IS EPISTEMIC VIGILANCE INDISPENSABLE?

Part of Michaelian's strategy is to propose an array of stronger or weaker possible interpretations of our claims and to argue at best for the weakest. He starts by suggesting that our evolutionary conjecture can be given two readings with importantly different implications: according to a stronger reading, epistemic vigilance is indispensable for communication to endure among humans. According to a weaker reading, vigilance is dispensable but still adaptive. We did write, ‘It is because of the risk of deception that epistemic vigilance may be not merely advantageous but indispensable if communication itself is to remain advantageous’, which suggests that we recognize these two possible readings without committing to either. In the light of Michaelian's discussion, I now believe that we could and should have been even less definite. Given the variety of forms of communication and of types of partners involved, neither reading would be plausible if taken to apply to human communication as a whole. Communication among people, in particular close kin, on matters where their interest are aligned does not require epistemic vigilance towards the risk of deception in order to be advantageous to all participants. On the other hand, several typical forms of human communication such as bargaining or arbitration could hardly be practiced in a manner advantageous to all in the long run without vigilance.

Michaelian's ‘two readings’ involve treating human communication as a static given and asking whether epistemic vigilance is indispensable for it as a whole. From an evolutionary point of view, it would make better sense to pose the question in a more fine-grained and dynamic manner. In the six millions of years or so since the separation of the ancestors of modern humans from the ancestors of their most closely related great apes, chimpanzees and bonobos, the semantic and social scope of human communication has hugely broadened, with presumably an ever increasing acceleration of this broadening in the past 40,000 years. Together with the huge expansion of human cognitive capacities and of human cooperation — the three developments being obviously related – this is the most striking aspect of hominin evolution. In others animals, we expect the evolution of communication to have been severely limited by, among other factors, differences of interests between senders and receivers (Dawkins and Krebs Reference Dawkins, Krebs, Krebs and Davies1978; Krebs and Dawkins Reference Krebs, Dawkins, Krebs and Davies1984). Why not in humans? Our suggestion is that humans evolved cognitive capacities and cultural institutions that (1) made it possible to communicate much richer contents, and (2) also made possible a degree of epistemic vigilance that reduced the risks of misinformation and made communication advantageous on a uniquely large scale. The benefits of vigilance may be negligible in some communicative interactions and essential in other interactions. All I feel confident to say is that, without vigilance, human communication would be a very different and probably much more restricted affair. Anthropological studies in societies where forms and levels of vigilance differ would certainly throw some light on the issue (see for instance Gilsenan Reference Gilsenan and Kapferer1976).

IS EPISTEMIC VIGILANCE EFFICIENT?

With the same strategy of suggesting stronger and weaker interpretations of our claims, Michaelian writes:

Needless to say, we conjecture that the stronger claim is true: people are generally vigilant towards the risk of deception (in the form of a low-key monitoring, not of active suspicion) and they are good enough at detecting it to make it a cost-effective investment of mental energy. If people were uselessly vigilant, this vigilance — or should we then call it ‘paranoid attitude’? — far from explaining anything, would be another and quite puzzling thing to explain.

That epistemic vigilance is a permanent dimension of our communicative exchange is not received wisdom. Many philosophers believe that people are generally not just trustful, but trustful as a matter of course, without any permanent even low-level attention to the risk of misinformation. Thomas Reid famously argued that we have been endowed by God with a disposition to speak the truth and a disposition to accept what other people tell us as true. Modern philosophers such as Burge, Davidson, Lewis or Millikan have defended similar views not because of any specific empirical evidence or argument, but because basic truthfulness and trust play a central role in their philosophy of language (for a detailed discussion of Millikan's views in this respect see Origgi and Sperber Reference Origgi, Sperber, Carruthers and Chamberlain2000). The psychologist Daniel Gilbert has produced experimental evidence in favour of such a Reidian (or, as he calls it, Spinozan) approach (e.g. Gilbert et al. Reference Gilbert, Tafarodi and Malone1993). In our article we discussed these philosophical arguments and Gilbert's experimental evidence and argued that they were not particularly compelling. We surveyed various sources of empirical evidence suggesting that people, including children, are as a matter of course paying a modicum of attention to the trustworthiness of communicators and to the plausibility of what they assert. I will readily grant that the import of this evidence is open to discussion. Still, it is strong enough to justify doing further empirical research on the topic rather than assuming that the facts of the matter can be decided on a priori or common sense grounds, or on the basis of just a few clever experiments.

Considering both the degree to which vigilance might be practiced and its degree of efficiency, Michaelian distinguishes six possibilities. In three of them, vigilance plays no role, either because people are not vigilant, or because vigilance is inefficient. He focuses on the three possibilities where vigilance is exercised and is not wholly without efficacy:

Michaelian concludes in favour of weakly effective vigilance. For my part, on theoretical grounds I would tend to think that vigilance is likely to be moderately effective, but I don't believe that we have, or even are close to having, clear empirical evidence to help us decide the issue. Michaelian is more confident, but this is because he redefines ‘epistemic vigilance’ in a narrow way that does not correspond to what we were talking about, and that, more importantly, is not particularly useful for addressing the puzzle presented by the unique scope of human communication.

Michaelian seems to attribute to us the view that ‘epistemic vigilance is a matter of processes devoted to screening out incoming false information on the basis of available behavioural cues’. Showing that vigilance in this narrow sense is not efficient would, he holds, be quite damaging to our conjecture. This is a misunderstanding.

Here is how this misunderstanding might have arisen. To suggest that a concern for reliability of communicated information is a permanent feature of human interaction, we cited experiments (Willis and Todorov Reference Willis and Todorov2006; see also Origgi Reference Origgi2012) suggesting that looking for signs of trustworthiness is the first thing people do when seeing a new face, even when they are presented with the picture of a face for a mere 100 milliseconds. We did not assume that people were correct in in forming judgments on that basis. We mentioned this research as evidence of people's concern, not to say obsession, with trustworthiness. In the same spirit, we talked briefly about the rich lie detection literature that suggests (even if there are problems of ecological validity with the experiments) that people are much less good at detecting lies from behavioural cues than they think they are, and are in fact close to random in their judgments in the matter.

Michaelian goes at length over a wider sample of this same literature defending its negative conclusion (which we do not dispute) as if it was a source of objection against the idea that epistemic vigilance is pervasive and for the most part efficient. Nowhere did we argue or imply that detection of lies through behavioural cues was important to the success of vigilance. Indeed, if people could not help but provide their audience with behavioural evidence of their bad faith when they lie, deception would be nearly impossible and the whole issue we are discussing would hardly arise. But if, as we suggested, there has been an evolutionary arms race between our abilities as senders and our abilities as receivers, being able to lie without behaviourally betraying that one is doing so should be an early evolving adaptation for the sender in that arms race, pushing the receiver to look for less easily suppressed evidence. So, I agree with Michaelian that there is no evidence that this kind of vigilance would be effective. I would surmise that very little energy is invested in this form of vigilance in ordinary conditions. Of course, if someone we have independent reasons to suspect of lying starts acting nervous, we notice and interpret this as possible evidence of bad faith. If someone talking to us starts stuttering and sweating profusely, we look for an explanation and moral unease caused by lying is one distinct possibility. On the other hand, we do not monitor for such kinds of evidence every time someone is talking to us. Indeed, it would not be cost-effective at all.

When we argued that there is a suite of abilities contributing to epistemic vigilance, we did not see lie detection on the basis of behavioural cues as a particularly important item in that suite. In any case, remember why we introduced the notion of epistemic vigilance: to explain how receivers can limit the risk of being misinformed and in particular deceived. Our point was that there had been selective pressure for any cost-effective relevant ability. Abilities likely to be involved include drawing inferences from remembered past communicative performance, understanding the sender's immediate purpose in communicating, or taking into account the sender's reputation. Michaelian seems to think these could not be bona fide items in the epistemic vigilance suite. Here is an example of his reservations. We evoked a series of experiments by means of which Olivier Mascaro investigated the development of epistemic vigilance among pre-schoolers (see Mascaro and Sperber Reference Mascaro and Sperber2009). In one of the experiments, children were told that an informant (a puppet) was mean and did not want them to find a sweet. Five-year-olds—but not four-year-olds – inferred that the informant was lying about the location of the sweet (see also Vanderbilt et al. Reference Vanderbilt, Lio and Heyman2011). This, Michaelian comments, ‘tells us nothing about children's ability to detect deception where they do not already have reason to take it to be present’. Right, if by ‘detection’ we understand ‘detection based on behavioural cues’, but what these results do tell us is that five-year-olds pay attention to such reasons and draw appropriate conclusions from them. This is a perfect instance of epistemic vigilance as we defined it (not arbitrarily but on theoretical grounds).

There are obvious costs in remembering someone's past communications, evaluating them in the light of later confirming or confuting evidence, tracking the reputation of people who might want to inform or misinform us, interpreting not just what a person says, but her motivations in saying it. Paying these costs makes sense if they help us evaluate the reliability of present and future communication (and sometimes in revising our evaluation of past communications that still influence us). Michaelian himself writes:

Again, I don't see why only what he calls ‘real-time detection’ should be involved in a sensible, theoretically useful notion of epistemic vigilance. Finding out after the fact that one has been deceived may limit the damages of this deception and, going beyond mere updating of one's beliefs, prevent future deception, that is, contribute to fulfilling the function of epistemic vigilance.

Also, what Michaelian calls ‘normal sensitivity to evidence’ needs unpacking. Every state of affairs, every event is bursting with evidence that we do not pay attention to because of its irrelevance to us. We are ‘normally sensitive’ to potentially relevant evidence, and not to evidence in general. Our sensitivity to evidence potentially relevant to the epistemic evaluation of communicated information cannot be explained by appealing to some unspecified notion of ‘normal sensitivity’. It can by appealing to the hypothesis that we have an evolved disposition to be epistemically vigilant.

In our article we distinguished vigilance towards incompetence (which has been richly studied, in particular from a developmental point of view, see e.g. Clément et al. Reference Clément, Koenig and Harris2004; Koenig and Harris Reference Koenig and Harris2007) and vigilance towards deception and argued that only the latter posed a major evolutionary problem. While the selective pressure for the two kinds of vigilance may differ and be stronger for vigilance towards deception, the psychological mechanism involved may be in part the same in both. In particular, we distinguished vigilance towards the source of information – who to believe? – and vigilance towards the content communicated – what to believe? Vigilance towards the content attends to the coherence of information communicated with background information, and to its internal coherence (Mercier and Sperber Reference Mercier and Sperber2011). We argue that some evidence on coherence is available as a by-product of a relevance-guided comprehension procedure, so that, up to a point, vigilance towards the content can be a low-cost, low-attention affair. This vigilance towards the content is equally relevant to the detection of deception as it is to the detection of incompetence. As any liar knows, the coherence that normally goes with speaking truly takes some effort to achieve and maintain in deception. Michaelian however ignores this vigilance towards the content, as if it was irrelevant to our conjecture, but it is not.

EXPLAINING HUMAN COMMUNICATION

Michaelian concludes his discussion of epistemic vigilance (understood as real-time detection on the basis of behavioural cues) with the claim that humans have a ‘slight sensitivity to dishonesty [that] stems from the occasional transparency of liars’.

So, what we took to be a serious evolutionary problem seems not to arise.

What does it mean to say that ‘the base rate of deception is low’? There are a few problems with such a statement. Michaelian cites a number of interesting studies but what they document is the rate of daily deception in given (contemporary, Western) populations as indicated by self-reports, or the rate of deception in controlled experiments, with obvious problems of extrapolation or generalisation. In what sense are these rates ‘base rates’? The expression would make sense if we could measure the frequency of deception in a communication system where receivers were gullible. If that base rate were sufficiently low, epistemic vigilance might indeed be superfluous. But there is no way to obtain such a measure. Otherwise what is aimed at is measuring the rate of deception in the world as it is, and since it is certainly not proven that there is no epistemic vigilance at work, a low rate might be to a large extent an effect of this vigilance.

Moreover, available data does not even vaguely indicate the aggregated expected disutility of deception for receivers. Vigilance would be much more important if people lied once in a while on some momentous matter than if they lied ten times a day on irrelevant details. Mere frequency of deception without a sense of its potential cost for receivers is not really relevant to the issue.

Still, do the studies cited by Michaelian at least suggest that there is little need for receivers to be vigilant towards the risk of deception? Well, not really. Here is an example. Levine et al. (Reference Levine, Kim and Hamel2010), cited by Michaelian, report three experiments investigating whether people would lie with or without a motive to do so. The authors write:

Even without extrapolating from this experiment, I don't see how a 60% rate of deception when it is in the interest of the communicator to deceive can be construed as evidence that epistemic vigilance is superfluous.

Michaelian agrees that available evidence on the base rate of deception ‘by itself is too limited to be conclusive’. He offers however another argument, a deduction actually, for assuming that this rate must be low:

I agree with the premise that we are largely insensitive to behavioural cues of deception, but not with the premise that therefore unless the base rate were low, communication would collapse. What is simply ignored in this argument is the possibility that other mechanisms for epistemic vigilance stabilise communication. By ignoring the possibility of epistemic vigilance and the effect it would have on the rate of deception, you ‘prove’ that the base rate of deception is low enough to make vigilance quasi superfluous.

To explain this alleged low rate of deception, Michaelian offers some considerations that have relevant implications for our disagreement. He writes:

I agree – and we mentioned the point ourselves – that lying involves cognitive costs that may make it preferable to tell the truth rather than lie in the absence of reasons to do so (see, however, Shalvi et al. Reference Shalvi, Eldar and Bereby-Meyer2012). But the second and third costs invoked in the above quotation make sense only if receivers exercise epistemic vigilance. Why otherwise should liars worry about their credibility? Why should they monitor the receiver and remind themselves to role-play? All this extra effort is clearly useful when addressing a vigilant audience. Not so if receivers are gullible.

Michaelian also evokes emotional cost:

I don't disagree with this conjecture (even though it should be noted that, unlike energy costs, emotional costs are not necessarily detrimental to fitness and that therefore invoking them in an evolutionary explanation is not self-evident). However, wouldn't the ‘socially imposed deterrents’ that Michaelian invokes require widespread epistemic vigilance to have any efficacy? The phenomena that he sees as making epistemic vigilance superfluous are best explained, I would claim, by the prevalence of epistemic vigilance.

From an evolutionary point of view, the fact that humans can engage in communication the way they do, expect others to be rather honest and indeed are rather honest themselves begs for an explanation. Our explanation is that communicators are generally honest because they are addressing a vigilant audience. Quasi-universal vigilance makes dishonesty less likely to be beneficial in the short run and more likely to be costly in the long run: falsehoods may be disbelieved, and dishonesty may have reputational costs.

There is however an alternative possible explanation that deserves discussion. It could be that the deterrence Michaelian invokes is just an application of whatever practices contribute to the imposition of morality in general rather than being the effect of dedicated epistemic vigilance mechanisms. More generally, one way in which our conjecture about epistemic vigilance might be false or at least off the mark would be if the evolution of honesty in communication was just an automatic consequence of the evolution of cooperation and morality.

The evolution of human cooperation and the related evolution of human morality have no doubt played a major role in the extraordinary broadening of human communication. It might even be thought that the difference of interests between senders and receivers is prevented from compromising the stability of communication by the evolution of moral dispositions. These dispositions would make senders quite generally honest; the risk of misinformation would be so low as to make it more cost-effective for receivers to be quite generally trustful. So might the argument go. Note however that theoreticians of the evolution of human cooperation and morality all assume that some form of what could be called ‘moral vigilance’ is essential. Those who favour group-selection accounts of the evolution of morality typically give a central role to punishment (e.g. Gintis et al. Reference Gintis, Bowles, Boyd and Fehr2003); those who favour a mutualistic approach give a central role to reputation (e.g., Baumard et al. Reference Baumard, André and Sperber2013; Sperber and Baumard, Reference Sperber and Baumard2012). In other terms, no one assumes that people's own individual moral sense is sufficient to cause them always to behave morally. What is generally assumed on the contrary that the moral vigilance of others plays an important role in everybody's relatively moral behaviour.

So, could honesty in communication result from, or be an aspect of, a more general moral sense bolstered by moral vigilance? Could what we have described as epistemic vigilance be just an effect of moral vigilance? This is conceivable. It might be investigated by looking for instance for dissociations between epistemic and other forms of moral vigilance both in development and in attention and inference. If some such dissociation were found, this would suggest that epistemic vigilance is sui generis; if none were found, it would suggest that epistemic vigilance is just moral vigilance applied to communication.

According to the approach we suggested however, there is not one single mechanism of epistemic vigilance; there are several mechanisms turned towards either the trustworthiness of the source or the believability of the content. For similar reasons, moral vigilance might also better be viewed as a suite of mechanisms attuned to different types of moral behaviour and different types of evidence of morality, rather than as a single evaluator of the moral conformity of people's behaviour. I would expect all these more specific mechanisms to have distinct evolutionary and/or developmental trajectories, and multiple kinds of dissociation to occur. Again, these are conjectures that I believe are sensible and worth exploring, not strong claims.

In this speculative context, I have found Michaelian's discussion and arguments, if not compelling, at least stimulating and useful, and I am grateful for them.