Why Believe What People Say?

Stevenson, Leslie: Synthese, 94:3 (1993;maart) p. 429

The basic alternatives seem to be either a Humean reductionist view that any particular assertion needs backing with inductive evidence for its reliability before it can rationally be believed, or a Reidian criterial view that testimony is intrinscially, though defeasibly, credible, in the absence of evidence against its reliability.

Some recent arguments from the constraints on interpreting any linguistic performances as assertions with propositional content have some force against the reductionist view. We thus have reason to accept the criterial view, at least as applied to eyewitness reports. But these considerations do not establish that any rational enquirer must have the concept of other minds or testimony. The logical possibility of the lone enquirer, who uses symbols and thereby expresses some knowledge of his world, remains open — but it is a question we have no need to pronounce upon.

The practice of accepting observation-statements is in fact extended to chains of testimonies believed to start in perception or in some other kind of justification, but the arguments for doing this are not so clear.

1.

Very often the only answer one can give to the question 'How do you know?' is 'Someone told me so'. (Let us throughout this paper use the words telling, saying, and testimony in a sense wide enough to include the use of writing, print, telephone, radio, television, etc. — not just face-to-face conversation.) If we were not entitled thus to rely on testimony, each of us would know very much less than we think we do — only what one has see for oneself, or what one can inductively support or deductively prove with one's unaided resources. With some claims one might, if one took enough trouble, check the matter out for one self and justify one's judgement by perception of proof. But in many other cases — such as assertions about the past, about present events too far away to be perceived, or about matters beyond one's scientific or mathematical competence — verification by the hearer is out of the question.

Our actual dependence on testimony is enormous, then. But when (if ever) is it reasonable to accept something because someone says so? I will concentrate on the first-person form of this question, looking not so much for conditions for the transmission of knowledge from A to B (formulated from the point of view of a third person, C), but asking from B's point of view what, of anything, can justify B in believing what A says. I first essay a brief history of the topic, and then I formulate some rival principles about the epistemology of testimony. I present an argument for the a priori but defeasible acceptability of testimony based on perception; I consider a limitation of this argument, and finally some possible extensions of it.

2.

One standard view of testimony is that no proposition can be justified merely by the fact that someone has asserted it, i.e., that testimony has no intrisic or primary evidential force (unlike perception, memory, and induction, according to most theories of knowledge). But this cannot exclude the assembling of evidence that some kinds of testimonies are reliable and hence, justifiably believable on inductive grounds. If one finds that reports on certain kinds of topics or made in certain sorts of circumstances or by certain kinds of people tend to correlate well with the facts of the relevant matters whenever one investigates them for oneself, one might reasonably come to rely on those kinds of testimonies thereafter. This is what has been called the reductionist view of testimony — that it can earn justifying force only by inductive success.

There is an alternative account which we can label criterial, by which belief on the basis of testimony is reasonable, i.e., prima facie justified, by definition. According to an unrestricted version of this, whatever anyone says about anything is, in the absence of contra-indications, worthy of belief. It has been argued that if a speaker knows that P and says that P, then (under certain typical conditions) his hearer comes to know that P too. According to a more restricted version, it is only in certain appropriate circumstances that A's asserting that P gives a good reason for believing it. For example, if it is known (or, at least, reasonably believed) that someone is or has been in a position to see something, this tends (unless overriden by other factors) to make their testimony about it credible. Eyewitness reports are treated in the courts, in historical enquiry, and in everyday life as having just such epistemic status, defeasibly justifying belief.

In the late medieval and renaissance periods there prevailed an epistemology which strongly emphasized the authority of testimony — at least from recognized sources — but seemed to find no proper place for inductive argument. We have long since learned to ridicule the cast of mind for which what Aristotle said, or what is written in the Scriptures, carries more weight than what can be observed. There has been such a major change in world-view that it is hard for us to understand how such an epistemology could be followed by rational people. As Hacking has shown in the early chapters of his book The Emergence of Probability (1975), the very word probable once meant approved of or attested to, before our modern concept of probability as believability based on statistical evidence was developed. There was even a tendency to treat what we nom call inductive evidence as a special kind of testimony — 'what Nature herself tells us' or 'what is written in the great book of the World' — which suggests the idea of a reverse reduction, namely, of induction to testimony. Remnants of this ancient conception perhaps survive in those uses of the words sign and mean, which analytic philosophers tend to find conceptually promiscuous — e.g., 'Red spots are a sign of measles' or 'Those dark clouds mean rain'. Our diagnosis of such usages as involving a primitive or 'natural' sense of 'sign' or 'meaning', quite distinct from that relevant to language proper, shows how the older view is not a serious candidate for us now: we live far on this side of the epistemological innovations of the seventeeth century, epitomized by Bacon and Descartes.

But in rejecting the medieval view, Descartes went to the opposite extreme; for in his Discourse on Method and Mediations he repudiates all reliance on the testimony of others and resolves to accept only what he can justify with his own unaided mental resources. "[A]s soon as my age allowed me to pass from under the control of my instructors", he tells us, "I entirely abandoned the study of letters, and resolved not to seek after any science but what might be found within myself or in the great book of the world" ([1637]/1954, Part One). "Letters" here seems to mean all the writings regarded as authoritative in the schools, and talk of "the great book of the world" must for him have become a dead metaphor. On Descarte's new individualistic approach, testimony can have evidential force only in a very secondary way, if at all.

Hume provides the classic example of this reductionist approach. At the beginning of the famous discussion of miracles in his first Enquiry, he claims that our assurance in any piece of testimony "is derived from no other principle that our observation of the veracity of human testimony, and of the usual conformity of facts to the reports of witnesses". Where there is contrary evidence, such as conflicting testimony from someone else, or the bad record of the witness, or (as Hume is specially concerned to emphasize) from the sheer improbability or miraculous nature of the alleged fact, we qualify or entirely withhold our belief. "The reason why we place any credit in witnesses and historians, is not derived from any connexion, which we perceive a priori between testimony and reality, but because we are accustomed to find a conformity between them" ([1748]/1962, Sec. 89).

In the Treatise Hume had already discussed in some detail our human propensities to believe. In one of the less frequently read sections (Bk. I Pt. III, Sec. IX) he wrote that "custom, to which I attribute all belief and reasoning, may operate upon the mind in invigorating an idea after two several ways" (my italics). One of these is that with which Hume's name will forevermore be associated — experience of constant conjunctions leading to habits of inference. But the other — most relevant here — is if "a mere idea alone... shou'd frequently make its appearance in the mind, this idea must by degrees aquire a facility and force... The frequent repetition of any idea infixes it in the imagination" ([1739]/1888, p. 116). Hume adverts here to a tendency in human nature which is exploited by advertisers, governments, campaigners, and all parents and teachers who rely on authoritative assertion and repetition. But his attitude to it is not exactly parallel to his treatment of induction. Hume treats our tendency to base expectation on past experience as beyond criticism, though also beyond rational justification; but of belief in testimony he writes:

No weakness of human nature is more universal and conspicuous than what we commonly call CREDULITY, or a too easy faith in the testimony of others... When we receive any matter of fact upon human testimony, our faith arises from the very same origin as our inference from causes to effets, and from effects to causes; nor is there any thing but our experience of the governing principles of human nature, which can give us any assurance of the veracity of men. But tho' experience be the true standard of this, as well as of all other judgments, we seldom regulate ourselves entirely by it; but have a remarkable propensity to believe whatever is reported, even concerning apparitions, enchantments, and prodigies, however contrary to daily experience and observation. ([1739]/1888, pp. 112-13)

For Hume, then, our propensity to believe each other can and should be restrained, so to stay within the limits justified by experience of "the governing principles of human nature". Stated thus generally, this is an epistemological norm which everyone must surely accept — the word 'credulity', denoting a tendency to accept what goes beyond reasonable standards, expresses this. But it remains to be seen exactly what shape this norm should take, in particular whether it should require one to have positive evidence for someone's reliability before one is entitled to accept what he says.

A different approach is suggested in the work of Reid, Hume's compatriot, contemporary, and critic. In Chapter 6, Section XXIV of his Inquiry into the Human Mind ([1764]/1970), Reid claims to find a close analogy between "the testimony of nature given by the senses" and "human testimony given by language" (note how the former phrase echoes the renaissance view noted above). He thinks "the general principles of our constitution" which fit us to receive information by both these means are very similar. Piously, Reid says that "the wise and beneficent Author of our nature" has implanted in our natures two principles that tally with each other. One is a "propensity to speak truth": lying does violence to our nature, and requires art or training, inducement or temptation. The other is "a disposition to confide in the veracity of others, and to believe what they tell us". If it were not for this latter "principle of credulity" (by which label Reid must intend no criticism), children could never learn what they need to know. Reid's allusions to God could presumably be replaced by talk of evolution without substantial change to his epistemology.

Unlike Hume, Reid claims a parallel in epistemic status between this "principle of credulity" and what he calls "the inductive principle", namely, the human tendency to expect experienced constant conjunctions to be continued into the future. Given the uniformity of nature, inductive inference serves us well; and because of our propensity to speak truthfully to each other, our credulous tendencies also servce us well. In each case, truth is generally (though not always) attained: and in the absence of any contra-indication in any particular case, we may reasonably follow our natural tendencies to believe. For Reid, then, our accepance of testimony is just as intellectually respectable as our iductive inferences. Admittedly, both need to be moderated by experience — each person finds out as he goes through life that certain kinds of testimonies are unreliable, and that some observed correlations are merely accidental. But Reid's position is that any assertion is credit-worthy until shown otherwise; whereas Hume implies that specific evidence for reliability is needed.

Many epistemologists have followed the Humean, reductionist approach. They have assumed that in a rational reconstruction of our knowledge, the subject must first justify belief in the external world by his own unaided resources before he can even begin to justify belief in other (rational) minds, or to argue (inductively?) that certain sounds or marks made by other bodies express assertions, or that the content of these assertions are (in some circumstances) likely to be true (see, for example, Ayer 1973, pp. 104-05). More recently, however, there has been a reaction towards a Reidian, criterial view of testimony. Such an approach is strongly suggested in many of Wittgenstein's remarks, posthumously published in On Certainty (1969): for example: "The child learns by believing the adult. Doubt comes after belief" (sec. 160); also "I believe what people transmit to me in a certain manner. In this way I believe geographical, chemical, historical facts etc... And can it now be said: we accord credence in the way because it has proved to pay?" (sec. 170). A negative answer seems to be invited to this last question, rejecting the reductionist approach; but it is difficult to discern argumentative structure in the jottings of Wittgenstein's last days. Coady (1973) is a rare example of explicit argument against reductionism. Evans (1982, Ch. 5.2) suggests that a reorientation of traditional views of testimony is necessary both for a sound epistemology and a sound philosophy of language; but he, too, was not given the time to develop his thoughts on the topic.

3.

  So much by way of preliminary identification of the options and their history: we must now address the substantive questions of how good are the arguments on either side. But we must first try to formulate more precisely the rival a priori principles about when the acceptance of testimony is defeasibly justified. How should the criterial view be best formulated? If it were suggested that:

If	B hears and understands A say that p,
then	B is justified in believing that p;

the obvious objection would be that if I have reason to suspect my informant may be lying, then I would not be justified in believing him. But if I think him sincere in his assertion, or at least have no reason to suspect his insincerity, is that enough? Could we accept:

If	B hears and understands A say that p, B has not reason to doubt A's sincerity,
then	B is justified in believing that p?

This would be far too literal. One obvious reason is that whenever I have strong independent evidence against P, I can hardly be right to believe P just because said A said so. P may go strikingly counter to (what have been widely taken to be) well-known particular facts (consider, for example, the allegation that Hitler did not authorize the 'Final Solution'). Or P may be miraculous, inconsistent with highly confirmed laws of nature (for example, the doctrine of virgin birth). This of course is the Humean point that if someone asserts something incompatible with what there is strong reason to believe, the testimony cannot outweigh the counter-evidence already possessed, unless one can make a rational case that it is more probable that that evidence is flawed than that one's informant is mistaken. Suppose we add a third clause as follows:

If	B hears and understands A say that p, B has not reason to doubt A's sincerity, B has no evidence against p,
then	B is justified in believing that p.

This is surely too liberal also, for even where one has no evidence against what one is told, one may have good reason to doubt the reliability of one's informant, that is, the likelihood that what he or she says about the matter is likely to be true (however sincerely they assert it). If someone claims to read in my palm that I will fall in love with a brown-eyed stranger in my fiftieth year, I can adduce no very good evidence against the content of that prediction — it is, for all I know, entirely within the bounds of psychological possibilty — but I think I have excellent reason for saying that nobody, not even Gypsy Rose Lee herself, can have any justification for such predictions. Often the reasons for doubting realiability are more specific to one's informant. If my window-cleaner gives me a tip that a certain share is going to rise rapidly, I would be well-advised not to give much credence, or invest much money, on that basis; but if someone with inside knowledge of the company tells me — or if it is revealed that the window-cleaner is privy to such knowledge — then the epistemological and financial situation might well be different.

Consider then the following formulation of a criterial epistemology of testimony:

(TC)	If	B hears and understands A say that p, B has not reason to doubt A's sincerity, B has no evidence against p, B has no reason to think that A's belief about p is not justified,
	then	B is justified in believing that p.

This represents a criterial view which is unrestriced as to subject matter: whatever anyone asserts about anything may be accepted, provided one knows of nothing counting against it, their sincerity in saying it, or the reliability of this person's beliefs on this topic. The justication is eminently defeasible, of course — for example, the very next person one meets may tell one the exact opposite — but that is no surprise: we are only in the market for defeasible guidance here.

The reductionist view, in contrast, demands more than the mere absence of doubt about the reliability of the speaker: it requires that one have positive evidence in favour. The criterial approach treats testimony as 'innocent' (i.e., trustworthy) unless shown guilty; the reductionist treats it as 'guilty' (i.e., not worthy of belief) until a good track-record is shown. Here is a first shot at formulating the principle behind the reductionist approach:

(Tr)	If	B hears and understands A say that p, B has not reason to doubt A's sincerity, B has no evidence against p,
	then	B is justified in believing that p if and only if B has reason to think that A's belief about p is justified.

The 'only if' is needed to distinguish the reductionist approach from the criterial, for the latter can happily accept that evidence for the speaker's reliability can be sufficient to justify acceptance; what is in dispute is whether it is necessary. The introduction of 'only if' into the conclusion requires the inclusion of 'for' into (iii).

(Tr) is still importantly ambiguous, however. On an extreme reductionist view, the required evidence should be completely available to B, consisting only in correlations between B's hearing of assertions by A (and/or others) and B's own observations of the relevant facts; there should be no relieance on testimony in assembling this purely inductive evidence. But, on a less stringent view, B would be allowed, in assessing A's reliability, to depend on the testimony of C, and perhaps many others. For example, if little Johnny cries 'Wolf', I might be able to justify believing him by appeal to a number of cases in which, prompted by just a cry from him, I have seen a wolf myself: I would thus meet the stronger requirement. But it might be that I could only satisfy the weaker one, for example, by relying on the word of Johnny's older sister Janey that in her experience he has been a reliable wolf-detector. Of course, I might have checked out Janey's wolf-recognitional skills for myself, but, there again, I could take her mother's word for it; or I might be confident about Janey's wolf-reports on the grounds of her perceptual competence with other animals, or other things generally (whether demonstrated or reported). In the light of such possibilities, the very strong requirement mooted above does not seem very attractive.

But the spirit of the reductionist approach remains that all justified reliance on testimony is at bottom a complex kind of induction — one is not to believe what someone says just because they say it, but only when one can assemble an empirical argument that what they say is likely to be true. One may appeal to other testimony to support belief in these correlations, but only in so far as it is itself supportable in similar fashion. There can thus be a regress of dependence, justifying A's testimony, or that of people like A, or testimony about topics like P, by appeal to the testimony of B, or of people of a kind which B exemplifies, or about topics like Q. But the reductionist idea is that this regress must be finite: to justify accepting another's assertion, it must terminate in correlations between sayings and facts, both of which one has oneself observed. Thus in building up knowledge of the world, or in rationally reconstructing justification for the beliefs one already has, one may proceed through a number of layers of dependence on testimony; but at no stage can one justify a belief merely because someone has told one so, even if one has no evidence against it or against the informant's reliability. To express this reductionist view we need to streghten (Tr) somewhat as follows:

(TR)	If	B hears and understands A say that p, B has not reason to doubt A's sincerity, B has no other evidence for or against p,
	then	B is justified in believing that p if and only if B has evidence that A's belief on matters such as p are likely to be true ; and if that evidence involves appeal to other testimony, B must have evidence for the reliability of those informants' beliefs, and any such regress of evidence must end in inductive correlations involving only what B has simply observed.

Reid's analogy between testimony and perception can now be expressed by comparing (TC) with the following principle for the defeasible acceptability of perceptual experience:

(P)	If	B seems to perceive that p, B has not evidence against p, B has no reason to doubt the reliability of her perceptual faculties,
	then	B is justified in believing p.

And while we are about it, we can formulate similar defeasible principles governing memory and induction:

(M)	If	B seems to remember that p, B has not evidence against p, B has no reason to doubt the reliability of his own memory,
	then	B is justified in believing p.

(I)	If	every time B has observed an event of type F, she has soon thereafter observed an event of type G, B observes another F, B has no reason not to expect a G this time,
	then	B is justified in believing that another G will soon occur.

But to formulate epistemic principles is not to accept them, let alone to argue for them. Scepticism about perception, memory, and induction can take precisely the form of questioning whether we should accept even the defeasible principles (P), (M) and (I). And there are interesting questions whether these three have to be accepted together as a conceptually interdependent package, or whether one might accept (P) without the others, or (P) and (M) without (I). (There does seem to be an undeniable presupposition of (P) by (M), and of (M) by (I).) Our topic here, however, is not these well-worked issues of epistemology but the less discussed one of the epistemic status of testimony. Let us assume, then, for the sake or argument in this paper, that (P), (M) and (I) are not only generally accepted (as they surely are, implicitly at least), but can be rationally defended against sceptical attacks. Our question now is: What is their relation to testimony? Does the acceptance of (P), (M) et (I) somehow involve accepting the analogous (TC), or is the way left open for the rival (TR) - or need one accept either?

...