Here are my notes on The Wisdom of Crowds by James Surowiecki
Under the right circumstances, groups are smarter than the
smartest people in them, even if the group doesn’t contain expert members.
Experts are as likely to disagree as agree. Experts’
individual consistency is also 0.5. Experts overestimate the likelihood that
they are correct – little correlation between self-assessment and performance.
Therefore, however well informed and sophisticated and expert is, his advice
should be pooled with that of others and the larger the group the better. So we
should stop “chasing the expert” and ask the crowd. We continue to chase the
expert because we assume average means least common denominator (Larrick and
Soll) and we are fooled by randomness (Taleb).
Criteria for good
collective decision making
The group must be big enough and diverse enough, the members
must be forming opinions independently, and the group must be decentralized.
To solve cognition problems you must 1) uncover alternatives
and 2) decide among them. Diversity is needed for both 1 and 2. Diversity adds
perspectives and weakens destructive characteristics of group decision making. A
successful system recognizes losers and kills them quickly. Homogenous groups
spend too much time exploiting and not enough time exploring (James March). Homogenous
groups are susceptible to groupthink, willing to rationalize away
counterarguments and often convinced that dissent is bad. Diversity is more
important than individual intelligence but members must be somewhat informed. There
has to be some information – can’t have a completely ignorant group.
means members can’t be dependent upon one another for information and can’t be
subject to influence from one another.
keeps mistakes from being correlated. Members can be biased/irrational without
making the group dumber. The more influence members exert on each other, the
more personal contact, the dumber the group will be. This is difficult to
want to learn from one another
are affected by environment/neighborhood/hierarchical position
become more influential as they get bigger
better for reputation to fail conventionally than to succeed
cascades can occur when members make decisions in sequence rather than
simultaneously. That is, the first deciders influence the subsequent even
if they are wrong. If the subsequent start following the crowd then the
cascade stops informing. (From The
Tipping Point, cascades move via social ties – mavens, connectors,
There also must be intelligent imitation not slavish.
people stop imitating and learn for themselves when the benefits of doing
so become high enough
people just keep imitating no matter what
To get intelligent imitation
needs to be initially a wide array of options and information
members must value their own judgment ahead of group’s – overconfident
people who go with their gut or systematically test and adopt
Corporations should incentivize employees to uncover and act
on private information. (Blasi and Kruse, High Performance Work Systems).
Decentralization- specialization plus coordination. The best
collective decisions are the product of disagreement and contest, not consensus
or compromise. For a decentralized system to be intelligent there must be a
means of aggregating all members’ inputs such as market prices or centralized
decision makers, e.g., Linux. The risk of decentralization is that information
won’t make it through the system to where it is most valuable. There must be a
balance that allows individual knowledge to be specific and local (tacit) but
also makes it globally useful.
People focus better on a decision when there are financial
rewards attached to it so decision markets are often successful.
Problems can be classified as
when there are definitive solutions (Who will win the Super Bowl?) or
there is a best possible answer (Where’s the best place to site this
when members’ behavior must be coordinated (driving or finding a party)
getting members to work together when self interest would dictate that
they should not (taxes, dealing with pollution)
For coordination problems, independence is pointless since
what one member is willing to do depends on what that member thinks other
members are going to do. The El Farol problem shows that even in this case
collective judgment can be good though it can result in many members not being
satisfied. With traffic jams the diversity of drivers makes coordination
difficult. Solution is more control: automatic highways with platoons of
synchronized cars or driver assistance to keep cars evenly spaced.
conflict, e.g., “first come first served,” queues – there is wisdom in
conventions but many conventions can be very stupid. For example B movies and
old movies cost the same as new. This convention is uncoordinated with
To solve cooperation problems, members need to adopt a
broader definition of self-interest than maximizing short-term profits and they
need to trust other members. There also needs to be a mechanism for preventing
free riders since many people are conditional consenters – only cooperating
because they believe that people who don’t will be punished. People exhibit
strong reciprocity: willingness to punish bad behavior even when they get no
material benefit. The evolution of capitalism has been toward more trust and
transparency because the benefits of trust are immense.
still prefer to work in proximity to colleagues but researchers who spend more
time collaborating internationally are more productive.
want recognition more than cash. We trust that allowing scientists to pursue
self-interest yields better results than command and control.
blend of collaboration and competition works because of open access to
flaw is that most scientific work never gets noticed because famous authors get
Rules for Small
Small groups can be good because they make people work
harder and think smarter. Non-polarized small groups make better decisions than
individuals. But small groups face many problems so there need to be rules for
good decision making
must have structure (ask each member for input) but not too much (one leader
doing all the asking)
making must not begin with a conclusion. This makes it unlikely for new info to
be incorporated. Don’t spend all the time talking about what everyone
advocates must be encouraged
polarize through discussion (counter to common wisdom) because members try to
maintain their place in the idea spectrum relative to the entire group. So if
entire spectrum shifts right, member must shift right just to stay in the
middle. To avoid polarization make sure the group has equal number of people
with strongly opposed views.
order of speakers matters – earlier comments are more influential. So don’t
choose earliest speakers on basis of status since that may not equate to more
knowledgeable. The same applies to group members that talk the most.