of the beginning of the car industry in America around 1900 [chosen as an example; it could have been any industry]. The point being made is that a lot of companies created lots of very different solutions and models, but these were gradually whittled down as problems became apparent (e.g. with steam or electric cars) and techniques were developed (e.g. mass production assembly lines), until only a few companies and technologies were left - these being the selected solutions. The importance of diversity in arriving at good solutions is stressed and discussed. The waggle dance of the title describes the behaviour of bees finding and choosing between food sources and an analogy is drawn between this and the car industry example. A point made is that the example problems here require twofold solutions - first various alternative solutions need to be discovered, and then they need to be selected from. The performance of individual experts is discussed, concluding they're not very good in lots of situations and that having the influence of more diverse opinions helps reach better decisions. The chapter concludes with a discussion of the negative effects of a lack of diversity on groups - groupthink and peer pressure - using the example of the ill-fated Bay of Pigs Cuban invasion in 1961.

Finding options and solutions - and diversity
I have sketched out the example of the beginning of the car industry in the Summary above. The author is giving this as an example of a good system to discriminate between emerging technologies and designs, and implies that the model is typical of many industries. He says that an important factor helping to develop a car industry that satisfies the market is diversity. By this he means: diversity in the entrepreneurs (or inventors) creating cars - so you will cover a large range of design alternatives; diversity in the people financing or managing them - so that people are likely to take a more radical idea seriously (if all the people holding the strings were similar they would likely invest the same way); and diversity in the people choosing the winners (the market in this case). A couple of points here: 1. diversity is meant in a general rather than a sociological sense - i.e. diversity of opinion, attitude, background, personality, etc.; 2. the comment about needing diversity by the people choosing the solution is pretty much a given in this example, because we are talking about a large consumer market which will contain a lot of diversity - it is usually in small groups where this diversity may be missing.

The waggle dance is this: when a hive of bees wants to find nectar they send scouts out in random directions. When a scout finds nectar he comes back and does a waggle dance - the intensity of the dance is related to the quality/size of the nectar source. Bees will thus choose to follow a scout depending on its waggle dance - so lots will follow the scouts who have found good nectar, fewer will follow those who have located sparser patches and some will get no takers. As a solution to the twofold problem of locating nectar with no initial information and then spreading the bees efficiently across the nectar locations, this is pretty sophisticated ... and analogous to the way in which the car industry evolves.

Some notes on diversity

• Taking the car industry and similar examples, they seem very wasteful systems - most ideas and endeavours end up as failures. However, it seems impossible to pick out successful products in advance, judging by how many new products are released and the small percentage that succeed. The opinion is given that good ideas and bad ideas come together - to get the successful ideas you need a system that encourages a heap of good and bad ideas. The important thing is that the system recognises losers and kills them quickly.
• Diversity works because it means more information is available to the group - it adds perspectives that would otherwise be missing. It also works to counteract any bias or groupthink in a group.
• Scott Page at the University of Michigan has done experiments on problem solving with computer-simulated software agents. He set up the agents with a number of skills and made some of them very smart and some not-so-smart; then he had a group of them solve problems. He found that a group of very smart agents were less good at solving problems than a group of some very smart agents and some not-so-smart agents. The theory here is that very smart agents/people tend to resemble each other - so if you have a group of very smart people (in a particular area) the best way to add to the group's intelligence is not to add another very smart person (who won't actually add much knowledge to the group), but to add someone less smart (in the relevant area) but who will have a different background/experiences and hence will bring new information to the group. That is, diversity for its own sake is valuable - although you do need intelligence in there as well.

Experts - how good are they ?
The thrust of this chapter is that a diverse group of people with varying skills and insights will be better than one or two 'top' experts when it comes to decision making. This conclusion leads us to question the value of experts. If you have a healthy scepticism of experts read section 3 of this chapter - you'll enjoy it. Basic conclusions are that expertise can be 'spectacularly narrow' and that it may not be possible to be an expert in something as wide-ranging as forecasting or decision making or strategy. Note, you can be an expert in narrow, skilled roles like piloting a plane, car repair or heart surgery, and the comments here don't apply to these types of roles. Along the way we look at the performance of fund managers (pretty rubbish against the market) and studies showing non-psychologists predict people behaviour better than psychologists. There are a lot of studies quoted in the notes for the performance of experts in a range of fields. Several points stand out:

• Consistency between experts: studies in a variety of fields have found that the between-expert agreement is less than 50% - that is, experts are as likely to disagree as agree.
• Internal consistency - one study found an internal consistency of medical pathologists was 0.5 - that is, given the same evidence a pathologist would give different opinions half the time.
• Overconfidence - like the rest of us, experts are overconfident of their own abilities and to about the same degree.
• Calibration - this measures a sense of how well you know the accuracy of your judgment; the answer is generally not very well for experts. An exception is made for weathermen and expert bridge players - if a weatherman predicts a 30% likelihood of rain he is more or less right.
• There was no evidence that if most experts are not that good, a few are excellent. The author admits that you can't rule it out, but makes the points of 1) how do you find them, and 2) since diverse groups are intelligent anyway, if the group covers a lot of the skills required you don't need that one expert anyway.

The conclusion here is that the opinions of experts should be pooled with those of a diverse group of others, and that you don't need to chase that one expert to get the best solutions. Some implications are - your company doesn't need that killer CEO to succeed (the best CEO is likely to be someone who appreciates and uses the value of diverse opinions); get a second (and third) opinion where necessary. However, the author states that most people do try and find the best expert - 'the crowd are blind to their own wisdom'.

Problems with lack of diversity - groupthink and pressure to conform
The psychologist Irving Janis used the term groupthink to describe what happens when a team are too much alike in worldview and mind-set. He studied a number of American foreign-policy disasters, including the Bay of Pigs, and concluded that groupthink was at play (pretty badly - read it for the details). Groupthink happens when a group are isolated from certain perceptions: information that may challenge their view is rationalized away or excluded; this bonds the group together, further re-enforcing their viewpoint and leading them to develop an illusion of invulnerability and an attitude that dissent is not useful. It is more brought about by an attitude of assumed consensus than any censorship.

Another problem in non-diverse (or homogeneous) groups is peer pressure or the pressure to conform. This differs from groupthink in that group members don't actually change their mind, they just say that they do for an easier life. A classic experiment by Solomon Asch demonstrates this - a line of people (actors) state an obviously wrong answer to a task; the subject, at the end of the line, will often go along with this to conform. An interesting corollary to this is that if the subject is given just one ally (someone else who says the obviously correct answer) this makes a huge difference and the subject is much freer to give their true opinion. The author concludes that diversity helps groups both by adding new perspectives and by reducing any need to conform.

3. Monkey See, Monkey Do: Imitation, Information Cascades, and Independence

Summary: This chapter is about the importance of independence in groups. Independence is important for two reasons: it allows errors to cancel out - whereas if a group are dependent on each other for information, then there is likely to be a systematic bias in errors and they will add up; secondly it is likely to bring in new information to the group. A point to note is that independence doesn't imply rationality or impartiality. The chapter shows that achieving independence is more difficult than you would think, mainly because we are social people affected and influenced by those around us. The chapter examines a number of impediments to independence - social proof, imitation, herding and information cascades - along with examples. It is also noted that some of these behaviours are good to an extent - the trick is in knowing, e.g. when to imitate the actions of others and when that leads to a counter-productive loss of independence. Note that diversity will help to foster independence, but that diverse groups can still fail to be independent.

Social proof and imitation
People are positioned in a particular social setting, mix with their own social networks, work long-term in organisations. How can they not be heavily influenced by their social setting ? Well true, and in everyday life this is helpful in regulating social norms. It is not so good for decision making if everyone thinks the same way and makes the same mistakes. A couple of examples are given. The first one describes a simple experiment whereby people looked up at the sky on the street to see if passers-by would copy them - when one person did it most people ignored this, but the more people that were initially looking up then the more passers-by stopped to look up. The second example concerns American football (whatever that is): the example concerns the decision making involved (once you have been awarded the ball) in choosing whether to kick for goal (worth 3 points) or to drive and go for a touchdown (worth 7 points, but less guaranteed) at various positions on the pitch. An economist did a statistical analysis of hundreds of games to work out the best strategy at each pitch position. The conclusion he made is that coaches were much too cautious - they kicked much more often than they should to maximise their points. The implication is that they all adopt this strategy because everyone else does. Social proof is the tendency to believe that if lots of people are doing something there must be a good reason. The author cites both these examples as demonstrating social proof at play - quickly leading to imitation. The tricky thing here is that if groups are wise, then following the group can be an intelligent thing to do. The trick is to recognise when too many people are following the group and not enough are involved in independently shaping (or challenging) the group decisions.

Herding
Herding achieves the same effect as imitation due to social proof - but the reasoning behind it is more subtle. In imitation, you imitate people because you think lots of people are doing it so it must be right. In herding you go along with the crowd because it is safest. If everyone else is doing something and you follow, then if they're all wrong your failure doesn't stand out; alternatively if you buck the trend and your idea doesn't quickly work out you will be very exposed. An example given is that of fund managers - they tend to follow the same kinds of strategy and invest in the same types of stock. This reduces the overall intelligence of the managers, since they are not bringing their own information to the table but imitating others. The reason given for this is that they have to do two things - invest wisely and convince people they are investing wisely. The second part makes it risky for the managers to do anything different. Although an alternative strategy may make more money long term, if it doesn't work out at first investors will see worse performance and they will see the reason for this as being someone who is not doing the same as all the other experts. A quote from John Maynard Keynes "it's better to fail conventionally than succeed unconventionally" gives a succinct summary.

Information cascades
An information cascade works like this: say you want to choose between solutions (an example given here is which of two new restaurants is best). The first few people to choose get certain information that one solution is best and take that option. The next few people have some independent information, but they also have the information that more people have chosen one solution - so they go for that one. It is possible for this process to run away, meaning that everyone chooses one solution because everyone else has done so (the restaurant ends up crowded, so it must be best). A large factor contributing to the emergence of a cascade is that people choose in sequence, so that people can follow the initial choosers - if everyone had their information and chose at the same time, you have better conditions for an intelligent group decision. The problem arises when people stop paying attention to their private information and instead imitate the actions of those before (who may have been doing the same thing) - in other words, the later choosers are not adding to the group information. Another example given is that of plank roads which took the US by storm in the 1840s, until it was discovered they didn't last.

An alternative description of the way cascades work is discussed (proposed by Malcolm Gladwell in The Tipping Point). This model stressed the importance of particular kinds of people - mavens (mediators), connectors and salesmen. In this model it is not so much the early adopters who are influential but that certain people are much more influential and people look to these people. This is interesting and I take this to mean that you can start a cascade by influencing the right type of people. Indeed, the author describes a beneficial cascade (the adoption of a screw standard in the US) where the manufacturer targeted the most influential people to adopt his design. This also highlights the point that cascades can be either beneficial or harmful - leading to the right or wrong decision; the author concluding that they are not the best way to make decisions - we want those diverse, independent groups.

When and how to imitate
The essence of the rest of this chapter is that imitation does work and is useful when it's intelligent imitation rather than slavish imitation. Effectively we don't want to have to learn everything from scratch, but we do want to be able to challenge established behaviour. Intelligent imitation is conditional (on good results) and means that people must be willing to stop imitating and learn for themselves when appropriate. Two things are quoted as necessary for intelligent imitation: an initially wide array of options and information; and the willingness of people to put their own judgment first. The author says that there are a large number of people willing to back their own judgment since people are generally overconfident - this may not always be good for them, but it is good for society as some of their ideas will be good and it will increase the information and possibilities available, and make cascades less likely to happen. People in general are also more likely to investigate on their own, the more important a decision is. There are several good examples and experiments described here. Since he is saying that people will challenge the group, why do cascades still happen ? Well, I guess the answer is a combination - there don't seem to be many options or much information; herding has already taken over; the decision making process is not set up to value independence.

The chapter concludes with the description of an experiment that shows how you can set up a decision making process to value independence - it does this by making the decisions simultaneously rather than sequentially. Two urns are provided - urn A has twice as many light marbles as dark ones; urn B is the opposite. People stand in a line and all pull out a marble from just one of the urns (the other is taken away) - they are rewarded for guessing which urn they are using. No-one but themselves can know what colour marble they pulled out, but they hear all the other guesses. So if the first person pulls out a light marble, they are likely to guess A; however if someone pulls out a light marble, but the 3 people before said B, then that person is likely to say B. Quite often a cascade starts here (78% of the time). Then the rules were changed so that people were rewarded if the group as a whole got the decision right - that is they put all their guesses together at the end and came to a decision. In this case people relied much more on their private information (the marble they pulled out), rather than other people's information. And the collective judgment became much more accurate. Fairly obvious, but a good demonstration all the same.

4. Putting the Pieces Together: The CIA, Linux, and the Art of Decentralization

Summary: The chapter starts by analysing American intelligence forecasting failures both at Pearl Harbour and more recently (culminating in 9/11). Both times inquiries blamed the intelligence agency's decentralized structure (there's a multitude of different agencies) and called for a centralised structure. The main point of this chapter is that decentralization is not the problem - decentralization provides specialised, local knowledge and decisions - it is the way in which decentralization is practised. The key component missing with the intelligence agencies is a method such that useful knowledge in one (decentralised) part of the system gets distributed to the rest of the system - that is, a way to aggregate the information and judgments. Along the way we look at how Linux achieves this aggregation and how the US tried to set up a decision market in intelligence information (and the subsequent furore).

What do we mean by decentralization ?
A decentralized system is one where 'power does not fully reside in one central location, and many of the important decisions are made by individuals based on their own local and specific knowledge rather than by an omniscient or farseeing planner'. Several examples are given (without discussion) - flocks of birds, free-market economies, cities. Two key points about decentralization are noted: 1. It encourages (and in turn is fed by) specialization - which both makes individuals more productive, and increases the amount of opinions and information in the system; 2. It encourages 'tacit knowledge' - this was described by Friedrich Hayek as knowledge that can't be easily summarised as it's specific to particular circumstances, but is very valuable. Decentralization assumes that someone is more likely to have a solution the closer they are to the problem. A quote from Exodus shows the age of this principle - Moses' father-in-law advises him to judge only in great matters, leaving other decisions to local leaders. The author summarises that the aim of good decentralization is to find the right balance between allowing the crowd to be specific and local, and aggregating all the information into a collective whole. There is an interesting discussion in the notes about how Hayek (a great believer in decentralization) would likely disagree with any attempt at aggregation and how successful it would be.

Linux example and aggregation
A quick description of the Linux open-source operating system is given. Linux is developed by programmers all over the world. Anyone is free to send in code for bug fixes or new functionality. Lots of people do, working on the bits they are interested in. The determination of which of this code can be used (some kept, some rejected, some combined) is done by Linux's creator Linus Torvalds and a select group of programmers - this is the aggregation of the decentralized information coming in. This leads the author to state that this aggregation, which is a form of centralization, is paradoxically needed (in this case) for the success of decentralization.

Intelligence gathering and decision markets
The last topic of the chapter concerns the FutureMAP program, an attempt by the intelligence community to set up decision markets where individuals could buy and sell futures contracts based on what they thought would happen. There were two parts to this. The first was a set of internal markets open to a small number of specialists (from different agencies) who would try and predict the likelihood of specific events - for example, one might be centred on Middle East events. The second was the Policy Analysis Market (PAM) - this was similar, but was open to the public. This caused a storm on the basis that people could profit from bad events, e.g. assassinations - and was killed by the US Senate. The author attacks this as throwing away an idea that may provide good information, and says that people are always making decisions or bets on the cost of life and death (e.g. life insurance companies, government analysis of likely world events). He does raise an interesting point - if something like PAM made accurate predictions, but the government acted to change predicted events, then the predictions would no longer be accurate. In this case you could, perhaps, rig the market to make a trade good if it provoked preventative action; or, perhaps, the market would become more sophisticated and take into account US actions to predictions. The point is also made that a simple way for the intelligence agencies to bring some aggregation and improvement to their decentralized parts would be to link their computers, which would allow teams to see significance in data considered noise by other teams with a different perspective.

5. Shall We Dance?: Coordination in a Complex World

Summary: The chapter starts by describing a study (1969 - 1985) of how New Yorkers used their city. In reference to pedestrian movements (captured on time-lapse cameras) the observations showed 'lots of small, subtle adjustments in pace and stride and direction' adding up to a smooth pedestrian flow. People 'are constantly anticipating each others' behaviour' and 'decide for themselves what they'll do based on their best guess of what everyone else will do'. This is the basic summary of a co-ordination problem - you not only have to decide what you think the correct answer is, you have to anticipate what other people think the right answer is as well; what everyone does affects everyone else. The author states that these types of problems don't generally have easy answers and solutions in the real world are often good not optimal. A lot of what the chapter suggests is that solutions spring up through custom, institutions, norms and history - and often (not always) these are pretty good. Often people are not actually trying to solve the whole problem, they are following rules that work for themselves (using private information) and the whole solution springs up out of many people doing this - examples of bird flocks and the free market are given. In fact, the denouement of the chapter is a description of the free market as a giant coordination problem. The point is made in passing that, for the simpler examples, you may be able to solve the problems by use of a central planner ... at the cost of people's freedom to make decisions.

The crowded bar
One problem that seems to have been well studied is the problem of how to decide whether to visit a bar - the nuts and bolts being that if it's too crowded it won't be any fun. So people have to decide how many other people they think will go. Computer simulations were initially run on the basis that different people would choose different strategies to determine whether to go (assuming a 60% attendance was optimum). Some people would assume the same number would turn up as last time, some would use an average of the last few times, etc. The simulations showed that over a period of time, the average attendance was exactly 60%, although it fluctuated week to week. A different simulation assumed that people would go back if they had a good time (less than 60% full) on their last visit - in this case the attendance converged to 60% (albeit with a large number of regulars). The point here (I think) is that it is not obvious which is correct, but that in the real world reasonably good solutions often simply turn up like this. Doesn't explain why the Hog's Head is always packed; I don't know why I keep going back there.

Schelling points, customs and queuing
We are getting into custom here. Thomas Schelling told people they had to meet someone in New York, but they didn't know where or when and they couldn't contact them. A large number of people said Grand Central Station at noon, so would have met. Similarly, if you try and match 'heads' or 'tails' with a partner most people will say 'heads'. These focal points are cultural and are called Schelling points - they show how people can sometimes coordinate with each other without direction or communication. This leads onto a discussion of customs like seating on trains and queuing. The normal rules are ingrained in people here - first come, first served for seating, and everyone takes their place at the back of a queue. Experiments are described where people break these rules by asking for a seat from someone seated, or by pushing in in a queue. Both times, the behaviour was accepted or tolerated about 50% of the time. A main point was that the experimenter found it very difficult to break convention by actually carrying out the experiment - this is called internalisation of a norm; there's less need for policing when it's very difficult for someone to bring themselves to break the norm. However there is some informal policing as the 50% of rejections showed - predictably the queue breakers suffered a more virulent reaction when their behaviour wasn't tolerated. It is stated that the seating and queuing conventions are not optimal - it takes no account of how much people want to sit down or how much of a hurry they are in. In theory you could have a system where people bartered for queue places, which would satisfy people's needs better. However, this would probably be so complex that it would outweigh any gain - the social ease of these customs is powerful.

Custom in business - movie theatres
There follows the example of movie theatres (translation: cinemas) following custom to always charge the same for film admission. This makes little sense, since unpopular movies play to empty audiences - why not cut the price. Similar examples of custom in business are companies not cutting wages in recession (but laying people off instead), pricing not adapting to rapidly changing circumstances (with exceptions like American Airlines or Walmart which continuously change price to conditions), and CDs being priced outrageously. These are more or less given as examples of businesses not coordinating themselves with customers and suffering less than optimal rewards because of it. There is not so much about why this is going wrong here in terms of lack of independence, etc. - I guess the point is that customs may be good, but you must be ready to look at the results and challenge them when appropriate, and not herd behind them.

Bird flocks ... and the free market
Finally we get a description of bird flocks. Apparently they are not acting in concert, but each bird has four rules based on their own local knowledge (the last of which is 'if a hawk dives at you, get out the way') - with the result that you see a beautifully coordinated flock. The parallel is made with the free market example of orange juice being available for you at a price you are willing to pay when you go to the supermarket. A whole bunch of farmers, manufacturers, distributors, etc. have coordinated with you to produce the right number of cartons for us to turn up and buy - and none of them know you! They are all essentially using local knowledge and their own rules (pretty much to maximise their own profits), without knowing the whole picture.

A little on economic theory
That last discussion on orange juice leads into a little on economic theory about how the free market can optimise the gain to everyone - sellers sell at lower than they would like and buyers buy at higher than they would like, but overall the price converges on the market price that maximises the total gain for everyone. This is established in economics theory (general equilibrium theorem), but how true is this in real life ? Well I've leave you to read this section at the end of Chapter 5, and feel free to pursue further through a lifelong study of economics. An interesting point made, though, is that for consumer markets it's a lot easier to arrive at that correct price than for asset markets like the stock market (and we'll see examples of stock market cascades later). The conclusion is that people with their own private information (and not a whole picture) coordinate in the free market to produce very good pricing and distribution solutions.

6. Society Does Exist: Taxes, Tipping, Television, and Trust

Summary: This is the longest chapter in the book and probably the widest in scope. It is about cooperation problems, which are introduced by an example from Italian football. After the 2002 World Cup all of Italy was convinced that Italy's elimination (by South Korea, including a disallowed Italian goal) was the result of conspiracy, rather than that incompetent refereeing decisions contributed. This comes from the Italian's attitude to football which is stated as cynical, and corruption is assumed normal in Italy; this is fed by media scrutiny of referees and leads to games being negative and foul ridden, with players trying to 'work' the referees. The point here is that Italian footballers need to do two things: i) to win the game for their own team (they have a competing interest with other teams); ii) to create entertaining games that means the whole football industry benefits by increased ticket sales, TV revenue, etc. (they have a common interest with other teams). The difference between cooperation problems and the coordination problems in the previous chapter are as follows: the coordination problems can be solved by people pursuing their own self interest; cooperation problems require people to also work together to achieve a common interest. A good example covered is tax paying, where from a narrow self interest point of view, people should avoid paying taxes if they can, because they will still benefit from everyone else's tax. Other examples given are paying money to charity; voluntary work; communities clearing the snow on the street; not shirking your work when it can be covered by other people; there is also a long section on how TV ratings are measured, showing a failure of cooperation. The author pushes the point that a key element to solving cooperation problems is trust. This leads to a discussion of the evolution of capitalism and how trust was an inherent part of getting to a system where complete strangers trade with each other with ease and efficiency; and also what can go wrong, as in Enron-type corruption. Several examples of experiments in behavioural economics (like the Ultimatum game) are given - these show that a key element to trust not breaking down is that people who break the rules are caught and punished.

Accomplishment and reward - the ultimatum game
An example is given of a US CEO who ended up being fired because of public clamour against a very high payment package. The author looks at how irrational this was, since the package has no impact on the public's lives (no public money was involved) and they gained nothing from forcing the CEO out. This leads to a discussion of how people will act against perceived unfairness, even if they gain nothing. This is demonstrated by the ultimatum game. Here, two people are paired and are given $10 to split between them. One person decides what the split should be and the other person either accepts it or rejects it; if it's rejected neither of them get any money. Rationally, the proposer would offer $1 and keep $9 himself - if this is rejected then the responder has passed up $1 for free. However this doesn't happen. Low offers are routinely rejected - people would rather have nothing than let the proposer have most the money. In fact the proposer realises this and the most common offer is £5 each. A variation of the game makes it look as if the proposer has earned the position by doing better on a test. In this case, the proposers make lower offers and these are generally accepted. The conclusion is that people want there to be a fair link between accomplishment and reward, so in this case they believe the proposer should keep more. The rejection of unfair offers is called 'strong reciprocity' - the willingness to punish bad behaviour with no personal benefit. This is described as a 'prosocial behaviour' since it pushes people to serve the common good (in the case of the CEO, it encourages more rigour in establishing a CEO's real value in future). There is a brief discussion on culture as determining what is seen as fair - the US are more likely to believe wealth is the result of skill and work than Europeans who are more likely to put it down to luck.

Why cooperate ?
The author initially quotes Robert Axelrod who argued that the reason for cooperation is the result of repeated interactions with the same people. This is the 'shadow of the future' as people know that if they don't cooperate they will be punished in future dealings. However as the author (and Axelrod, later, seems to say) this is not quite satisfactory because people cooperate with strangers. They still tip in restaurants they will never visit again, they donate to charities, trade fairly on one-off deals. The author expresses preference for Robert Wright's answer: that people have learned that trade and exchange are games where everyone can gain as opposed to zero-sum games with a winner and loser - read the notes on this, Robert Wright's argument is much more subtle than this one sentence. The author notes that different cultures have different ideas about trust and cooperation and this primes him up to argue that one of the reasons for this is how far advanced capitalism is in a culture ...

Capitalism - evolution and trust
Section four of this chapter gives some examples of how capitalism evolved in Europe in the 18th and 19th centuries. The key points are:

• Trade used to be more between people who knew (or knew of) each other; and outside of this, honesty in business transactions was much less certain.
• Trading groups (who traded amongst themselves) sprang up with certain rules to encourage fair trading and punish offenders - for example the Quakers in the UK.
• People began to trade with these groups because they were trusted.
• These groups prospered and the link between trust-worthiness and prosperity was established in people's minds.

Putting trust at the centre of the system can create a problem, since this makes it easier to become corrupt and exploit people. This is why there is a system of laws and regulations to hold the framework in place. As we'll see later one of the things necessary for people to cooperate is that they trust wrongdoers will be caught and punished. This leads to a discussion of the Enron-type corruption in the late 1990s. The author's summary of what went wrong goes like this:

• Stock prices were vastly inflated so people could make far more money in the short term than by long term fair dealing. The 'shadow of the future' had been eliminated since people could make a short term fortune from stock options, and there was no need to worry about the long term health of their companies.
• Investors didn't realise this or didn't care because they were making money on the back of heavily rising share prices.
• Institutions supposed to offer checks and balances (accountants, underwriters, the stock exchange) weren't punished for doing a bad job - in fact they got more business. So, for example, when Arthur Andersen was the auditor in accounting disasters, it didn't lose work for it and companies it subsequently audited weren't scrutinised more closely and sceptically.

Taxes
Paying taxes is a cooperation problem. Everyone needs to cooperate to provide the money for national defence, social policy, etc. Everyone benefits from this, but they will still benefit if individually they avoid taxes. Nevertheless tax avoidance rates have generally been low in the US. The point developed is that people will pay taxes as long as they believe others are doing so, or that they will get caught and punished if they don't. However, the 1990s did see an increase in tax avoidance systems and more scepticism from the Americans of the fairness of the tax system. A magazine article is quoted called 'Are You A Chump', which asks whether people are being mugs by not avoiding taxes when lots of people are. The point being made is that people will stop cooperating if they believe wrongdoers are getting away with it. The psychology behind this is demonstrated by a social experiment called the public goods game. Here there are four people in a group, each with twenty tokens. There are four rounds and on each round a person can either contribute tokens to the pot - which means each person in the group gets 0.4 tokens - or keep them; people know how much goes into the pot, but not who put what in. If everyone keeps all their money they will each end up with twenty tokens; if everyone invests all their money everyone will end up with 32 tokens - but, to maximise your income (in theory) people could keep all their money and free ride off everyone else. What generally happens is that people put about half their tokens in to start with, but as they see others free riding, contributions drop on subsequent rounds. A study showed that people fall into three categories - about 25% always free ride; a few people are altruistic (they contribute despite free riders); and most people are conditional consenters - they contribute initially, but stop when they see others free ride. To make this work better, the key is therefore to stop people appearing to be suckers. The game was changed so that people could see what everyone was contributing and so that free riders could be punished (paying one third a token would take away a token from a free rider). In this case, people did pay to punish free riders; and the free riders started contributing. This pretty much leads to three rules of successful tax paying, that can probably be generalised to any kind of cooperation:

• People trust, largely, other people to pay their taxes
• People trust that their efforts will be used wisely (the government will spend taxes efficiently and well)
• People trust that tax evaders will be caught and punished

PART II

7. Traffic: What We Have Here Is a Failure to Coordinate

Summary: This chapter discusses traffic flow and how it can be improved. It starts with discussing congestion charging and how this makes clear the congestion cost (driving in rush hour creates more congestion and costs more), leading to drivers making a more informed decision as to whether to drive at that time and hence a reduction in congestion. The second section describes some conditions that cause congestion, talks about why coordination is difficult, and dips a little into traffic engineering. The last section talks about some potential technical innovations to help solve congestion issues.

I am not entirely sure about this chapter. Traffic flow is a coordination problem, but largely not for the drivers. We don't want the drivers to be independent, diverse or decentralised - we want them to travel at the same speed at the same distance from each other. To be fair, I think that is what the author is largely saying, and his solutions focus on the traffic engineers and technology - so as long as we have a diverse, independent bunch of traffic engineers ! The drivers do want independence in terms of when they take the car and congestion charging can give them more pause for thought here (although note my comment later on that congestion could be its own cost). I think the author has missed a couple of points here. In the chapter on coordination, he spoke about culture sometimes being a cause of coordination failure. Here, I think that despite flexible working practices and technology making home working easy for many, the culture of standard office hours is ingrained enough that most people still venture out into the morning rush hour (the work run ruinously coinciding with the school run). The second point would be that why do other forms of transport not spring up and compete better than they do, despite high congestion. A lot of the answer is the initial and final journey segments - for train and air journeys you need to drive to the station or airport. What innovative ways are there of reducing the reliance on the car for these segments ? Why are new train stations not built more often ? Answers on a postcard please.

Influencing the number of cars on the road
Congestion pricing is discussed using London and Singapore as examples. London has a £5 toll for cars between 7am and 6:30pm. Singapore started this way but is much more sophisticated now. Each car has a smart card in it and drivers see the money disappearing as soon as they are in a pay zone; different times and areas have different rates. And both these systems have worked in that congestion is better; especially so in Singapore. Here are a few points about this:

• A potential disadvantage or unfairness is that the rich get access to congestion-free roadways and everyone else has to adapt. This is mitigated in both examples, since the money collected is used for investment in better public transport.
• The idea of charging more at times of scarce resource (e.g. at rush hour when the roads are clogged) is not new. For example telephone charges have peak and low rates or flights are cheaper at inconvenient times.
• Note that pricing people off the road and having too little traffic is bad as well as having too much traffic.
• Although the charges are imposed with the intent to engineer the correct car rates, the decision as to whether to drive is in the hands of the individual - given the right price, they will all make individual decisions that create the right levels of car flow. Mexico is quoted as a system where the authorities did try to impose when people could drive by only allowing certain registration numbers to drive on a certain day - people generally found ways round this (or, presumably, were very inconvenienced when they really did need to drive).

What causes traffic jams
Traffic is mathematically described as a wave motion, and the mathematics is an advanced subject. Two quick points to note here are that once traffic slows or a jam is started the resulting congestion moves backwards up the road behind the problem point; and that once jams form they are quite stable and take a long time to loosen. There is some discussion here on what actually triggers traffic jams and whether they can occur spontaneously (with no trigger). As well as obvious causes like roadworks or accidents, it seems jams can arise from very small issues like a car going too slowly. One experiment put a car on a free flowing multi-lane highway with instructions to drive slower than the current traffic. Congestion quickly started to build up - lanes started going at different speeds, drivers changed lane more often, more space was left between cars (because everyone was changing lanes). The author says here that drivers find it harder to coordinate with each other. They have a lot less information than pedestrians on a sidewalk; they only have information about other driver's intentions via brake lights and indicators. And here, I think, is the main problem. Drivers don't have private information to make decisions and come to a collective smart solution. There are a very limited number of things drivers can do once they have decided to hit the roads - e.g. they can't detour over the fields and nor do we want them to. One thing that helps (which the author doesn't mention) are congestion monitoring devices linked to satnav, so that drivers can be warned about upcoming traffic and be advised as to the best alternative; similarly traffic news on the radio adds to their information.

How to reduce traffic jams via technology
A few ideas are sketched out here:

• Driver homogeneity - cars need to be equally spaced and to drive at equal speeds to create smooth traffic flow. Experiments have been done with automated car driving, where coordination with other cars is achieved via radio transmissions. The experiments sounded impressive, with platoons of cars traveling at 65 miles per hour for hundreds of miles. Not practical anytime soon though. This is also a top down solution, taking control completely from the driver.
• Driver assistance - here the car speed would be automatically adjusted via a radar and sensor system monitoring the car in front. Simulations show that you only need 10 to 20 percent of cars with this type of system to improve things considerably. Apparently this is starting to be incorporated now.
• Install a more intelligent method to regulate cars entering the highway. The system would need to monitor the flowing traffic and link this to lights allowing drivers onto the highway, so that they entered at breaks in the traffic flow. Hmm.

8. Science: Collaboration, Competition, and Reputation

Summary:Mostly this chapter is a paean to how science has managed to organise itself into a body where information is freely exchanged and available; and where collaboration between independent, diverse people (who are, paradoxically, competing with each other for recognition) is routine. The chapter starts with the example of how labs worked together to identify the SARS virus, and then leads on to discussing how scientists collaborate on papers, and how scientific information is published for acceptance or rejection by the scientific community. The chapter finishes with a discussion of the changes that may be seen with more scientific work done by companies - a possible issue being companies suppressing results they don't want seen; note the patent system ensures companies still make available information to protect their discoveries, hence others can build on these (but not directly exploit them). Along the way, a few flaws are pointed out in the way science progresses: the main one is scientific hierarchy, where scientific papers are much more likely to be read and referenced if they are by the most recognised scientists; hence information can be largely lost if it is not published by one of the top scientists. It does, however, turn out that the top scientists collaborate much more and more widely (i.e. internationally).

Having given a quick summary, I'll leave you to read the chapter, if you don't mind.

9. Committees, Juries and Teams: The Columbia Disaster and How Small Groups Can Be Made to Work

Summary: This chapter is about the performance of small groups - the mistakes they can make and how their performance can be improved. Two differences small groups have from some of the larger groups we have looked at (e.g. stock market investors) is that they are very aware of their identity as a team, and that the influence of individuals can play a significant role in group decisions. The chapter looks in detail at one example of poor small group decision making - the NASA team that decided not to pursue possible problems the Columbia shuttle could have had re-entering the atmosphere, after a debris strike at launch. A very detailed report was published into the subsequent disaster - the author concentrated on this example because this report was so detailed in the way it examined the performance of teams (see the Notes for references). The errors highlighted in the Columbia example are linked into research on the performance of small groups, and ways to improve small group performance are given. This chapter is relevant to people who work in (or run) small teams, and I found it impressive. It is also stated that few organisations have found out how to make small teams consistently effective, and hence for the group to be more than the sum of its parts.

Columbia disaster
Here is a quick summary of events and decisions as presented in the book:

• The Columbia launch was on 16 January 2003. A large piece of foam broke off an external fuel tank and hit the shuttle's left wing area. No clear picture of the impact or damage was taken.
• Members of the Debris Assessment Team (DAT) tried to evaluate the consequences of the debris strike.
• The DAT requested on-orbit pictures of the Columbia wing from the Mission Management Team (MTT), but this was refused.
• Prior to a meeting held by the MTT on 21 January, the DAT briefed Don McCormack, a member of MTT. They had reached no conclusions, but stressed there was reason for concern. At the meeting the MTT leader, Linda Ham, asked Don McCormack for an update on the foam strike two-thirds of the way through the meeting. He replied that it was being investigated and Linda Ham answered "And I really don't think there's much we can do so it's not really a factor during the flight." The comment made by the author is that she has already decided it's not an issue, and has decided for everyone else as well. The meeting moved on.
• Given the refusal to obtain on-orbit pictures, DAT estimated the size and speed of the foam and then used an algorithm to predict what damage this would do to the thermal protection files on the wing. The final conclusions they reached were that the shuttle was likely to be safe. This conclusion was qualified by saying their analysis was profoundly limited by their tools and lack of knowledge. This was presented to Don McCormack.
• At a meeting on 24 January McCormack summarised the DAT conclusions. In exchanges between McCormack and Linda Ham, the conclusion was emphasized ("no safety of flight issue") rather than the uncertainty in reaching it. Once the meeting ended, the author states, the fate of the Columbia was pretty much sealed.
• The author talks a lot about questions-with-implied-answers, unchallenged assumptions, and the lack of debate and minority opinions with reference to these events (more of this below).
• Columbia re-entered the atmosphere on 1 February, and was destroyed.

Columbia disaster: MTT team errors
Here is a potted list of the errors attributed to the MTT team, with relevant discussion from other research:

• The group didn't start with an open mind, but rather with assumptions that the foam strike wouldn't cause any damage.
• Confirmation bias - the initial assumption made the team more liable to unconsciously seek information that confirmed the assumption and discount evidence to the contrary (or not chase it up very thoroughly).
• The team thought they knew more than they did. For example, one of the reasons given for turning down the request for on-orbit photos was that the resolution wouldn't be good enough - it was shown later that none of them knew this and none of them went to the right people to ask about this. Also, one reason for not pursuing the whole issue was that nothing would be able to be done - again, NASA engineers later came up with two strategies to get the astronauts back (but not the shuttle; that was always doomed).
• Seeking consensus over dissent. At its most extreme this leads to groupthink, but at a lesser level it can accentuate certainty over doubt. An example is the way the DAT conclusion was emphasized, rather than focusing on the uncertainty in the analysis which (if pursued) would give more emphasis to dissenting views.
• Verdict-based, rather than evidence-based. The author bases this point on research into juries. Juries have been found to take two approaches - evidence-based, where no vote is taken until they've gone through a lot of discussion of the case; and verdict-based where a vote is taken before any discussion and the main point of the following discussion is getting the minority to agree (the mission here is to reach as quick and decisive a decision as possible). This is linked into the MTT group by the way questions were asked - they were asked with implied answers, having the effect of trying to get people to confirm the existing assumption (you could call these leading questions).
• The absence of debate and minority opinions in meetings. The author says that the meetings were both too structured (strictly based on the leader asking someone a question and them answering it) and not structured enough (no attempt to get other people to comment on particular questions - leading to a limited supply of analysis and information). It is stated that a consistent finding from small-group research is that group discussions are more effective when there is a clear agenda, and when leaders take an active role in allowing everyone to speak.
• Lack of diversity. NASA employees these days are much more likely to have a uniformity of background - science degree, graduate studies, join NASA - and hence are less likely to have divergent opinions. Research is quoted on mock juries showing that the presence of a minority view makes a group's decisions 'more nuanced and rigorous' even when the view itself turns out to be ill-conceived. The dissenting view forces the majority to consider their own views more seriously.
• Lack of aggregation - if a team are going to make decisions there needs to be a way of aggregating their collective wisdom. The MTT team didn't vote on anything.

Small groups: group polarisation and group dynamics
Group polarisation is a phenomenon where group discussion radicalises people's opinions. This occurs under certain conditions (which it is stated aren't entirely understood - this is an area of active research) and holds for both the group's opinion and the individuals within it. It seems that when people are alike they are likely to shift their view further in that direction - for example a group of risk takers find themselves advocating even more risky positions, whereas a group of risk averse people become even more cautious. Some reasons given for group polarisation are as follows:

• Social comparison: people note their relative position in a group and try to maintain that position - so if you are in the middle of the group and the group shifts, you will tend to shift with it.
• If the group already tends towards one position, then their will be more arguments in favour of that position, so the group will shift that way.
• People with strong views are more likely to have coherent arguments in favour of them and also more likely to voice them.

The discussion on group polarisation merged into one on group dynamics. The following comments are made on group dynamics:

• The order in which people speak in a discussion is important - earlier views are more influential and define the framework of the discussion. It's difficult for a dissenter to break down this framework. This is a problem, since people who speak first are often not the most knowledgeable (especially with a problem with no clear right answer) - the example is given of juries, where two-thirds of foremen (who lead the discussions) are men; do they know the most about guilt ?
• This leads to the similar point of talkativeness. People who talk a lot are thought of as more influential, and the more someone talks the more they are talked to by others in the group. And, again, there is not an obvious correlation between talkativeness and expertise - people who imagine themselves leaders can overestimate their knowledge and project an air of unjustified confidence.
• Deference is again linked to the above points. In groups where members know each other, higher status people will talk more and be deferred to. Research is quoted saying that this still holds even on subjects where there is no reason for the higher status people to have any more insight than the rest - higher status people would talk much more convincingly in defence of their position even when it was wrong and others were right.
• Ideas need champions in a small group. Having a strong champion for an idea is essential for group adoption ... and if champions are selected by status or talkativeness, the smartness of the group is lessened.
• Groups have a hard time incorporating new information. Research is quoted where all group members had two pieces of information, two group members had two more bits of information and one of those had two more pieces. In unstructured discussions the information talked about the most was the information all the group had; and the new information was often modified to suit the existing information (have you ever looked at a map with a pre-conceived idea of where you are and tried to fit your surroundings to the map, even when they obviously don't fit ?). And of course it is the new, rare (and perhaps unexpected) information that can often add the most value.

Research is quoted on depolarising groups by having an equal number of people with opposing views. It was found that the groups generally moved away from the extremes (note, though - just sometimes the extreme is correct), and that they were more accurate when tested on matters of fact - and generally the group outperformed its best member.

So, how do we sort these small groups out, you ask ? Group meetings should have a clear agenda and leaders should give everyone the chance to speak. Group members should have diversity of background. Minority and dissenting views should be welcomed, and the opportunity should be given to challenge assumptions. If possible, depolarise the groups by putting people with different approaches together. You are not looking for consensus, but the correct decision. Leaders can have strong opinions, but try not to imagine you know more than you do (check and see if those Columbia photos would actually give decent information). Some kind of aggregation is necessary to obtain group decisions.

Note that groups can still be used as above, even when they don't have a definitive problem to solve. Managers can treat groups just as a bunch of people to assign tasks to ... but what they should be doing is using the wisdom of the group to look at the way tasks are performed, the priorities of the group, the processes in place, etc. The group is smarter than the manager !

10. The Company: Meet the New Boss, Same as the Old Boss?

Summary:This chapter starts by looking at why companies exist and what their purpose is. There is a discussion on different company models - summarised as a top down corporate model, a small business type model with a group working together, and a group of people who come together for a particular project (e.g. independent film production). Problems in these models are given, respectively, as: information flow through the top down company is restricted; the small business is limited by its resources; the independent model costs money to set it up. There is no obvious correct model and some combination of these models is probably needed. A discussion takes place as to why all companies don't outsource everything (like the independent film makers) since then they will have access to the widest and most diverse set of people and information: the answer is (as hinted above) transaction costs - companies save by already having people, contracts and equipment in place (although sometimes outsourcing is still most efficient). An answer to why companies exist is given as to reduce the cost of getting people together to accomplish future goals. This says nothing about how companies operate. A history of company models through the 20th century in the US follows, showing that often lip service was paid to employee empowerment but it was frequently illusory. Much is made in the chapter about the fact that companies compete in the market place, but use non-market mechanisms (controls, plans) to run the company. The conclusion is that the more market-like the company is run - local decision making, corporate decisions helped by aggregating employee knowledge, one man (the CEO) not making major company decisions in isolation - then, generally, the better. The caveat to this is that the company has a common goal - creating artificial internal competition is not good. Diversity ,decentralisation and aggregation are key here. I'll give a quick bulleted summary of key points (because I'm flagging) under two headings showing good and bad corporate practice.

Poor corporate practice

• Mistaking consensus for collective decision making. The search for consensus leads to 'tepid, lowest-common-denominator solutions which offend no one'. An example is given (applying to many US companies in the 60s and 70s), where to make the decision making process inclusive, issues were discussed at the lowest levels of the company and then gradually moved up the managerial chain (up to 15 levels, compared to 5 in Toyota). At each level the decisions became more watered down and the top executives were even more isolated from the workers. The point is also made that people lower down passed on their responsibilities when they gave advice to their managers; the manager was now in charge of presenting the advice and so power was not really delegated, leaving less incentive for initiatives from the ground.
• Centralisation, which discourages the free flow of information through the company. A large number of bosses and information-flow-paths are going to lose and distort ideas. Problems are handed up the chain and decisions handed back down (maybe) causing a lack of use of local knowledge and initiative.
• Incentives being skewed against dissent and independent analysis. Research is quoted showing that to move up the executive chain it helps if you don't report problems.
• Incentives to hide information or lie. This refers to incentive systems being based against expected budget and performance targets, leading people to manipulate expected targets - that is, their aim is not to be as productive as possible but to try and make targets low to maximise their rate.
• Hostility by bosses of opposition from subordinates. The 'illusion of perfectibility' is placed on the decision makers and people are encouraged to play along; this is exacerbated by a natural inclination among employees to avoid conflict.
• Creating (deliberately or not) internal rivalry and competition between divisions. Different company divisions can have an incentive to look for more resources than they deserve to the detriment of the company. Where divisions are truly separate you can have major inefficiencies and blockages to information - e.g. separate IT systems ! You should be able to trust your fellow workers more than workers from another company - Enron is quoted as the classic case of where this wasn't true.
• Lack of diversity in managers, in both mindset and otherwise.
• Research shows stock options are good for a company (they increase a sense of ownership) ... but only if they are widely distributed, and not just to the execs.
• CEOs wielding power way above their ability (which means making all the ultimate decisions themselves). The author is good on the failures of CEOs - as stated earlier he was the editor of Best Business Crime Stories, a collection of business articles chronicling the fall from grace of various CEOs. His thesis is not that CEOs are not very smart, but that they are endowed by hype (and probably salaries) that are not justified. A poll said that 95% of investors would invest in a company based on their opinion of the CEO. Yet this is not borne out. Stories are given of CEOs who achieved outstanding performance at one company, yet fail completely at another. It is often difficult to measure CEO performance, as to directly relate a CEO decision to company performance is not trivial - one example, though, quotes research showing 80% of new products released (presumably signed off by CEOs) do not last 12 months; another shows corporate profit margins reducing over the 90s, as executive pay was soaring. As stated before 'chasing the one expert' is not correct - at any one time track records of particular CEOs may be superb, but this could be down to being in the right company at the right time (i.e. some luck is involved here, and the CEO is often not massively influential).

Good corporate practice

• Decentralisation and local decision making. Research is quoted showing that the more responsibility people have over their environment the more engaged they are and the better their performance. A second argument is that people will coordinate themselves with their peers when they are empowered to do so, saving on supervision and controls. An example of one way of doing this is given, whereby teams of workers are put in charge of their own production process - as opposed to each being told to work on one specialism. They may have a specialism, but to at least be aware of what those around them are doing helps generate more ideas and information, and to spot potential problems earlier.
• An inter-disciplinary approach to problems (across divisional boundaries).
• Less rigid boundaries between bosses and subordinates; the merit of an idea not determined by status of the proposer.
• Diversity.
• Group management by CEOs- i.e. CEOs building a strong managerial team around them and using them. The board of directors shouldn't defer decisions to the CEO.
• Internal decision markets, as one of the best methods of aggregating employee knowledge. A favourite of the author; and you can see it frustrates him that they're hardly used despite strong performance when they have been used, and in research. Positive examples are given, including one where employees at a drugs company were used to accurately predict which drug candidates would be accepted by the Federal Drug Agency (which should be hugely profitable in determining which drugs to pursue). It is stated that decision markets should inform major corporate decisions, not make them. Have a quick search on the net if you want a choice of software to set up decision markets. Some of these are right along the lines in this book. An example is inkling at http://www.inklingmarkets.com. They describe a prediction market as "Prediction markets allow a group of people to express an opinion over a period of time about the probability of an event occurring. A question is posed and people buy and sell shares in stocks representing possible answers to that question. The highest priced stock at the end of a period of time is the group's prediction." Software can be provided for enterprise markets, small business markets, personal markets or for academic research. It even lets you set up a free trial in your company - so there's no excuse.

11. Markets: Beauty Contests, Bowling Alleys, and Stock Prices

Summary: This chapter is largely about the stock market, which you might think of as a prime example of a large number of people making independent decisions, arriving at prices for numerous stock. And the fact that individuals find it very hard to beat the market testifies to the quality of the decision making. However, the market goes wrong - bubbles and crashes happen. The chapter looks at why this happens, the basic conclusion being that the market is skewed towards dependent decision making at these points. The bowling alleys of the title are an example of a stock market bubble in all things bowling alley-related in '50s America - bowling alleys proliferated, everyone thought it would last for ever, stock prices rocketed, companies built loads of alleys (based on optimism rather than intelligent analysis), and eventually, of course, it crashed. Think tech stocks of the '90s. The beauty contest of the title refers to a quote from John Maynard Keynes, where he likened investment to newspaper competitions where you had to pick out the six prettiest faces from 100 pictures. The prize goes to whoever most closely matches the average choice of all competitors. So you are thinking about who everyone else thinks has the prettiest face, rather than who you think has the prettiest face. And this, to an extent, is dependent decision making rather than independent decision making. Along the way we also briefly look at what tends crowds to riot and parallel that with stock market bubbles. As in the last chapter, I'm going to pick out some key points and list them under bullet points.

Key points

• Investors are not rational, believe it or not. Examples given from research are that investors pay too much attention to recent news, they hold onto losing stocks too long, they are over confident, they trade more than they should, ... - and this applies to professional investors as well. This doesn't mean that the overall result is not the best solution, the biases may cancel out - the problem is when biases are systematic (that is, there is a skewed irrationality). One example given is that people are (psychologically) very reluctant to short stocks - this is reinforced since it has often been frowned on (and even illegal) on the basis that people are profiting from misfortune. The point made is that the aim is to get the correct price, not make the price go up artificially, and a better way to do that is if the investors are really diverse and independent. Prices historically go up, so this may not have skewed the market too much, but the author hypothesizes that when stocks start to be misvalued the lack of shorting helps to exacerbate this - he says tech stocks were almost impossible to short in the '90s bubble.
• Here's an example of irrationality, solved by a psychological trick. The irrationality is that people want to save for their future, but don't. They constantly think they will save next month, next year, etc, but they just sign up to put the minimum into their pension scheme (if at all). The trick is to allow them to choose, at the beginning, different rates for future months - so they sign up for a small amount in the first couple of months, with larger amounts in the future. By the time they get to the future, they don't make the effort to change it. This is a similar ploy (just slightly more subtle) than getting people to opt out, rather than opt in. If people have to opt out of, e.g. being an organ donor, rather than opt in, there will be a lot more organ donors. So, the point is ... if there is a systematic irrationality for psychological reasons, you may be able to skew it back by changing the structure.
• What do we actually mean by getting the stock price right ? The share price is simply the verdict of all the buyers and sellers of that stock. The author talks about the true value of a stock. What he means is something like the following: The total value of stock (the share price times the number of shares) should be able to be recouped by the investors in a certain time period if all future earning are paid out to the shareholders. He states this period as 20 years. The actual time period will depend on the price to earnings ratio [annual P/E ratio = share price / earnings per share] - so if you have a P/E of 20, you are earning a twentieth of the stock value per year, and hence the company could pay the stock back in 20 years. The industry suggests that 20 (well, 15 - 20 is what I read) is about a fair value for the P/E which is why the author chooses 20 years. If the P/E is high, maybe that is because the investors think the company will expand in the future to still make that 20 year payback; and if it is low, maybe they are uncertain about the long term viability. Anyway, under this analysis, the stock price is correct if the total value equates to the company's future earnings over the next 20 years. Which is pretty tough to predict. The author suggests that the market may not get this right, but that it will get it more correct than any other mechanism.
• One reason why the market is difficult to predict is that there is no point at which the 'correct' answer is revealed. This can encourage speculators, who don't care about the correct price, as long as they can sell for a profit. They can make money when they're wrong. So, if you're dead right about the market, you can still lose money - again we have a quote from John Maynard Keynes, 'the markets can stay wrong longer than you can stay solvent.' We're into bubble territory here, which leads us on to our next point.
• A bubble is where price and value lose all connection; the same holds for a crash although it's more abrupt, with everyone selling. Collective decision making is not working in these cases (or at least, not giving the 'correct' share prices) - you could say that the crash is a necessary correction, but it is an over-correction. The author makes the point that getting the stock prices right is good for the overall economy. The argument goes something like this: the stock market helps determines how much capital is available to firms by setting share prices (firms can raise money based on their share price, by selling stock or issuing bonds); this means the stock market effectively allocates more capital to some firms than others; you would expect those companies with high prices to use the money wisely if they deserve the high share price (so that they can make good on those future earnings people are predicting); if they don't, then the economy will not grow as much as it could. Note that bubbles and crashes are not seen in the real economy (TVs, food, etc.). The reason is that in the real economy people are mainly concerned about the value of an item to them; whereas in an asset market people are concerned about the resell price, which they hope will go up. So ... people buying shares are not only concerned about the value of the share to them, but about what everyone else thinks the value will be. This is a key point, leading us onto dependent decision making.
• A classroom experiment in share dealing is described, where a bubble was created. The value of the shares was clearly understood, yet they shot up. The students in the experiment explained that they knew the price was over-valued, but they thought they could make a profit by selling on to 'the greater fool'. The author suggests that in real bubbles, not everyone knows they are in a bubble and people start to believe the hype - you see newspaper articles explaining and justifying why the high prices are 'a new paradigm', etc. Some people buy shares based on what everyone else is buying - this is dependent decision making, and is not necessarily bad; if we trust the wisdom of crowds, then it is wise to follow the market decisions. The author says that a healthy market has a mix of these dependent decisions and independent decisions (where people with various viewpoints and backgrounds are analysing the company value). And bubbles arise where the market is too skewed in the direction of dependent decision making - that is, everyone is copying the group. The point is made that there is originally some justification for higher prices (the Internet did offer new and impressive business opportunities) - the problem comes when they take off and everyone follows.
• This point is about information, how it is presented and how this can help create bubbles. The example is given of CNBC (a US market news channel) which was watched by millions during the '90s tech bubble. Markets prices reacted instantly to any stock featured on the show (within 15 seconds). So people were reacting just because a stock was on the show, on the basis that other people would react; the show also bombarded people with opinions from traders. If this news (watched by millions) is saying 'biotech stocks are rising' or 'panic in tech stocks', then this will exacerbate herding in investors. People's decisions become dependent, and they are reacting to information that has its importance over-played. An experiment is described showing that more information doesn't ensure better decisions, if the way it is presented over-plays its importance. The number of jelly beans in a jar experiment was run, the classroom giving a very accurate average. The experiment was then repeated, but they were told to think about the fact that there was air space at the top of the jar and that it was made of plastic not glass - the average guess was then 15% out. The information was true, but it suggested there were more beans than they thought and gave a common skew to the group. This is a good point - it means that if you want a better decision you need to present facts neutrally without hype or comment; it also means that if you want a particular group decision, you can skew it by the way you ask the question and present related information (think politics).
• Finally, a quick diversion on rioting crowds. In a mob, some people will never riot, some (instigators) are ready to riot at the slightest provocation, and many people are somewhere in the middle (their willingness to riot depends on what other people are doing). So, to get a riot ... you need instigators and a large number of people who can be swayed - once a crowd does riot its behaviour is shaped by its most violent members. Hence the more independent people (not following the crowd), the less likely a riot. [This also, I think, links to a point made earlier saying that if people firmly dissent from authority (in this case what the instigators are saying), then other people are less likely to go along with it]. Anyway, the rioting crowd is paralleled with a stock bubble - the more independent people, the less likely a bubble and the more intelligent the market.

12. Democracy: Dreams of the Common Good

Alternative comment: This chapter's about democracy ... and I'll let you read it yourself. I wouldn't like to be sued for giving a complete summary of the whole book. I'll add my own very brief comments on this topic, though. Firstly democracy works. It gives a much better solution than any other - would you rather be in North Korea, Hitler's Germany, a religious theocracy, ... ? Secondly, although it may give better solutions, it often seems obvious it is not giving optimal solutions. Sticking with the UK, you would be hard pushed to argue that centuries of democracy have produced the best health and education systems, the best criminal justice policy, the most intelligent defence procurement, and let's not even mention the EU (and the Common Agricultural and Fisheries Policies). You could make the argument that one reason why results are not more optimal is because politicians don't properly tap into the wisdom of their population (either deliberately or subconsciously).

One reason for this, I think, is that politicians form an elite. They are not diverse and the more established parties are much closer to each other than they appear - they deliberately emphasize small differences to make it look like they are more different than they are. This is like the risk averse fund managers who all follow the same strategy, even when they think another approach is correct. It is, for example, the small parties, think tanks or radio show callers who propose more radical ideas, which are ignored. Staying on this point, and hypothesizing freely, maybe it is the more mature democracies that get like this as the elite political class beds in - like the markets, you would expect a correction eventually, although maybe politics can stay wrong (divorced from the population's opinions) for a long time. My second and final point (for anyone who's got this far) is that politicians and governments are masters at giving selective, slanted information; they are not trying to get wise answers from the crowd in elections or referendums (and wouldn't expect to the way information is presented) - they are trying to get elected. Anyway, there is room for thought, disagreement, and further ideas here. I leave you to it.

Conclusions

Despite my glowing review, is he correct ? After all, by his own argument, you shouldn't trust one expert (the author) alone; he probably used confirmation bias in the research he quoted to back his thesis - although he did put out counter opinions, and offered intelligent counters against these. Well, I'm only one person as well, but I can offer the following examples where I have found his thesis to give good results:

• Whenever I have two spelling possibilities for a word, I use Google to look up each version. For example is it traveled (101 million matches) or travelled (40 million matches) ? The answer is therefore traveled, and this pretty much always works. [Actually in this case, both are acceptable, but let's not get picky].
• I seem to get sports injuries far too often, and they are not always easy to diagnose or treat. Any of the following may be needed: rest, physio exercises and/or stretches, massage, painkillers, cortisone injections, orthotics (for your shoes), heel raises (for leg length discrepancy), osteopathy (manipulation), ultrasound, X-rays, MRI scans, surgery. And you need to do two things - get rid of the symptoms and eliminate the root cause. Different people have specific expertise in each of those categories, and even those in the same field have different approaches, skill levels and opinions. My approach is to see a number of people if I can't get an injury cleared up quickly. And when I see someone new, I explain the treatment and opinions I have already received - I may have largely accepted an opinion already and just be trying to get it actioned, I may be seeking further information like the root cause, or I may still be trying to get to the bottom of the problem. Effectively I am using a team to get to the best solution - my aim is to put as much independent information in front of each professional and eventually it all gets put together and we come to the solution. Some professionals embrace this and say it helps them, and some hate it. The point is, it gets me to the solution - it can also help the professionals to see opinions from other fields (honestly). In contrast to this, the following is, sadly, common: someone sees one person for months or even years and feel a bit better after each treatment but soon go back to the same state - e.g. they are seeing an osteopath for back pain, when they really need to be doing physio exercises or getting a scan or a second and third opinion, etc. It is also common for people to gradually give up activities when they are told to by their doctor or the first person they see - when the answer and treatment is actually out there.
• In my work (as a project manager) I make an effort to pass information, decisions and queries to a large number of people - whether by email, webpages, telephone, meetings or whatever. Some of these people will be working on specific areas of a project or related function and in theory don't need to know a lot of this. I try and tread a fine line between highlighting information and not making people read a whole bunch of stuff (which they won't anyway). Anyway, the point is, I am looking for contradictions - person A thinking they are doing one thing and B assuming A is doing something slightly different, or person C thinking D is doing something and D thinking E is doing it, etc. It is surprising how often contradictions turn up quickly - and when you find them early they are not a problem. You also get other useful opinions and information turning up. This approach, I find, far outweighs the cost of people accessing information that they don't strictly need.

Finally (honest), given that I think this is an excellent book and I believe it is broadly correct, has anyone taken any notice ? It has had great reviews , so you'd think people may try and apply it. I would say, on a large scale, no. Masses of companies and teams and governments haven't suddenly started tapping into group wisdom any more. If you Google 'decision markets' you only get 17,000 hits. But there are shoots out there. There are companies selling decision market software and they seem to be achieving steady sales to companies, and there are positive articles and examples on the Internet. So let me recommend you give it a go in your job or with problems generally, or that you recommend this approach to your company. Don't overdo it in your personal life, though - I'm sure you can think of questions it's inappropriate to poll your mates on. Then again, could be fun ...