by Hugo Mercier
going to be repatriated.”1 Here, the speaker identifies the relevant
source of information, allowing his interlocutor to gauge its ac-
curacy more easily. Proper credit (if the information turns out
to be true) or blame (if it doesn’t) can be given not only to the
speaker but also to Bill Smith. This motivates members of the
network to be more careful when starting rumors, as false rumors
jeopardize their reputation not only with those with whom they
share the rumor but also with every individual to whom the
rumor then spreads.
By contrast, inaccurate rumors are accompanied by vague
sourcing (“ people say that . . .”) or, worse, by inaccurate sourc-
ing that increases the credibility of the rumor. The rumeur
d’Orléans was made more plausible by its sources: “A friend’s
father is a cop, and he’s investigating a kidnapping case . . .” or “My
cousin’s wife is a nurse, and she treated the victim of an attempted
kidnapping . . .”2 The obvious prob lem with these sources is that
166
c ir c ul a r r e p o r t in g 167
they are simply false: no such cop or nurse exists. A less obvious
prob lem is that the sources remain the same throughout the
chain of rumor transmission.
In theory, the information about the (imaginary) credible
source should have become increasingly diluted as the rumor was
passed along, going from, say, “a friend’s father” to “a friend’s
friend’s father,” “a friend’s friend’s friend’s father,” and so forth.
But this is not what the researchers observed. The length of the
chain was never acknowledged. Instead, most of the people re-
ported the rumor as being validated by “a friend’s father” (or the
equivalent cousin’s wife, etc.). As sociologist Edgard Morin,
who led the team studying the rumeur d’Orléans, put it: “Each new
transmitter [of the rumor] suppresses the new link, and rebuilds
a chain with only two or three links.”3
Sourcing can greatly help or hinder the work of open vigilance
mechanisms. Why does it work so well in some cases, and so
poorly in others? To better understand, we must start by appre-
ciating the omnipresence of sources.
Omnipresent Sources
Paying close attention to sources might seem to be the remit of
professionals. Since Thucydides and his History of the Pelopon-
nesian War, historians have reflected on which sources their work
should be based on, distinguishing primary from secondary
sources, debating the reliability and in de pen dence of their
sources— engaging in historiography. More recently, journalists
have also learned to practice source criticism, not relying on a
single source, finding in de pen dent means of evaluating their
sources’ credibility, double- checking every thing. Clearly, in some
domains, people must take great care to find, track, evaluate, and
cross- check their sources. Absent this learned, reflective practice,
168 ch ap t er 11
the information provided by academics or journalists cannot be
relied upon.
Yet sourcing isn’t restricted to professionals. We all do it, all the
time, but usually in an intuitive, rather than reflective, manner.
For instance, imagine asking your friend Aluna about a movie
you consider watching. She tells you one of the fol owing:
(1) It’s good.
(2) I saw it last week, it’s good.
(3) I heard it’s good.
(4) Osogo told me it’s good.
(5) The
Chicago Sun- Times says it’s good.
Even though the opinion (“it’s good”) remains the same, you
weigh it differently as a function of how it is presented. The least
convincing would likely be (3), because it provides little informa-
tion with which to evaluate the opinion. How you weigh the
others would depend on your judgment on the tastes of Aluna,
Osogo, and the Chicago Sun- Times movie critic. Even when no
source is explic itly provided, as in (1), you would likely be able
to draw some inferences: if Aluna utters (1), she is more likely to
have seen the movie than to base her opinion only on the trailer,
or on a movie review. The provision of information about sources
gives more fodder to our mechanisms of open vigilance.
Specifying the source of our statements is so impor tant that
many languages make it grammatically mandatory. In En glish,
you must indicate the tense of the verb to make a grammatically
correct sentence. In Wanka Quechua, a language spoken in the
south of Peru, you must specify how you acquired a given piece
of information:
(1) Chay- chruu- mi achka wamla- pis walashr- pis: “Many
girls and boys were swimming” (I saw them)
c ir c ul a r r e p o r t in g 169
(2) Daañu pawa- shra-si ka- ya- n- chr- ari: “It (the field)
might be completely destroyed” (I infer)
(3) Ancha- p- shi wa’a- chi- nki wamla- a- ta: “You make my
daughter cry too much” (they tell me)4
The bits in bold— the evidentials— are used to tell whether
the speaker owes their belief to direct perception, inference, or
hearsay. Wanka Quechua is far from being unique: at least a quar-
ter of the world’s languages possess some kind of evidentials.5
Some languages have relatively simple systems with only two
evidentials, such as Cherokee, which distinguishes between first-
hand and non- firsthand information.6 Other languages, like
Kaluli, spoken in Papua New Guinea, have complex systems,
with a great variety of evidentials to choose from.
Whether they are conveyed through evidentials, with explicit
mentions (“Peter told me”), or left implicit (“This movie is good”
suggests direct experience), sources are omnipresent in language.
Why?
For open vigilance mechanisms, an obvious role of source in-
formation is to make a statement more convincing, for example,
by specifying that it stems from direct perception or from a reli-
able individual. But why would sources make a statement more
convincing? After all, if you don’t trust the speaker enough to
accept their statement (“Paula is pregnant”), it’s not immedi-
ately obvious why you should trust them when they give you
source information (“I’ve seen Paula”). Indeed, when the speaker
is thought to be dishonest, source information doesn’t help. If
your poker partner tel s you, “I’m looking at my hand, I have a
royal flush, you should fold,” they won’t be any more convinc-
ing than if they told you, “You should fold.”
Fortunately, in most interactions we don’t suspect our inter-
locutors of such dishonesty. But that doesn’t mean we trust them
170 ch ap t er 11
entirely; far from it. In many cases we don’t think them competent
or diligent enough to warrant changing our minds only on the
basis of their opinion. It is in these situations that providing
sources makes statements more convincing. For example, I trust
my wife a lot. I trust her with
our kids. I would trust her with my
life. But if, while we’re shopping, I think there are eggs in the fridge
and she says there aren’t, I don’t believe her. It’s obviously not that
I suspect her of lying but that I don’t have reasons to believe she’s
in a better position than I am to know what the egg situation is. If
she tel s me, “I checked the fridge before leaving; there were no
eggs left,” I’m convinced that we have to buy more eggs.
By default, statements are attributed to the speaker’s ability
to draw inferences, which becomes the main locus for estima-
tions of competence: we believe people more if we think them
better at drawing inferences (in the relevant domain). By provid-
ing sources we outsource ( pardon the pun) competence to
other cognitive mechanisms— chiefly, perception—or to other
people. These other sources become the locus of the estimation
of competence and can be used to convince our interlocutors
when they believe that our senses, or some third parties, are more
reliable than our inferential abilities.
Yet people do not always provide information about the
sources of their beliefs in order to persuade their interlocutors.
Sometimes, source information has the opposite effect. If Bill
says, “Someone told me Paula is pregnant,” you might be less in-
clined to believe him than if he had simply said, “Paula is preg-
nant.” Why would Bill give you reasons not to believe him?
We use what we know about our interlocutors— how compe-
tent and diligent they are—to evaluate their messages, but we
also use what we know about messages to evaluate our inter-
locutors. If we know for sure, or later find out, that Paula isn’t
pregnant, our opinion of Bill as a competent and diligent source
c ir c ul a r r e p o r t in g 171
of information decreases. But it decreases less if he hedged his
statement (“Someone told me that Paula is pregnant”) than if
he took full responsibility (“Paula is pregnant”).7
Conversely, Bill might want to get more credit than he de-
serves for an idea by obscuring its actual source.8 If he tel s you,
“I think Paula is pregnant,” and you knew that Paula wasn’t preg-
nant three months ago, you might attribute to him the ability to
recognize early pregnancy based on subtle cues, and maybe some
more general social skil s. But if he tel s you, “Paula told me that
she’s pregnant,” he won’t get much credit: he just had to listen
to what Paula was saying.9
Two Degrees of Separation
Providing information about sources serves two broad functions:
to convince our interlocutors, and to engage in reputation man-
agement (i.e., to take more credit than warranted or, on the
contrary, to limit our exposure to reputational fallouts). The in-
terplay of these goals helps explain inaccurate sourcing and its
effects.
Take the sources that often accompanied the rumeur d’Orléans
(“A friend’s father is a cop . . . ,” etc.). Why did speakers provide
such sources? And why did their interlocutors accept them?
For those who provide them, these sources play a dual role:
to increase credibility, and to limit exposure if their interlocutor
questions the validity of the rumor. However, it seems as if this
second goal contradicts the thesis I have defended in the last
chapter, namely, that people spread rumors mostly so that they
can score social points: How could they both not be exposed if
things go wrong (the rumor is rejected), and get credit if things
go well (the rumor is accepted)? Those who spread rumors can
achieve this apparently impossible feat because the value of such
172 ch ap t er 11
wild rumors isn’t so much in the practical implications of their
content as in being something people want to hear about— a
type of mind candy. So someone who transmits the rumor can
manage both to keep their distance from the content (they don’t
say they have witnessed anything themselves) and to get credit
for giving their interlocutors a gift they can use to score social
points in turn.
From the speakers’ point of view, an external, somewhat cred-
ible source hits a sweet spot as a way of increasing plausibility,
decreasing exposure, and improving the overall credit they can
obtain from spreading the rumor.
We see the same pattern in the spread of diff er ent types of false
information. In the 1980s a fear formed in the United States that
snuff movies (movies depicting actual murders, torture, or rape)
were being regularly shot and widely distributed. Nearly all the
people who spread these accusations remained at least one step
removed from any actual witnesses: they had never seen a snuff
movie themselves, but they knew someone who had.10
Similarly, advocates of conspiracy theories rarely rely on first-
hand knowledge. Few people claim to have witnessed the
shooting by Stanley Kubrick of the fake moon landing, or to
know the guy who actually killed JFK. Not even David Icke says
he has seen with his own eyes the human- reptilian beings that
control the earth. Instead, he claims to have “started coming
across people who told [him] they had seen people change into
a non- human form.”11
Hidden Dependencies
Before launching the second Iraq War in 2003, the administra-
tion of President George W. Bush engaged in a vast program of
justification. One of its key arguments, used by Bush as well as
c ir c ul a r r e p o r t in g 173
by high- ranking officials such as Condoleezza Rice and Colin
Powell, was that the Iraqis had attempted to buy “significant
quantities of uranium from Africa.”12 Several intel igence agen-
cies, across at least two countries— the United States and the
United Kingdom— said they had documents in their possession
proving Saddam Hussein’s attempt to buy from Niger hundreds
of tons of uranium oxide, a material that can be pro cessed and
used to build nuclear weapons. This convergence of reliable
sources— the Central Intel igence Agency (CIA), the Defense
Intelligence Agency (DIA), the UK intelligence services—
allowed these accusations to play a central role in the justifica-
tion for the war.
In fact, the evidence all rested on documents peddled by a
former Italian spy to several intel igence agencies. The concor-
dance between the agencies’ assessment was only as good as this
set of documents. And the documents were straight-up forger-
ies. Not only had Hussein not attempted to buy uranium from
anybody, but he had given up on his nuclear weapons program
more than ten years earlier, in 1991.
These forged documents wreaked such havoc in part because
of the White House’s eagerness to justify the war, but also because
the intel igence agencies failed to disclose their sources. The Brit-
ish agencies in par tic u lar played an impor tant role, as they were
seen as providing more in de pen dent evide
nce than the vari ous
American agencies. But they never revealed to their U.S. col-
leagues the basis for their accusations, depriving them of the
ability to work out that all the evidence came from the same set
of documents. As a U.S. intel igence official quoted by the Los
Angeles Times put it: “This became a classic case of circular re-
porting. It seemed like we were hearing it from lots of places.
People didn’t realize it was the same bad information coming in
diff er ent doors.”13
174 ch ap t er 11
Let’s forget for a minute the documents were forgeries, as this
is not the aspect of the story I’m interested in here. If the docu-
ments were real, for each individual agency, disclosing the source
would make its case more convincing. However, given that all
the agencies relied on the same source, their case was more con-
vincing when the sources weren’t disclosed and their opinions
were thought to have been formed in de pen dently of each other.
As explained in chapter 5, a convergence of opinions is a reliable
indicator of the opinions’ validity only to the extent that the opin-
ions have been formed in de pen dently of each other. If they all
rely on the same source, they are only as strong as the one
source.14 In this case, the combined agencies’ case would have
actually been less convincing had they disclosed their sources—
even though doing so would have made each individual case
seem more convincing.
When the agencies failed to reveal their sources, there was a
hidden de pen dency between their opinions. Such hidden depen-
dencies are a particularly tricky prob lem for our mechanisms of
open vigilance. For each informant— here, an intelligence
agency, but the same applies to any other case— their statements
are made less convincing by the absence of a source. As a result,
our mechanisms of open vigilance have no reason to be on the
alert: they are on the lookout for attempts to change our mind,
not attempts not to change it. When someone fails to mention
a source that would make their statement more convincing,
we’re not particularly vigilant. If many people do the same
thing, we might end up accepting all of their statements, with-
out realizing they all stem from the same source, ending up
more convinced than we should be. Not identifying hidden
dependencies is one of the rare failures of open vigilance mech-
