Alvin I.

Goldman

Epistemology and the Evidential Status of Introspective Reports
I: Trust, Warrant, and Evidential Sources The question of trusting introspective reports is a question about evidential warrant or justification. It is therefore a question of epistemology, and it behoves us to approach it within the framework of epistemology, which addresses evidential warrant across a broad spectrum of topics and sources. This paper examines the scientific status of introspective reports from the vantage point of general epistemological theorizing. Two opposing forces often arise concerning a candidate source of warrant. A positive force may be a widespread habit of treating it as a source of evidence, a habit that probably derives from fundamental psychological dispositions. Opposing this force are theoretical challenges that give rise to sceptical doubts. Perception and memory are generally viewed as sources of evidence, a view that undoubtedly arises from psychological dispositions. If I see an object that appears to be a book, I am psychologically disposed (in normal circumstances) to believe it is a book; and I will credit anyone who undergoes such a visual experience with having evidence that a book is there. Thus, vision is treated as one evidential source. Memory is a second such source. People are psychologically disposed to believe memory’s deliverances, and to credit such deliverances as providing evidence for the remembered event. Memory deliverances are regularly relied upon in legal contexts (though not without challenge). Third, we inveterately use inductive inference to extrapolate from past observations, and we consider it a source of evidential warrant. Philosophers have historically found reasons to doubt or challenge such sources of warrant, and contemporary psychology can add to these doubts. In the seventeenth and eighteenth centuries, philosophers raised doubts about the
Correspondence: Alvin I. Goldman, Department of Philosophy, Rutgers University, 26 Nichol Avenue, New Brunswick, NJ 08901, USA. Email: goldman@philosophy.rutgers.edu

Journal of Consciousness Studies, 11, No. 7–8, 2004, pp. 1–16

Copyright (c) Imprint Academic 2010 For personal use only -- not for reproduction

productive of the same results across multiple tests. Science would also be threatened by sceptical terrorism. Induction as a source of evidence was sceptically attacked by David Hume (1777/1975) in the eighteenth century. or default. Science similarly depends on memory. GOLDMAN reliability1 of the senses. Ultimately. For example. then. Take deduction. [1] I use the term ‘reliability’ in the philosopher’s sense. seeming to see a book on the table is a deliverance of vision that counts as prima facie warrant for there being a book on the table. induction. evidential sources. epistemology warns us. and the looming danger to human knowledge is severe. Thus. of course. 1995). warrant on proposition p. Copyright (c) Imprint Academic 2010 For personal use only -. Reflection on the notion of evidential sources suggests that some evidential sources are more basic than others. meaning truth-conduciveness or accuracy.I. In the psychologist’s vocabulary. and deduction. but the validation process must be conferred by other. you cannot prove deduction to be reliable to someone who does not start out trusting it. it seems necessary that there be one or more ‘basic’ sources of evidence. Berkeley. for example. and has never been securely rescued from this attack. For example. leading him to posit an idealist metaphysics in which the apparently physical world exists only in the mind of God. they have no evidential leg to stand on. especially basic or fundamental sources like perception. so can it be legitimately regarded as a source of evidence? Contemporary epistemologists are cautious about such sceptical threats. Without such memory-based justification..g. and psychological studies of illusion might add to those doubts. Various doubts about memory have been raised in contemporary psychology. see Schacter. ‘recovered’ memories) and doubts at the theoretical level (viewing memory as a ‘constructive’ process poses worries about its reliability. sceptical threats are easy to construct. viz. or defeated by other evidence.not for reproduction . It means that a given deliverance confers only provisional. argued that the senses give us no solid evidence for a world outside the mind. Can scientists avoid reliance on memory by recording all data at the times of observation? They will still need justified memorial belief about the interpretation of the language or notation in which their data are recorded. ‘reliability’ means something else. No non-circular defence of induction seems to be in the offing. we should not be hasty in trashing our sources of evidence. As Lewis Carroll (1895) showed. It is roughly equivalent to the psychologist’s term ‘validity’. a scientific instrument can be a source of evidence only because it is validated. If we demand of a basic source of evidence that it be non-circularly validated. no scientific hypothesis can be verified. overridden. memory. which can be overthrown. even deduction will be toast. The ‘prima facie’ qualifier is crucial. in his delightful dialogue between Achilles and the tortoise. presumably independent. Past philosophical manoeuvres show that if the bar is set too low. either from source M itself or from a different source.2 A.. What is a ‘source of evidence’? A method or process M is a source of evidence just in case any deliverance of M to the effect that p counts as prima facie warrant for the truth of p. doubts both at the applied level (e. If perceptual observation is relegated to the evidential scrap heap.

If a method frequently produces false beliefs. a non-basic source could have its reliability inductively supported by certifications of its accuracy in a substantial number of tests. or fundamental. Hume (1777/1975) rejected the basicness of testimonial warrant. when these have been calibrated (using basic evidential sources). Under classification (3). but rather is a basic. For example. it can be undercut by additional evidence pertaining either to the speaker or to the facts alleged in the assertion. deduction.EPISTEMOLOGY & EVIDENTIAL STATUS OF INTROSPECTIVE REPORTS 3 What is a basic source of evidence? A method or process is a basic evidential source just in case (1) it is a source of evidence. or derivative from. in the psychologist’s sense of that term). I shall not try to advance here a full theory of what it takes to be a source of evidence. Burge. other sources of evidence. What considerations should exclude a method or faculty from the ranks of an evidential source. there is the general perspective of everyday belief and evidence. 1992. source of evidence of its own (Coady.2 Given the longstanding controversy over such matters. Second. (2) as a non-basic source of evidence. Classifications of candidate evidential sources can be undertaken from at least two perspectives. this PF-warrant is only provisional. where the indicated ‘certifications’ are given by a basic source.not for reproduction . even when there is a psychological disposition to believe its deliverances? One relevant factor is global unreliability (or invalidity. instead claiming that a hearer’s grounds for trusting a speaker is wholly based on the hearer’s personal observations of the speaker’s past veracity plus inductive inference from those observations. Of course. Examples of non-basic sources of evidence would be readings on instruments or measuring devices. I confine myself to such questions as how evidential sources can be challenged and what is the structure of relations between basic and non-basic evidential sources. at least a more basic source. there is a lively debate over the reception of testimony — the process of relying on the reports or assertions of others. In contemporary epistemology. the deliverances of a source would not confer even PF-warrant on the proposition so delivered. In the case of a (categorically) basic source. First. there is the perspective of science. presumably of a broadly inductive rather than deductive variety. and made an inductive inference from this track record. Copyright (c) Imprint Academic 2010 For personal use only -. memory. But is actual unreliability the relevant excluding factor? Or is the relevant factor evidence of unreliability? Perhaps it is [2] The ‘basing’ or ‘derivation’ in question is some sort of inferential relationship. (3) as not a legitimate source of evidence at all. its status as evidential source does not require such certifications. which might impose additional restrictions on acceptance as a scientific source of evidence. Any candidate source of evidence can be classified in one of three ways: (1) as a basic source of evidence. 1993). A hearer is prima facie warranted (PF-warranted) in believing what a speaker asserts even without having checked his past track record. and (2) its status as such a source is not wholly based on. By contrast. several recent philosophers contend that testimonial evidence does not have to be derived from such perceptual and inductive evidence. he isn’t justified in trusting that speaker. Unless a hearer has personally checked a speaker’s previous statements against the facts. induction. and perception (in its various modalities) would be strong candidates for being basic sources of evidence. Presumably. it does not merit even provisional trust.

Why not introspection. Not all legitimate sources can derive their legitimacy wholly from other such sources. a method need not be perfectly reliable to qualify as an evidential source. for the physical world. perception. on pain of either an infinite regress (of sources) or a kind of circularity. then M should not be admitted into the ranks of evidential sources. emotions) based on their ‘awareness’ of these occurrences. This is especially plausible for the physical domain. and the standard objection to coherentism is that justification or warrant cannot be conferred by certain beliefs upon others if the former have to derive their justification from the latter. between the global defeat of a candidate source and the individualized defeat of one of its deliverances.4 A. memory seems to be a basic evidential source. If we have evidence that method M is globally unreliable. where there is more than one sense modality. epistemologists go along with people’s general disposition to form beliefs about their current mental states (sensations. that there are distinct basic evidential sources for distinct domains. however. This is the unsatisfactory circularity to which the text alludes (although what is in question here is the epistemic status of sources rather than individual beliefs). for the future.I. is it a basic evidential source? Again. or inner awareness. Presumably. which might disqualify it from the status of evidential source. For the domain of the past. there must be at least one basic evidential source. If some of M’s deliverances attest to P and other deliverances of M attest to not-P. for the domain of the mental? Although the following proposal has not previously received discussion. [4] Some epistemologists would try to avoid the indicated regress by appealing to Neurath’s boat meta- phor.4 This just shows. M cannot be terribly reliable. neither of which is satisfactory. GOLDMAN the latter. Self-awareness or introspection is the source that confers PF-warrant on these beliefs. it is arguable that some domains have two or more basic evidential sources.5 Vision does not wholly derive its [3] However. We must carefully distinguish.3 In making this acceptance. II: Introspection as an Evidential Source Introspection is accepted as an evidential source by most epistemologists. induction. some of these deliverances must be false. It is plausible. however. Copyright (c) Imprint Academic 2010 For personal use only -. philosophers of psychology as a group probably comprise an exception to this generalization. propositional attitudes. But Neurath’s boat comes down to a coherentist theory of justification (or evidence). Individualized defeat simply makes trust in that particular deliverance unwarranted (on balance). each of which might be a basic evidential source. Evidence might defeat an individual deliverance of M without thereby undercutting the candidate source as a whole and depriving it of its status. If this happens too often. One type of evidence of M’s unreliability is internal inconsistency. Assuming for the moment that self-awareness or introspection is an evidential source. however. that at least one evidential source must be basic. It does not show that there is a basic source for mental states of affairs in particular. many epistemologists would say that if there are any evidential sources. [5] In his instructive discussion of these topics. Alston (1993) tends to treat sense-perception as a whole as a single source of evidence.not for reproduction .

Copyright (c) Imprint Academic 2010 For personal use only -. Under this scenario. Observing this behaviour corroborates the accuracy of the introspection. Here is an example of a strengthening scenario. but this does not mean that it precludes independent validation (nor the possibility of limited invalidation. however. The problem for the second type of validation is that it isn’t clear what (non-verbal) behaviour attests to consciousness. if there were a satisfactory functional analysis of consciousness. If so. I introspect this plan and verbally report it. which has better prospects. as I discussed in two earlier papers (Goldman. this raises the prospect of behavioural evidence being used either to strengthen or weaken the evidential power. I see nothing objectionable in this scenario. To the extent that their deliverances conflict. The possibility of the former is illustrated in the previous paragraph. they strengthen or reinforce each other’s evidence-conferring power. the possibility of the latter is much more problematic. In the present discussion. the sequence of behaviour unfolds. a plan first to wrinkle my nose. I registered pessimism about the independent validation of introspection. This is why. Is the introspection accurate? A predictive consequence of my having such a plan is its imminent execution. There must have been such a plan. plans) and the behavioural validation of introspection as a guide to conscious states. because otherwise the sequence had a very low probability. I am considering the validation of introspection as an evidential source for mental (rather than conscious) states.. Each has the capacity to confer PF-warrant on its own. 1997. in the 1997 and 2000 papers. Notice that according to the present conceptualisation of a basic evidential source. But there is no such analysis of consciousness. The unusual behavioural sequence described above may be good evidence for there having been a plan. a distinction needs to be drawn. and then scratch my left ear with my right hand. it is perfectly intelligible that the degree or strength of a basic source’s evidence-conferring power should be raised or lowered by corroboration or conflict with the deliverances of another basic source. is it the only basic evidential source in the domain of the mental? It is surely plausible that behaviour is also a basic evidential source for the mental domain. In discussing the prospects for behavioural corroboration of introspection. being a basic source does not require independent validation. How far such reduction can go in the case of a basic evidential source is an open question. moreover. the validation or corroboration in question would be entirely feasible. Suppose I form a plan to execute a series of bodily movements. but is it good evidence for there having been a conscious plan? More generally.not for reproduction .EPISTEMOLOGY & EVIDENTIAL STATUS OF INTROSPECTIVE REPORTS 5 evidence-conferring status from touch and touch does not wholly derive its evidence-conferring status from vision. they reduce or qualify each other’s evidence-conferring power. What about introspection? Is it a plausible candidate for being a basic evidential source? If so. and getting one would probably depend on prior use of introspection to determine conscious states (Goldman. or trustworthiness. of introspection. We need to distinguish the behavioural validation of introspection as a guide to mental states (e.g. To the extent that vision and touch agree in their deliverances about physical objects. then curl my left pinky finger. Lo. or weakening). 2000). 2000).

that term is not an apt characterization even of the 1997 position. p.I. First. a term Piccinni (2003. introspection might fall below the threshold. 142) applied to the position staked out in Goldman (1997).g.. a wholly different matter. Actually. Given high scientific standards for reliability of evidence. so the epistemology of introspective report inherits properties of testimony. Perhaps science should also be more demanding in what it countenances as legitimate ‘data’ or evidence. III: The Scientific Status of Introspective Reports The foregoing account of introspection’s warrant does not resolve the evidential role of introspective reports in science. Returning to introspection. but not in trusting someone whose claim is based on divine revelation. that diminishes the listener’s warrant. it doesn’t follow that scientists should do so. Third. An introspective report is a species of ‘testimony’. Scientific practice may be a separate issue. requiring different and more demanding epistemic criteria. for two reasons. Arguably. introspection’s status as a basic evidential source guarantees PF-warrant only for the beliefs of an introspector. and insisted that avoidance of this scenario is an extra test introspection has to pass (above and beyond reliability per se). at least to some extent. there may be special problems associated with the verbal reporting of mental experience which should give scientists pause. e. If the speaker’s source is questionable.6 A. the listener’s warrant may depend on his warrant vis-à-vis the speaker’s source. but one that admits the possibility of independent corroboration by other evidential sources. I admitted the possibility of getting evidence for introspection’s substantial unreliability. viz. epistemologists who sanction introspection as an evidential source are typically offering epistemological principles for common-sense thought and belief. A hearer might be PF-warranted in trusting someone who makes an observation-based claim. Obviously. This says nothing about the warrant of a scientist who hears or otherwise learns of a subject’s report. Second. testimony is not equally a source of evidence across all topics. So my current position could not be characterized as preaching reliance on introspection as a matter of faith. science erects more rigorous requirements for inference than those used by laypersons..not for reproduction . GOLDMAN I am therefore inclined to classify introspection as a basic evidential source. 543). The evidential credibility of testimony partly depends on which source or sources the speaker uses (or claims to use). The 1997 paper endorsed a certain ‘negative’ constraint on the acceptability of methods. ‘that there not be (undefeated) evidence of [the method’s] unreliability’ (1997. Even if laypersons routinely accept introspective reports. So even the 1997 paper did not advocate accepting introspection on blind faith. its credibility varies from topic to topic. Furthermore. scientists might have reasons to doubt the very existence of introspection as a bona fide process. p. In either case. So let us consider these three possible challenges to Copyright (c) Imprint Academic 2010 For personal use only -. Going beyond introspection’s reliability. scientists might justifiably believe things that challenge introspection’s reliability. they might justifiably view reports purportedly based on introspection as untrustworthy. an experimental subject.

and (3) arguments for the inaccuracy of verbal reports. it hardly follows from the special case of commisurotomy that confabulation is the method of all self-attribution. they do not have privileged access to the causes and must infer them. says Wilson. But to the extent that people’s responses are caused by the conscious self. but it may be comparatively rare. p. which constructs explanations of the agent’s behaviour by creating plausible theories. 2002. others may have been persuaded of only the narrower claim that there is no introspective access to cognitive processes. now concedes that the article contained some overstatements. However.not for reproduction . the Nisbett and Wilson argument was wrong about such cases’ (2002. So one oft-cited critic of introspection now takes a substantially qualified position. or to cognitive processes. Gazzaniga therefore postulates a left-brain-based module called the ‘Interpreter’. This is not a major concession. Nisbett and Wilson concluded that people have ‘little or no introspective access to higher order cognitive processes’. 2000). I can have privileged. Wilson. Some philosophers drew the conclusion that introspection is entirely bogus. and instead confabulated such causes. in short. introspective access to a specific memory that just came to mind. Gazzaniga advances a confabulation model of self-attribution based on split-brain experiments. Confabulation seems to be one way self-attribution can occur. and instead offered various spurious explanations. But Wilson now concedes that even this is not correct. Cognitive scientists who doubt the existence of introspection often trace self-attribution of mental states to an entirely different source: inference or confabulation. p. 106). says Wilson. Gopnik (1993) has appealed to Copyright (c) Imprint Academic 2010 For personal use only -. 105). [T]o the extent that people’s responses are caused by the adaptive unconscious. First.EPISTEMOLOGY & EVIDENTIAL STATUS OF INTROSPECTIVE REPORTS 7 introspection: (1) arguments for the non-existence of introspection. 1992. because all that their original article meant to deny was that people have introspective access to the causes of behaviour. they have privileged access to the actual causes of these responses. At any rate. preferences among four identical pairs of panty hose were influenced by the positions of the items in a display. This is introspective access to the content of a thought (Wilson. When a commisurotomy patient’s left hemisphere is ‘blind’ to what the right hemisphere sees. Subjects denied that position influenced their choices. Nisbett and Wilson (1977) argued that subjects were unaware of the true causes of their own behaviour. much more accepting of the introspective method. not just in commisurotomy. Introspectionism endorses a fundamental asymmetry between first. although that piece of information is inaccessible to anyone else. (2) arguments for the unreliability of introspective beliefs. In one of their studies. confined largely to clinical cases. the second author of the paper. Introspection may still be the predominant source of self-attribution.and thirdperson methods of mental-state attribution. This is allegedly what happens in all self-attribution. the talking left hemisphere still confabulates a story about what is in view (Gazzaniga. The precise moral readers drew from the Nisbett and Wilson paper may have varied. just as Nisbett and I argued. Baynes and Gazzaniga.

Gopnik concludes that the impression of immediate self-awareness is an illusion. children’s errors in the cited studies were errors about past beliefs. Self-knowledge is a matter of theoretical inference rendered fluent by expertise. Many experimental psychologists concluded that introspective judgments about experience are inconsistent with measures of performance. From the Wurzburgers’ theoretical perspective. They review many instances from the developmental literature showing that no such parallelism obtains. they were not. inconsistencies between introspective reports about ‘imageless’ thought in the early twentieth century already revealed introspection’s unreliability. As I have indicated. conflicting reports showed that many of the subjects must have introspected incorrectly. see Hurlburt and Heavey. 9). the alleged parallelism between first. no clear relation between accuracy and reported vividness was found. Monson and Hurlburt (1993. GOLDMAN alleged evidence of parallelism in children’s belief attributions to argue against introspection and in favour of inference in both self-attribution and other-attribution.I.not for reproduction . In a deceptive-container task. Rosen et al. argued Monson and Hurlburt. science can certainly produce evidence to challenge introspection’s reliability. At the level of pure description. three-year-olds made analogous errors about their own immediately past beliefs as they make about other people’s beliefs. According to the standard historical account. for example. Gopnik’s case is unpersuasive for several reasons. thereby challenging introspection’s reliability [6] For a more extended defense of introspection as the primary method of self-attribution. For example. Surprisingly. which carry as if in a nutshell the entire meaning of a situation’ (Titchener.and third-person attribution has been roundly refuted by Nichols and Stich (2003). On the other hand. the two sides were in perfect agreement. However. (1997) found that young children have a great deal of difficulty characterizing the pretences of others. Ch. in psychology’s terminology). namely the existence of ‘vague and elusive processes.8 A. these ‘vague and elusive processes’ were imageless thoughts. Subjects in both the Wurzburg and the Cornell laboratories reported identical observations. their disagreement came at a theoretical level. see Goldman (forthcoming. Another premature demonstration of introspection’s alleged unreliability is discussed by Ericsson (2003) and Jack and Roepstorff (2003). 2001) reviewed the Wurzburg and Cornell results and showed that the two factions in fact agreed with each other’s reports of the phenomenon that was called imageless thought. some attempts to do this have been premature or badly interpreted. Assuming similarity in thought across subjects. the facts here are questionable. whereas Gopnik and Slaughter (1991) found that children of the same age did perfectly well on questions about what they themselves pretended or imagined. pp. and introspectionists would not claim that self-awareness is possible for past mental states. A body of research attempted to relate recall accuracy for a presented picture to the reported vividness of the memory. Copyright (c) Imprint Academic 2010 For personal use only -. from Titchener’s theoretical perspective. 1910/1980. 505–6). Second. First.6 The second scientific reason for distrusting introspective reports appeals to its unreliability (or invalidity.

and hence a stronger candidate for acceptability. some information can get lost or become distorted in the ‘translation’. Synaesthestic experiences are difficult to express verbally. Cytowic had a patient. However. When people try to express in words experiences like the appearance of a face.) Visual discrimination may be inaccurate where the visible objects are too far away. MW. because the verbalizations can be inaccurate or misleading. i. however. the (non-verbal) beliefs that awareness generates. however. these problems and roadblocks must be kept in proper perspective. The likelihood of noise entering the translation process is particularly great when the reported experience is inherently difficult to put into words. The psychologists in question took the findings to refute the correspondence between reported and actual vividness. but this should not totally Copyright (c) Imprint Academic 2010 For personal use only -. Was he being metaphoric. But why not abandon the presumed correspondence between vividness and memory accuracy.. Introspection-in-conditions-O* may be more reliable. or verbally interpreting a sensory experience by translating tactile sensations into images a clinician could understand? That is difficult for the hearer to resolve. Epistemologists find this type of manoeuvre attractive in other domains. As Jack and Roepstorff rightly indicate. than introspection tout-court. But it does pose a challenge to the reliability of introspective reports. one can reasonably take into account vision’s range of acuity. Cognitive science has posed many roadblocks to introspection’s reliability. Schooler (2002) points out that an introspective belief. Introspection tout court could be replaced by a more carefully configured source: introspection-in-conditions-O* (where O* is the complement of O).e. the findings only show that one cannot sustain both of the following relationships: a correspondence between vividness and memory accuracy and a correspondence between reported vividness and actual (experienced) vividness. In wondering whether vision is an acceptable evidential source. or meta-representation. the taste of a wine. MW described the taste of spearmint as ‘cool glass columns’. or the intuitions that lead to insights. and patients are likely to ‘embroider’ their descriptions. one could build range restrictions into the specification of the candidate source. requires re-representing the contents of a mental state. in whom taste and smell evoked tactile perceptions in the hands and face. Alternatively. one could reconfigure the source that is considered as a candidate for scientific acceptability. In other words. I do not intend to cast general doubt on such attempts.EPISTEMOLOGY & EVIDENTIAL STATUS OF INTROSPECTIVE REPORTS 9 (validity). preserving the accuracy of the vividness reports? Our third possible source of error concerns problems of verbal reporting. (This is routinely done in calibrating instruments. A natural reaction is to treat this evidence as reducing introspection’s overall reliability. Cases like this do not challenge the reliability of introspection itself. Although I have cast doubt on certain attempts to challenge introspection’s reliability. Suppose an experimental paradigm indicates that introspection in conditions of type O yields false beliefs. A similar problem is discussed by Cytowic (2003) in connection with synaesthesia. As with any recoding process. thereby undercutting introspection’s reliability. the description may distort the reality. There is a related effect of verbalization called ‘verbal overshadowing’ (Schooler & Engstler-Schooler.not for reproduction . 1990).

Many theorists of science insist that science should invoke only public evidence. Finally. then there can be no expectation of valid (i. However. and I think this is right. Jack and Shallice (2001) contend that the Ericsson–Simon verbal-reporting procedures are overly constraining.not for reproduction . personal Copyright (c) Imprint Academic 2010 For personal use only -. reliable) responses. introspection might again pass muster (Thanks here to Tony Jack). Ericsson. This is the aim of Ericsson and Simon’s research program on protocol analysis (Ericsson & Simon. not tokens. Dennett takes a different tack from my suggestion that scientists are PF-warranted in accepting subjects’ verbal reports. This is the sense of privacy I focused on in Goldman (1997). 1997. if behavioural–functional methods are usable to check up on your introspections. at least on the surface. there is a second sense of ‘public’ in which introspection is public. 2003). does the privacy/publicity distinction consist in? In one sense of ‘private’.I. A final factor in assessing introspection’s acceptability for purposes of science is an additional constraint: publicity. By introducing appropriate range qualifications. nobody else can apply it to those same token states.. 2003) instructs scientists to adopt agnosticism about the truth of subjects’ verbal reports. IV: Dennett and Heterophenomenology Dennett’s approach to verbal reports contrasts sharply with mine. as I now am inclined to agree. Reproducibility or replicability is a matter of event types. But what. His heterophenomenological approach (Dennett.e. 1980. Their theoretical and experimental analysis aims to specify types of verbal reports about thinking that offer the highest validity. Arguably. with the understanding that this default attitude should be reversed if there are experimental or theoretical reasons to do so. So in this sense of ‘public’. In counselling agnosticism.10 A. GOLDMAN undercut its evidential viability. Dennett (2003) disputes an earlier descriptive claim of mine (Goldman. there is a third sense of ‘public’ in which it refers to replicability or reproducibility. That seems to preclude scientific reliance on introspection. But the Ericsson–Simon work illustrates how scientific work on introspection fits with the epistemological perspective adopted here. we may define a visual source with a suitably high level of accuracy. one might apply the strategy ‘offensively’ by experimentally identifying conditions that optimise introspection’s reliability. an ostensibly private affair. 1993. exactly. this notion of ‘public’ is most crucial to science: science’s ability to reproduce a phenomenon of interest and make the same finding about it. My suggestion is that the provisional attitude of scientists should be wary acceptance of subjects’ verbal reports. 1991. They contend that the closest connection between thinking and verbal reports is found when subjects verbalize thoughts generated during task completion. By contrast. when investigators try to obtain more information by asking subjects why they responded in a certain way (as in the Nisbett and Wilson 1977 study). Instead of using the range-qualification strategy ‘defensively’ to exclude conditions shown to lower introspection’s reliability. introspection is a private method because everyone applies introspection to their own mental states.

Were subjects really doing this ‘mental rotation’? By varying the angular distance actually required to rotate the two figures into congruence. Not terribly helpful. he reported. let us consider Dennett’s conjectures. as I see it. Referring to mental rotation experiments by Roger Shepard. . and timing the responses. Suppose subjects had said something along the lines he imagines. Dennett (2003) chooses an oddly perverse example. Subjects always say that they are rotating their mental images. Copyright (c) Imprint Academic 2010 For personal use only -. A subject’s statement would be equivalent to: ‘I visualized one of the figures being rotated’. Dennett offers conjectures about purely hypothetical reactions of psychologists to purely hypothetical introspective reports. 1974/1982). it would have lacked commitment to a specific cognitive architecture of visual imagery. what would that prove? The problem. of course. Would psychologists have been persuaded of the analogue transformation hypothesis that Shepard and Metzler advanced? If not. 25) The example is odd because Shepard and Metzler’s subjects never reported rotating mental images. Still. or visual imagery. Dennett writes: Most subjects claim to solve the problem by rotating one of the two figures in their ‘mind’s eye’ or imagination. Instead. explained the latencies of their subjects’ responses in terms of hypothesized mental rotations (analogue transformations). are able to rotate such mental images at an angular velocity of roughly 60° per second (Shepard & Metzler. Their experimental responses consisted of lever-pullings that indicated ‘same’ or ‘different’ — these being judgments about the congruence or non-congruence of two figures (Metzler & Shepard.not for reproduction . e. Shepard and Kosslyn would have never needed to do their experiments to support subjects’ claims that what they were doing (at least if described metaphorically) really was a process of image manipulation. so if agnosticism were not the tacit order of the day. Shepard was able to establish a remarkably regular linear relation between latency of response and angular displacement.. .EPISTEMOLOGY & EVIDENTIAL STATUS OF INTROSPECTIVE REPORTS 11 communication) that scientists make a practice of relying ‘substantially’ on subjects’ introspective reports. (Dennett. . one expects to find examples in which subjects actually made introspective reports that psychologists actually used or ignored. and in each of these fields there is evidence for the validity of these reports’ (2003. As a statement about the ordinary experience of visualization. Since Dennett’s discussion is concerned with how psychologists treat introspective reports. certainly not reports of mental rotation. Practiced subjects. p. is that the intended meaning of those verbal reports would not have been strictly relevant to the intended meaning of the Shepard–Metzler (or Kosslyn) hypothesis. ‘I mentally rotated my image of one of the figures’. since Pylyshyn and others were quick to compose alternative hypotheses that could account for this striking temporal relationship. These were hardly introspective reports. to see if it could be superimposed on the other. 131). This didn’t settle the issue. 2003. but their hypothesis was not supported by mental-rotation reports by their subjects. p. But [7] It is interesting that Wilson (2003) now makes a very similar claim to mine: ‘A casual perusal of jour- nals in virtually all subdisciplines of [psychology] reveals many dependent measures in which participants are asked to report their internal states.g. 1971).7 In disputing my descriptive claim. Shepard and Metzler.

This leaves entirely open the question of whether they (the psychologists) would have accepted introspective reports of visualization as accurate reports — of visualization. There is a general moral here. p. Their introspective reports are typically couched in descriptions. Their subjects were asked to ‘report accurately about which region [a black region or a white region] appeared to be figure at any moment.’ (1991. truthful). But architecturally neutral interpretations are eminently more plausible. as the mental rotation example illustrates. the verbal reports concerned how the stimulus array appeared to them at various times. this leads him to conclude that introspective reports should not be trusted. their introspections might be perfectly adequate (i. do ordinary subjects have introspections about such cognitivist matters. They would not fall into the range of subject matter for which introspection may be regarded as reliable. always prefer the latter. Shepard and Metzler recognized. The investigators evidently assumed that the subjects’ reports of how the stimulus array appeared were accurate accountings of how the array did appear. The following. mere verbal reports of visualization would hardly have sufficed. . even extremely rapid ones. What is the evidence for my claim that the standard practice among cognitive scientists is to give wary.I. which comports with the discussion in Section III. But rarely. GOLDMAN cognitive architecture is precisely what Shepard and Metzler (and Kosslyn) were interested in.e. if ever. . It is certainly true that if ordinary subjects offered introspections about the cognitive processes that interest cognitive scientists — the causal or computational processes underlying some specimen of thought — cognitive scientists would pay little or no heed to them. Naturally. or conceptualisations. and a choice is available between an architecturally loaded interpretation of the report and an architecturally neutral interpretation. and I pick no quarrel with that conclusion under an architecturally loaded interpretation. They would not be credited with any evidential worth vis-à-vis the scientists’ cognitivist hypotheses. They were also asked to report all reversals. Observers reported what they were seeing by pressing one response key whenever (and for as long as) the surround appeared to be figure and pressing the other response key whenever (and for as long as) the centre appeared to be figure . Thus. and they offer much better prospects for truth or accuracy. Peterson et al. however. . Admittedly. laypersons’ descriptions of their mental lives are open to more than one interpretation. (1991) studied figure/ground reversals in perception. seems like a reasonable rule of thumb: ‘When considering an introspective report. 1078). To establish something about cognitive architecture. .not for reproduction . But relative to the coarse-grained descriptions they use. So of course Shepard and Metzler did the experiments they did rather than ask subjects for introspective reports of visualization. much less fine-grained than those of interest to cognitive scientists.. . Copyright (c) Imprint Academic 2010 For personal use only -. provisional credence to subjects’ verbal reports (suitably interpreted)? Here are two simple illustrations. in other words. His proclivity is to interpret ordinary introspective reports in architecturally loaded terms. and these reports were the principal experimental evidence utilized.’ This is just the opposite of Dennett’s practice.12 A.

96–7. Subjects were instructed not to answer a question if they weren’t confident of an answer. and we are giving you total. These other statements are worth exploration. and Copyright (c) Imprint Academic 2010 For personal use only -. like perceptual appearings and sensations? In the figure/ground study. 1991. Dennett. i. or even well-grounded.not for reproduction . These two examples are characteristic of investigators’ default propensity to assume that subjects’ verbal reports are trustworthy. 2003. the investigators ostensibly granted their subjects ‘dictatorial authority’ over whether a certain region of an array appeared (at a given moment) to be figure or ground. If they left a question blank. we first need to ask what it means to give the subject ‘dictatorial authority’ over how things seem to him. In an early study of meta-cognition. 22) You [the subject] are not authoritative about what is happening in you. dictatorial authority over the account of how it seems to you.EPISTEMOLOGY & EVIDENTIAL STATUS OF INTROSPECTIVE REPORTS 13 the subjects’ reports corresponded to their actual visual experiences (with respect to perceived figure and ground). It presumably means that the subject is automatically warranted in believing propositions about how things seem to him.e. p. quoted approvingly in Dennett. What is included under the heading of ‘seemings’? At a minimum. a feeling that they knew the correct answer although they couldn’t remember it at the moment. he also says other things that present a rather different picture. (2003. then he justifiably believes this. In making this assumption. as expressed in their communication. p. (Originally published in Dennett. 2003): You [the scientist] reserve judgment about whether the subject’s beliefs. why not extend such warrant to propositions about (current) desires. beliefs. of course. are true. plans. but only about what seems to be happening in you. This should meet with Dennett’s approval. So if subject S believes that he (now) believes p. pp. about what it is like to be you. he displayed trust in the subjects’ introspective reports.. they were asked to indicate whether or not they had a feeling of knowing the answer. Hart was interested in how their feelings of knowing correlated with their actual knowledge (memory). So here too the subjects were automatically warranted in believing that the seemings (visual appearances) were as they were believed to be. What about other seemings. His procedure assumed that the subjects’ reports of having (or not having) a feeling of knowing were accurate reports of the presence (or absence) of such feelings. Here are two formulations from his contribution to Part 1 of ‘Trusting the Subject’ in this journal (Dennett. But if automatic warrant is conferred in this fashion on these mental-state propositions. officially preaches scientific agnosticism toward subjects’ introspections. Hart (1965) investigated the accuracy of ‘feelings of knowing’. 22) To interpret these passages. However. but then you treat them as constitutive of that subject’s subjectivity. He gave subjects a test of recall involving 50 factual questions.

But Dennett’s own principles seem to imply that the investigator is warranted in regarding the subject as authoritative in his (self-descriptive) beliefs. i. Why doesn’t it follow that the investigator is also warranted in believing the contents of the subject’s (self-descriptive) beliefs? If we track all the apparent consequences of these passages.. verbal reports would not be reliable if they concerned these states. Dennett would be licensing the subject with a tremendous amount of warrant about his own mental life.9 On the question of operational conditions..14 A. events. The second would seek to specify the propositional contents for which it is reliable. GOLDMAN other (personal-level) mental states?8 If this natural extension is made. This is the problem of calibration. Since cognitive scientists generally agree that the vast majority of mental states. that the subject is undergoing or has undergone the reported seemings)? True. This is a topic I won’t venture into.e.I. 1998. 1999.. What about the warrant of an investigator who observes a subject’s report? Setting aside insincerity (not a major issue for Dennett). This does not much threaten the actual truth ratio of spontaneous introspective reports. for example. The resulting position is internally incoherent. One would seek to specify the operational conditions under which introspection is (sufficiently) reliable. however. which arises for any scientific instrument and cognitive capacity. Thus far. and processes are not conscious. I allude here to research on change blindness and inattentional blindness (O’Regan et al. Simons & Levin. Mack & Rock. are quite symmetrical. we find a position strikingly at variance with the scepticism or agnosticism about verbal reports that Dennett usually embraces. testimony can complicate the situation. the license pertains only to the subject’s own warrant. it may be confined to attended events. Copyright (c) Imprint Academic 2010 For personal use only -.. Work by Marcel (1993) indicates that there can be conflicts across different reporting modes. suggests that the range of accurate introspection is even narrower than conscious events. V: Fixing the Range of Introspective Reliability I have conceded — indeed insisted — that introspective reports are not reliable across all conditions or descriptions of mental life. Many studies have established that even if one is ‘seeing’ a certain scene. [8] Notice that from a functionalist point of view there is no principled distinction between beliefs and desires. because they almost always do address conscious events. I would subdivide the problem into two parts. Recent research. why isn’t a subject’s warrant for his beliefs transmissible to an investigator? Why isn’t an investigator who observes the report also warranted in believing their contents (viz. So why should authority be granted to beliefs about one’s beliefs but not about one’s desires? [9] A third dimension is the medium or mode of the report. 1997. 1998). one can easily miss events or dimensions of the scene to which one doesn’t attend. it is pretty clear that introspection is capable of detecting only states of consciousness.not for reproduction . the types of mental or cognitive descriptions for which introspection is prone to be accurate (assuming operational conditions are right). A crucial problem for the theory of introspection is to fix its range of reliability. Rensink et al. Their roles in producing behavior.

A. Selecting appropriate content constraints is extremely tricky. Coady. ‘Content preservation. 157–66. Carroll. (1992).P. reliability should not be expected for descriptions of cognitive architecture. ‘Who’s on first? Heterophenomenology explained. 2nd edition. may not be reliably introspectible. These cannot be directly introspected. M. C. ‘Verbal reports as data. K.C. (1991). Dennett. ‘What the tortoise said to Achilles.E. The Reliability of Sense Perception (Ithaca. and the split-brain: The two minds/one body problem. and temporal relations among mental events may not be reliably detectable. Burge. Ericsson and Simon’s work indicates that maximum reliability is attained when reports are concurrent rather than retrospective. We still need a more precise characterization of what counts as ‘personal-level’. introspection. L. I also benefited from discussion of similar material given at the University of Delaware (Department of Philosophy). 215–51. But here we move into the territory of the reported contents. [10] Thanks to the issue editor (Tony Jack) and two anonymous reviewers for helpful comments on a first draft of the paper. (2003).’ Philosophical Review 102. Ericcson.’ Journal of Consciousness Studies 10 (9–10). pp. If a (naïve) subject were to ‘report’ that her concurrent conscious thoughts are mediated by a language of thought.’ Psychological Review 87. In general. 19–30. (2003). W. 1355–68 (Cambridge.C. This architectural topic is surely outside the window of introspective reliability.’ Mind 4.J. Testimony (Oxford: Oxford University Press). ‘Consciousness. ‘The clinician’s paradox: Believing those you must not trust.S. (1993). Gazzaniga. R. MA: Little Brown). pp.not for reproduction . (1895). Baynes. Dennett. (1993). D. prospects for their reliability cannot be very good. (2003). (2000). even if conscious and attended. 457–88.’ Journal of Consciousness Studies 10 (9–10).’ Journal of Consciousness Studies 10 (9–10). But timing also enters into reliability in other ways. is within the range? One possible answer is: personal-level descriptions. T. NY: Cornell University Press). pp. These reflections indicate that a systematic treatment of introspection’s reliability is well beyond our current grasp. exactly.EPISTEMOLOGY & EVIDENTIAL STATUS OF INTROSPECTIVE REPORTS 15 Perceivers asked to watch a videotape of a basketball game and count the number of times one team took possession of the ball failed to notice a person in a gorilla suit stroll onto the centre of the court and do a little jig. D. As previously indicated. Copyright (c) Imprint Academic 2010 For personal use only -. including its computational properties. M. although one can gain introspective access to them by retrieval or activation. MA: MIT Press). 278–80. ed. Timing factors are another major parameter. Unless verbal reports are restricted to attended events (or event dimensions). H. pp. the micro-structure of cognition. Ericcson. K. pp. (1980). is beyond introspection’s range of accessibility. Brief episodes. which brings them into working memory (consciousness). no cognitive scientist should regard that as reliable evidence for the existence of a language of thought.A. but in no way do they suggest that a systematic treatment is in principle beyond reach. ‘Valid and non-reactive verbalization of thought during performance of tasks: towards a solution to the central problems of introspection as a source of scientific data.S. 1–18. Cytowic. such as standing desires and beliefs. K.’ in The New Cognitive Neurosciences. & Simon.10 References Alston. What.A. Consciousness Explained (Boston. This cannot be the full story because some personal-level descriptions refer to non-conscious states of affairs. & Gazzaniga. pp.A. pp.

(2001). (in preparation). ed. Goldman..I.. (1777/1975). pp. (1999).’ Child Development 62. Strangers to Ourselves: Discovering the Adaptive Unconscious (Cambridge.’ Journal of Consciousness Studies 10 (9–10). 1075–89.16 A. An Enquiry Concerning Human Understanding. pp. GOLDMAN Ericcson. A. Nichols. A.A. ‘Telling more than we can know: Verbal reports on mental processes. 36–71. A. Nidditch & L. Goldman. & Hurlburt.Theories in Cognitive Psychology: The Loyola Symposium. ‘Preschoolers’ attributions of mental states in pretense. ‘Shape recognition contributions to figure–ground reversal: which route counts?’ Journal of Experimental Psychology: Human Perception and Performance 17 (4). & Engstler-Schooler. (1997). C.. ‘Change-blindness as a result of “mud-splashes”. pp. (2003).N. D. A. & Weidenbacher. pp. J. 400–3.. Wilson.T. A. 208–16. A. A Textbook of Psychology (New York: Scholars’ Facsimiles and Reprints). Language. Mental Images and Their Transformations (Cambridge. D. I. (1992).T.J. and Societies Reconstruct the Past (Cambridge. pp.’ Journal of Educational Psychology 56.’ Psychological Review 84. A. & Heavey. Copyright (c) Imprint Academic 2010 For personal use only -.T. 98–110.J.’ Child Development 68. E. Sexuality. Monson. (1993). 231–59. Selby-Bigge (Oxford: Oxford University Press). Hume.A. publicity and consciousness. T. & Singer. R. T. (1993).’ Journal of Consciousness Studies 7 (5). (1974/1982). O’Regan.G. Clark. 368–73.D. Brains. (1977). Memory Distortion: How Minds. J. J. S. pp. Hart. T.W. (1910/1980). T. pp. ‘Memory and the feeling-of-knowing experience. R. 644–9. pp. ed. & Roepstorff. Experimental and Theoretical Studies of Consciousness. R. ‘To see or not to see: the need for attention to perceive changes in scenes. (2003). Gazzaniga. (1990). R. J. ‘A comment to suspend the introspection controversy: Introspecting subjects did agree.Y.’ Philosophy of Science 64.J.T. ‘Mental rotation of three-dimensional objects. ‘Science. ‘Can science know when you’re conscious? Epistemological foundations of consciousness research. & Shallice. J. A. R.’ Cognition 79 (1–2).K.N. Emotions. Jack.1–14. Harvey. R. 168–80. Inattentional Blindness (Cambridge. D. J. Hurlburt. Mack. J.not for reproduction .S.D. & Metzler. H. p. and Intelligence (New York: Basic Books). v–xx.M.’ Behavioral and Brain Sciences 16. pp.’ Psychonomic Bulletin and Review 5.N.. pp.’ Science 171. pp. & Rock.I. ‘Slippage in the unity of consciousness’. MA: MIT Press). R. R.K.I. (1993). (1971).P. A. O’Regan. Gopnik. 3–22. ‘Introspective physicalism as an approach to the science of consciousness.A. ‘Knowing when to ask: Introspection and the adaptive unconscious. Goldman. D. ‘Why trust the subject?’ Journal of Consciousness Studies 10 (9–10). Simons. (2001). (1965). Metzler. pp. 141–56. (1998). Wilson.A. T. (1993). Schacter. ‘Verbal overshadowing of visual memories: some things are better left unsaid.D. Jack.I.T. S. MA: MIT Press). Rosen. Schwebel. (2003).I. J. ‘Data from introspective reports: upgrading from common sense to science.K. Protocol Analysis: Verbal Reports as Data (Cambridge. Schooler. H. pp. eds. (2003). 701–3.’ Journal of Consciousness Studies 10 (9–10). & Slaughter. Titchener.’ Cognitive Psychology 22. 525–45. & Simon..I. ed.’ Psychological Science 8 (5). (1991). pp. Marcel. A.’ in Sampling Inner Experience in Disturbed Affect. (1998).’ Trends in Cognitive Sciences 5 (9). M. (2002).E. Piccinini. & Levin. pp. G. ‘Transformational studies of the internal representation of three-dimensional objects. pp. (2000). (1991). (1995). MA: Harvard University Press).H. R.J. The Simulating Mind (New York: Oxford University Press). P. C. pp. 161–96. ‘How we know our own minds: the illusion of first-person knowledge of intentionality. (1997). pp. Shepard. MA: Harvard University Press). pp. 131–40. M. ‘Telling what we know: describing inner experience. & Wilson. & Clark. Nature’s Mind: The Biological Roots of Thinking.A.A. C. 34. Cooper. & Stich. Mindreading (Oxford: Oxford University Press). Nisbett. V. Peterson. Hurlburt. (1997). Solso. & Shepard. MA: MIT Press). E.. 1133–42.. reprinted in R.L. Rensink. Shepard & L. J. ‘Failure to detect changes to people in a real-world interaction. in Ciba Symposium 174. Rensink. 15–26 (Plenum Press). ‘Young children’s understanding of changes in their mental states.A. D. K. Gopnik.’ Nature 398.