by Hugo Mercier
62 ch ap t er 4
checking, when it happens to work with poor material. When
more accurate views—in evolution by natu ral se lection, in the
efficacy of vaccines, and so forth— spread, it is in part thanks to
argumentation, but argumentation is most power ful among
people who can discuss issues at length and share a lot of com-
mon ground. For these sound beliefs to expend outside of a circle
of experts, we must be able to recognize that sometimes others
know better.
5
WHO KNOWS BEST?
on january 5, 2013, Sabine Moreau, resident of the small Bel-
gian town of Erquelinnes, was supposed to pick up a friend at the
train station in Brussels, fifty miles away. She punched the ad-
dress in her satnav and started driving. Two days and eight
hundred miles later she had reached Zagreb, on the other side
of Eu rope, having crossed three countries on the way. That was
when she de cided that something must be wrong, made a U- turn,
and found her way back to Erquelinnes.1
As I argued in the last chapter, we put more weight on our own
beliefs than on communicated information— when every thing
else is equal. Every thing else often isn’t equal. Others can be
ignorant, mistaken, or poorly informed, giving us reasons to
discount their opinions. But others can also be more competent
and better informed. A major part of the open- mindedness of
open vigilance mechanisms comes from being able to identify,
and then listen to, those people who know better, overcoming
our initial reaction driven by plausibility checking to reject infor-
mation that conflicts with our prior beliefs.
In this chapter, I explore a variety of cues that help us identify
who knows best: Who has had the best access to information?
Who has a history of getting things right? Whose opinion is
shared by the most people?
63
64 ch ap t er 5
These cues told Sabine Moreau she should believe her satnav.
The satnav has access to precise maps, had proven reliable over
many past trips, and is widely believed to be accurate. Obviously,
she went too far in letting these cues trump her intuitions. But for
one Sabine Moreau, how many people end up lost or stuck in traf-
fic jams because they didn’t follow their satnav’s suggestions?
Eyewitness Advantage
The most obvious cue that someone else is more likely than us
to be right is access to a sound source of information. You be-
lieve that your friend Paula is not pregnant. Bil , who you know
has just seen Paula, tel s you not only that she is but also that she
is quite far along in her pregnancy. Assuming you have no rea-
son to think Bill is lying to you (a question we turn to in the next
chapter), you should change your mind and accept that Paula is
pregnant. Testimony from the right source can also amount to
privileged access—if you know Bill has just called Paula, you
should also believe him when he tel s you she is pregnant.
Intuitions about the value of informational access develop
very early. In a classic study, psychologist Elizabeth Robinson and
her colleagues systematically varied the information
children— some as young as three— had about what was in a
box.2 Some children saw what was in the box; others simply
guessed. The children were then confronted by an adult who told
them that what was in the box was diff er ent from what the
children had just said. Like the children, some of the adults had
seen what was in the box, and others had just guessed. The
children were most likely to believe the adult when she had seen
what was in the box, while they had only guessed; the children
were least likely to believe the adult when she had guessed what
was in the box, while they had seen it.
w h o k n o w s b e s t ? 65
If we do not already know what information our interlocutors
have had access to, they often tell us. If Bill knows you think Paula
isn’t pregnant, he might preempt your doubts by saying, “Paula
just told me she’s pregnant.” Such information about the source
of our beliefs is ubiquitous in conversation. Even when none is
mentioned explic itly, some can usually be inferred. If Bill tel s
you “That movie is great!,” this suggests he saw the movie rather
than, say, read some reviews.
Again, even young children are sensitive to such reported in-
formational access. In a series of experiments, my colleagues
and I asked preschoolers to help a (toy) girl find her lost (toy)
dog. A (toy) woman suggested the dog had gone in one direc-
tion, saying that she had seen it go that way. Another (toy)
woman pointed to a diff er ent direction, without specifying why
she believed this to be the right place to look for the dog. The
children were more likely to believe the woman mentioning
a reliable access to information, and the same was true when
the second woman provided a bad reason rather than no rea-
son at all.3
Reliable Expertise
When a friend offers a way of fixing our computer prob lems, rec-
ommends a restaurant, or provides dating advice, it is not
enough to know where she got her ideas from. Maybe she had
firsthand experience of the restaurant, but the value of that
experience depends on her taste in food—if she can’t tell a
McDonald’s from a Michelin-starred restaurant, her firsthand
experience isn’t worth much. How can we figure out who is
competent in what domain?
The most reliable cue is past per for mance. If someone
has been consistently able to solve computer prob lems, pick
66 ch ap t er 5
exquisite restaurants, or give sound dating advice, then it is
prob ably worth listening to what they’re saying in each of these
domains.
From an evolutionary point of view, what makes past per for-
mance a great cue is that it is hard or impossible to fake. It is
difficult to consistently solve computer prob lems, find exquisite
restaurants, and give sound dating advice if you don’t possess
some under lying skill or knowledge that allows you to keep per-
forming well in these areas.
To evaluate others’ per for mance, we can rely on a wide vari-
ety of cognitive devices. Humans are equipped with mechanisms
for understanding what other people want, believe, and intend.
Thanks to these mind- reading mechanisms, we can understand,
say, that our friend wants her computer to work again. All we have
to do, then, is keep track of whether she successfully reaches
her goal.
We can also rely on the mechanisms described in the last chap-
ter: plausibility checking and reasoning. Someone who gives
you the right answer to an insight prob lem (like the one with the
triplets) or who offers a novel and convincing mathematical dem-
onstration should be deemed more competent, at least in these
domains.4
Seeing someone, from a professional athlete to a craftsperson,
&nbs
p; do something well can be very pleas ur able, giving rise to so- cal ed
competence porn— for example, the consistently articulate and
witty exchanges of screenwriter Aaron Sorkin’s protagonists. The
plea sure we derive from watching someone perform flawlessly
actions that do not benefit us directly is likely related to the
learning possibilities afforded.
Having recognized who is good in what area, one possibility
is to imitate them. Some nonhuman animals, such as the house
w h o k n o w s b e s t ? 67
mouse, are already selective in who they imitate, being more
likely to copy the actions of adults than of juveniles.5 But imita-
tion has its limits. Copying what your friend does when she
fixes her computer prob lems is unlikely to help fix your own
prob lem. Following your foodie friend around might lead you
to some places that are not really to your taste, and to a depleted
bank account. That’s when communication comes in handy.
Once you have inferred, through past per for mance, that a given
friend is good with computers, you can ask them about your spe-
cific prob lem. You can request from your foodie friend recom-
mendations that fit your taste and your bud get. Drawing on your
friends’ expertise to get answers to your own prob lems makes
more sense than simply imitating them.6
Einstein or the Mechanic?
Looking at past per for mance is a power ful strategy for establish-
ing competence, but it is not as simple as it seems. One diffi-
culty is that per for mance can be largely a matter of luck. A no-
torious con temporary example is the trading of stocks in financial
markets: it is fantastically difficult to tell whether the per for-
mance of, say, a hedge fund is due to the intrinsic competence
of its traders or to dumb luck.7 Even strong per for mance over
several years is not much of an indicator: given how many hedge
funds there are, it is statistically unavoidable that some will per-
form well year after year through chance alone. The same logic
applies to skills that were undoubtedly more relevant during
earlier stages of human evolution, such as hunting large game.
Once a given level of competence is reached, who makes the kil
on any given day is partly a matter of luck, making it difficult to
assess the hunters’ individual competence.8
68 ch ap t er 5
Fortunately, per for mance in many domains— fixing computers,
say—is less erratic.9 But even when per for mance can reliably
be detected, an issue persists: How do we generalize from ob-
served per for mance to under lying competence? When your
friend fixes the printer prob lem on her computer, what should
you infer? We intuitively discard some options: that she is good
at fixing things on Mondays, at fixing gray things, or at fixing
things on tables. But that leaves a wide array of plausible possi-
bilities: that she’s good at fixing printer prob lems on her com-
puter, fixing prob lems on her computer, fixing printer prob lems,
fixing Macs, fixing computers, fixing electronics, fixing things,
understanding complex prob lems, or following instructions.
The fact is that psychologists do not know how people should
generalize from observed per for mance to under lying compe-
tence. Some psychologists argue that competence on many
cognitive tasks is related through IQ. Others think we have
diff er ent kinds of intel igences. Robert Sternberg, for example,
has developed a theory of three aspects of intel igence: analytic
skil s, creativity, and practical skil s. Howard Gardner argues for
eight or nine modalities of intelligence, ranging from visual-
spatial to bodily- kinesthetic.10 Other psychologists yet suggest
our minds are made up of a great many specialized modules—
from a face-recognition module to a reasoning module— and
that the strength of these modules varies from one individual
to the next.11
What ever the correct answer to this complex prob lem turns
out to be, it is clear that humans are endowed with intuitions to
guide their inferences from per for mance to under lying compe-
tence. These intuitions are already on display in young children.
Preschoolers know that they should direct questions about toys
to another child rather than an adult, and questions about food
w h o k n o w s b e s t ? 69
to an adult rather than a child.12 When they are asked who knows
more about how elevators work, preschoolers pick a mechanic
over a doctor, while they pick a doctor over a mechanic when
asked who knows more about why plants need sunlight to
grow.13
Adults also appear to be quite good at telling who is good
at what. As we saw earlier, intraindividual variability in hunt-
ing per for mance means that sustained observation over long
periods of time is necessary to tell who is the best hunter. Yet
people are capable of such observations. When Hadza— a
group of traditional hunter- gatherers from southern Africa—
were asked to evaluate the hunters in their community, their
rankings correlated well with hunting per for mance (as mea-
sured by the experimenters, for instance, by testing archery
skills).14
Moving from the plains of Tanzania to the pubs of South West
England, a recent experiment asked groups of participants from
Cornwall a series of quiz questions on a wide range of trivia prob-
lems, from geography to art history.15 Participants were then
asked to nominate a group member to answer, on their own,
some bonus questions—if they answered wel , the whole group
would benefit. Even though the participants hadn’t received any
feedback on who had given the correct answers in the initial
round of questions, they didn’t rely on rough heuristics, such as
picking dominant or prestigious individuals. Instead, they were
able to accurately select the most competent members in each
trivia category. More significantly, research on po liti cal discus-
sions shows that U.S. citizens are able to figure out who, among
the people they know, is most knowledgeable about politics—
and they are more likely to broach po liti cal topics with these
more knowledgeable acquaintances.16
70 ch ap t er 5
Rational Sheep
That a given individual knows more than we do, either because
they have had access to better information or because they are
more competent, is not the only cue tel ing us that other people
might be right and we might be wrong (or simply ignorant). To
evaluate an opinion we can look beyond the individual compe-
tence of whoever holds it and take into account how many
people hold it.
Accepting something because it is the majority opinion has
a bad press. For millennia people have been castigated for indis-
criminately following the crowd. This distaste for majority
opinion has led some intel ectuals to pretty extreme conclusions,
as when phi los o pher Søren Kierkegaard claimed that “truth al-
ways rests with the minority,”17 or Mark Twain concurred that
“the Majority is always in the wrong.”18
By this logic, the earth is flat and ruled by shape- shifting liz-
ards. Without being as pessimistic as Kierkegaard or Twain, some
experiments suggest that people put little stock in majority opin-
ion. Take the following quiz:
Imagine an assembly that contains ninety- nine members.
They have to decide between two options: option 1 and op-
tion 2. One option is better than the other, but, before the
vote, we don’t know which.
To decide between the two options, they use majority vot-
ing. The ninety- nine members vote, and if one option gathers
fifty votes or more, then it wins.
Each member of the assembly has a 65 percent chance of
selecting the best option.
What do you think are the odds that the assembly selects
the best option?
w h o k n o w s b e s t ? 71
Martin Dockendorff, Melissa Schwartzberg, and I put this
question and variants of it to participants in the United States.19
On average the participants believed that the assembly was barely
more likely than chance to select the best option. Majority vot-
ing would thus make the assembly no more likely to select the
best option than each individual member— quite an indictment
of demo cratic procedures.
In fact, there is a correct answer to this question. Its formula
was discovered in the late eigh teenth century by the Marquis de
Condorcet,20 an extraordinary intellectual who defended the
French Revolution yet ended up having to kill himself to escape
the guillotine. Thanks to the Condorcet jury theorem, we know
that the odds of the assembly being right are in fact 98 percent
(making a few assumptions to which I will return later).
The power of aggregating information from many sources is
increasingly recognized. A century after Condorcet, Francis Gal-
ton showed how averaging across many opinions is nearly guar-
anteed to lower the resulting error: the error of the average is
generally lower, and never worse than the average error.21 Much
more recently, journalist James Surowiecki bril iantly pop u lar-
ized the “miracle of aggregation” in The Wisdom of Crowds.22 Car-
toonist Randall Munroe made this logic intuitive with his xkcd
“Bridge” strip (figure 2).23