By James Surowiecki
Key Points
The
thesis of the book, as the author states at the beginning:
Under
the right circumstances, groups are remarkably intelligent, and are often
smarter than the smartest people in them. Groups do not need to be dominated by
exceptionally intelligent people in order to be smart.
SUMMARY
Chapter 1
The
book begins with a series of examples of “crowd wisdom”, ranging from the TV
show “Who Wants to be a Millionaire?” and its “ask the audience”, to the stock
market indicating the company most likely to be at fault hours after the
Challenger disaster. These cases all demonstrate the four conditions that
comprise wise crowds - independence, diversity of opinion, decentralization,
and a way to aggregate the results. Similar results are to be found in sports
betting, and in Google’s results, determined by examining the number of links
pointing to any given page. One way to take advantage of this wisdom of crowds
is through the use of ”prediction
markets”, such as the Iowa Electronic
Markets, where people buy and sell probabilities as if they were
stocks. In the right circumstances, prediction markets are an excellent way of
turning the knowledge of many people into reasonably accurate predictions.
Chapter 2
The
importance of diversity is covered in the second chapter. A crowd can’t be wise
if everyone always picks the same answer as everyone else. Examples include
product markets, where there is usually an initial wide range of different
attempts in a new market, which is quickly winnowed down to the successful
designs; and honey bees, which send out scouts in all directions, but only
return to those areas where flowers have been found. Diversity is important to
“wise crowds”, because it expands the range of possible solutions proposed. In
large groups, diversity comes naturally, but in smaller groups, it’s necessary
to support and actively encourage it, to avoid the dangers of “groupthink”.
When people give in to their conformist tendencies, and are afraid to stick
their necks out, the quality of decisions suffers.
Chapter 3
Independence
of action and thought is important for the wisdom of crowds. If everyone thinks
alike, then they’re less likely to arrive at a good answer to a given problem,
because they’re less likely to fall into “groupthink”. “The more influence we
exert on each other, the more likely it is that we will believe the same things
and make the same mistakes”.
American
Football coaching is cited an example of the “herd mentality”, based on the
work of David Romer examining the “best 4th down
strategy” (pdf). It turns out that statistically, most teams
would be better off trying to make the touchdown or 1st down, rather than going
for the field goal, in many cases. However, since the accepted wisdom is to
kick, going against the grain of the relatively small pool of decision makers
(professional football coaches) would not be an easy choice to make
consistently, especially for the risk averse.
Herding
behaviour often occurs because people seek safety in numbers, but it can lead
to problematic results when independence is required. “Information cascades”
are what occurs when an initial decision is made by a few people, and then more
or less accepted uncritically by more and more people. This isn’t necessarily a
recipe for disaster, as we can’t all evaluate everything in our lives, but must
trust others to come to good conclusions. However, at times, it can be
disastrous when the original information and decisions were wrong, but continue
to be accepted by an ever-wider circle. Luckily, for most people, the more
important a decision is, the more likely they are to examine the facts
themselves, rather than simply fall in line. Information cascades actually work
reasonably well much of the time, but the basic problem is that they are a
sequential, rather than parallel process. If you’re trying to harness the
wisdom of crowds, you must attempt to have all decisions made at the same time,
rather than one at a time.
Chapter 4
This
chapter covers decentralization - where it works, where it doesn’t and what can
go wrong. Decentralized, aggregate behaviour is a key aspect of things like
free market economies, flocks of birds, and is something that has been touted
as a virtuous way of running a company as of late, with small, self-organizing
teams. Decentralization allows people, or more generally, components of a
system, to act freely and independently of one another, and still interact to
produce coordinated results.
Linux
is cited as an example of a decentralized system with a central aggregator -
Linus Torvalds. As most people know by this point, Linux is worked on
collaboratively by many programmers throughout the world, but often, different
people come up with competing solutions to the same problem. This is good at
finding and testing diverse approaches to see, in practice rather than in
theory, which one actually works the best. Ultimately, however, the ‘best’
solutions are not selected by popular vote, but by Linus, who is responsible
for taking the results of the decentralized development process, and
aggregating them into something useful by selecting the ‘best’ bits and pieces.
Also
discussed is the decentralization of the intelligence community, and the
negatives involved in the difficulty of sharing information, cited as one
factor in the failure of the intelligence community to predict and prevent the
9/11 attacks. The problem, however, was not decentralization, but
decentralization with no way to aggregate the results into something useful.
One
such way of aggregating information was a proposed futures market based on
potential events in the Middle East, and elsewhere, which was, however, not
allowed to get off the ground due to squeamishness about the idea of buying and
selling bets about, say, a leader’s chance of being assassinated in any
particular year. This market could have been a useful tool, perhaps not in
predicting precise events, but in collecting information about the general
state of things in places where information is at times difficult to gather,
and unfettered freedom of expression suppressed.
Chapter 5
This
chapter covers what are known as “coordination problems”, which are defined as
problems that don’t necessarily have an objectively “correct” answer, but which
are framed in terms of coordinating actions with everyone else’s actions. For
instance, driving on a freeway requires that you coordinate your speed and
actions with those of other drivers, and possibly even the time of day when you
drive in order to avoid getting stuck in traffic. Groups are not guaranteed to
come up with optimal solutions, but often do.
One
solution to coordination problems is central planning - having one omniscient
authority that makes some calculations and tells everyone how to act as a
consequence. This is, however, often not possible, feasible, or desirable.
Coordination
problems are often quite difficult to solve, with one example being a bar,
where, if it’s more than 60% full, no one enjoys themselves, but do if it’s
under that capacity. Several computer models have been built with agents that
follow simple strategies and do manage to coordinate well enough to keep the
bar at around 60%.
In
some cases, cultural references help us solve coordination problems, both by
giving us reference points (ask two people to meet at a given time without
communicating the time to one another, and they’ll likely pick 12 noon), or
norms, such as “drive on the right”. Conventions also lower the amount of
thinking you have to do about certain situations - it’s easier just to follow
the rules or guidelines rather than make a conscious decision after weighing
all the possibilities. This often frees us to think about more important
things.
Corporations
are supposed to operate in order to maximize profits, and should be immune to
things like social conventions - yet it turns out that they’re not nearly as
rational as might be imagined. One example cited is movie theatres, which
charge the same price for the latest hit, as for flops that are on their way
out. Charge too much for hits, and you risk losing out on concessions, where
movie theatres actually make a lot of their money, but by that logic, lowering
the price for less popular movies would get more people into the theatre.
Markets
can also be effective coordination mechanisms. Experiments conducted with
students, who know only the maximum price they will pay, or minimum they will
sell for, show prices rapidly converging on an optimal price, even though that
price is higher than buyers would like, and lower than what sellers would
prefer. Real markets often lack lots of information, and indeed the students
found the experiment “chaotic and confusing” - and yet, the market worked.
Markets aren’t perfect, of course, but they are often the best, if not perfect,
way of coordinating disparate buyers and sellers.
Chapter 6
Cooperation
problems are superficially similar to coordination problems, but with a key
difference: coordination problems can be solved with all players acting in
their own interests, whereas cooperation problems require players to “look at
the bigger picture”, as part of an organization or society.
Behavioural
studies have demonstrated that people will forego a reward in a simple game in
order punish someone perceived to be playing unfairly, even when doing so does
not benefit them at all. In other words, people, being social animals have a
sense of ‘fairness’, even if this isn’t rational in economic terms. This
extends to a sense that rewards should be correlated to efforts and
accomplishments, and this sense is part of the reason why large organizations
can exist in the first place.
Trust
is often secondary to long term relationships in terms of promoting ‘fair’ behaviour:
if you know you’ll see someone again and again, you’re less likely to attempt
to cheat them.
Capitalism
works in part because it’s possible to trust those beyond an established circle
of friends and family, and only works where there are institutions that promote
this trust. When you are reasonably certain that you can buy a product and that
it will work as advertised, you don’t need to inspect in detail each and every thing
that you purchase. This makes the flow of goods and services, and increases the
general welfare of a society.
Chapter 7
This
chapter discusses the idea of ‘coordination problems’, using traffic as an
example, beginning with a discussion of London’s “congestion pricing”. Because
traffic was so bad, a market-based solution was found that pushed people to
evaluate their access of downtown London via a car: during the day, it costs a
certain amount of money to drive into central London. This accomplishes two
things: rather than dictating to drivers what they can and cannot do, it leaves
everyone free to do as they so choose, but puts a direct cost on accessing the
downtown area during certain hours. People who really do need to go there at
that time will pay the money, but find the roads less crowded. Other people,
without such strong necessities, will take the time to walk, cycle or use
public transportation. London is hardly alone in using such a system; Singapore
has used congestion charges since the 1970ies, although clearly implementing
that kind of unpopular policy is easy in an authoritarian country.
The
discussion continues, touching on the subject of traffic flow, and the ideal
conditions that produce, a smooth, steady flow, rather than traffic jams, or
erratic start and stop conditions. Surprisingly, having just the right amount
of cars on the freeway is important: too many create obvious problems, but too
few cause problems as well; with too few cars, people tend to speed up and slow
down more erratically than with a steady stream of traffic.
