Abhinav Saxena
Glossary

Epistemic vigilance

Also called: vigilance towards testimony, epistemic vigilance theory, vigilance against misinformation

Definition

The set of capacities by which hearers guard against being misinformed, assessing both whether a source is competent and honest and whether what it says is believable.

By · Updated · 2 min read

Communication is only worth having if what we are told is mostly true, and speakers do not always have reason to tell the truth. Dan Sperber and his co-authors argued in 2010 that humans therefore have epistemic vigilance: a suite of cognitive mechanisms, developing from early childhood, that check incoming information so that being told things remains advantageous on balance.

Two targets

Vigilance is directed at the source and at the content. Towards the source we ask, mostly without noticing, whether this person is in a position to know and whether she is well disposed towards us. Towards the content we check what is said against what we already believe, and treat a claim that conflicts with it as needing a reason. The authors stress that vigilance is not the opposite of trust. It is what makes trust affordable, in the way that a pedestrian’s watchfulness is what makes it safe to walk among traffic.

Its place in the philosophy of testimony

The idea bears on an old dispute. Reductionists in the line of Hume hold that a hearer needs positive reasons to believe a speaker; anti-reductionists in the line of Reid hold that we may accept what we are told unless we have reason for doubt. Epistemic vigilance suggests a middle course: acceptance is the default, and it is policed by monitoring that runs in the background and comes forward when something is off.

Vigilance and AI

Our vigilance is tuned to people. Its cues are human ones: hesitation, evasion, a motive to deceive, a record of past dealings. A language model offers almost none of these. It does not hesitate when it is wrong, it has no stake that we can read, and its prose is as assured for a fabrication as for a fact. The usual alarms stay quiet, which is part of why automation bias and the ELIZA effect are so easily produced.

The remedy I defend is to move vigilance from the manner of the answer, where it has nothing to work on, to the kind of question and the system’s record with that kind (should you trust ChatGPT?).

Related terms

TestimonyComing to know something because someone told you; the source of most of what any of us knows.TrustRelying on someone while accepting vulnerability to their failure, in a way that, unlike mere reliance, can be betrayed.Epistemic dependenceThe condition of holding a belief on the strength of someone else’s knowledge or evidence, which one does not have and often could not obtain oneself.Automation biasThe tendency to accept what an automated system says or recommends in place of one’s own checking, so that its errors pass unnoticed and its silences are taken for all-clear.Sycophancy (in language models)The tendency of language models to tell users what they appear to want to hear, shifting answers to match a user’s stated views or pushback without new evidence.DeepfakeAn audio or video recording, made with machine-learning methods, that realistically depicts a person saying or doing something they did not say or do.Echo chamberA social structure in which outside voices are not merely absent but actively discredited, so that members come to distrust every source that does not belong to it.

Further reading

  1. Sperber, D., Clément, F., Heintz, C., Mascaro, O., Mercier, H., Origgi, G. and Wilson, D. (2010). Epistemic vigilance. Mind & Language 25(4): 359–393.doi.org/10.1111/j.1468-0017.2010.01394.x