Lecture Supplement on Alvin Goldman’s “Discrimination and Perceptual Knowledge” [1976][1]

     Copyright © 2013 Bruce W. Hauptli

I. Introduction:

In this essay we are introduced to Goldman’s initial attempt to develop a version of externalism often called “reliabilism.”  In his Problems of Knowledge: A Critical Introduction to Epistemology, Michael Williams maintains that such views rely upon:

subjunctive conditionals; they say something about what S’s beliefs would have been if (contrary to fact) p had not been true or circumstances had been slightly different.  This may seem far-fetched.  Why should we care about what someone’s beliefs would have been as opposed to what they actually are?  To see why, compare the concept of knowledge with that of reliability, which also has an important subjunctive component.  A reliable car will run properly in a variety of weather conditions.  My car is not reliable simply because it started this pleasant morning, but because it would have started even if the weather had turned nasty.  We can think of knowledge as a kind of reliability in our beliefs: truth reliability.[2] 

The radical externalist’s key notion is ‘reliability’.  This is not quite so straightforward an ‘empirical description’ as naturalistic epistemologists like to think.  Reliability is a normative standard governing adequate performance.  Because of this, any talk of reliability involves an implicit reference to a (possibly variable) range of conditions.  What that range is (or ought to be) is something we decide, not something we simply discover.[3] 

In his The Reliability of Sense Perception, William Alston notes that David Armstrong, Fred Dretske, and Robert Nozick “always shy away from questions of justification and connect reliability directly with knowledge” while “Goldman and Swain...develop a notion of the justification of belief in terms of reliability.”[4]  In his Contemporary Epistemology, Ralph Baergen maintains that:

...one could be a reliabilist and either an internalist or an externalist.  To be an externalist reliabilist, simply claim that having introspective access to the reliability of the process (or to the input, or whatever) is not required for justification.  And to be an internalist reliabilist, simply add the sort of internalist constraint to this.[5] 

We will use Goldman’s essay here to come to understand externalism better, and to see the sorts of problems which arise as one tries to pursue this orientation.  Of course, however, the core element of this reading will also be to see how hard it may be to try and overcome the Gettier examples as one tries to develop a full analysis of knowledge.  Whereas Goldman offered an account of many different sorts of knowledge in his “A Causal Theory of Knowing,”[6] here he limits his discussion to non-inferential perceptual knowledge. 

One reason he needs to abandon his earlier view (that beliefs must be causally connected with the facts to be known if one is to have knowledge) is that he believes his earlier account is susceptible to “Gettier counter-examples.”  The “barn facsimile case” on p. 87 points to the flaw in his earlier analysis: in both the “normal” and the “deformed” case Henry’s belief that he is seeing a barn is caused by his perception of the barn.  The existence of the facsimiles in the deformed case shows that a causal connection alone is not sufficient for a true belief to be knowledge—a successful causal analysis needs to “discriminate” between these two cases. 

Thus Goldman advances a view which will require that beliefs be produced by “reliable cognitive mechanisms or processes” if they are to lead to knowledge:

86 roughly, a cognitive mechanism or process is reliable if it not only produces true beliefs in actual situations, but would [also] produce true beliefs, or at least inhibit false beliefs, in relevant counterfactual situations. 

A “counterfactual” situation is one like the barn facsimile case—one which does not, but could, obtain.  Look at his discussion of the twins Judy and Trudy (pp. 91-92) to get a further handle on his revised view.  Sam can perceptually know that he is seeing Judy if he has

91-92 …some (visual) way of discriminating between them (though he may not know how he does it, i.e., what cues he uses).  If he does have a way of discriminating between them, which he uses on the occasion in question, we would say that he knows it is Judy.  But if Sam frequently mistakes Judy for Trudy, and Trudy for Judy, he presumably does not have a way of discriminating between them. 

In getting started in our understanding of his view, it is important to focus for a moment on the parenthetical clause above)—Goldman wants to maintain that Sam (and knowers generally) need not know what cues they utilize, and need not know the processes or mechanisms they employ).  Here even more clearly than in the earlier article, we can see that Goldman wishes to advance a version of externalism (he would want to say a reliabilism).  As he notes at the end of this essay:

102 I am trying to fashion an account of knowing that focuses on more primitive and pervasive aspects of cognitive life, in connection with which, I believe, the term ‘know’ gets its application.  A fundamental facet of animate life, both human and infra-human, is telling things apart, distinguishing predator from prey, for example, or a protective habitat from a threatening one.  The concept of knowledge has its roots in this kind of cognitive activity. 

In rejecting the “Cartesian-style” justificatory conception of knowledge (where one needs, and must be aware of, a justification if one is to be said to know), Goldman maintains that all that is needed (besides belief and truth) is that the belief be produced by a reliable cognitive mechanism!  Thus he would allow for non-inferential knowledge in the case of infants, chicken-sexers, some animals, and (metaphorically) in supermarket doors! 

Of course, the real work comes in specifying what such a reliable mechanism is like; and, as noted above, Goldman will build his account around whether or not the mechanism can adequately distinguish mechanisms which produce true beliefs in actual situations while also produce true beliefs, or at least inhibit false beliefs, in relevant counterfactual situations.  Of course, a key element here will be specifying which counterfactual situations are relevant—if one treats this class too broadly, knowledge becomes impossible!  That is, if one believes one must eliminate “skeptical alternatives” (Cartesian demons, concerns that the world may have been created only a few moments ago with all the geologic records and memories we rely upon, etc.), then no reliabilist’s account stands a chance of getting off the ground.  Goldman claims, however, that his account will help us understand not only non-inferential perceptual knowledge, but also skepticism:

101-102 my theory protects the possibility of knowledge by making Cartesian-style justification unnecessary [thus protecting it from many skeptical arguments].  But it leaves the door open to skepticism by its stance on relevant alternatives.  This is not a failure of the theory, in my opinion.  An adequate account of the term ‘know’ should make the temptations of skepticism comprehensible, which my theory does.  But it should also put skept8icism is a proper perspective, which Cartesianism fails to do. 

Now let’s turn to the article and see how well he does in his efforts to develop a theory which will meet the goals sketched above. 

II. The Text:

87-88 After offering his barn facsimile example, Goldman points out that it undercuts his earlier analysis. 

88 In addition, he claims that while Peter Unger’s “non-accidentality” analysis would seem to draw the right conclusions in the two cases here, it “…is not very satisfying, however, for the notion of ‘non-accidentality’ itself needs explication.[7]” 

He also contends that his example poses a problem for the sort of analysis offered by Keith Lehrer and Thomas Paxson, which we will be turning to next.  I’ll skip over this at this point, but encourage you to return to this discussion as you are evaluating their essay. 

88-89 Having established that the barn facsimile case poses a problem to a number of attempts to provide an analysis of knowledge, Goldman begins to develop his response to it.  Suggesting that knowledge requires that S knows that p if he can distinguish actual states of affairs [89] “...from a relevant possible state of affairs in which p is false.”  In terms of the example:

89 since, by assumption, a state of affairs in which such a hypothesis holds is indistinguishable by Henry from the actual state of affairs (from his vantage point on the road), this hypothesis is not ‘ruled out’ or ‘precluded’ by the factors that prompt Henry’s belief.  So, once apprised of the facsimiles in the district, we are inclined to deny that Henry knows. 

  Let us be clear about the bearing of the facsimiles….[their] presence…does not ‘create’ the possibility that the object Henry sees is a facsimile.  Even if there were no facsimiles….it would be possible that the object on that site is a facsimile…. 

That is, Cartesian demons or other traditional skeptical “hypotheses” might yield the example! 

The qualifier ‘relevant’ plays an important role in my view.  If knowledge required the elimination of all logically possible alternatives, there would be no knowledge (at least of contingent truths).  If only relevant alternatives need to be precluded, however, the scope of knowledge could be substantial.  This depends, of course, on which alternatives are relevant. 

89-91 While we may want to say that Cartesian and other “alternatives” are idle, it is specifying how to draw this claim out that requires a lot of detail.  An essential element of Goldman’s “barn facsimile example” is that, he believes, it accords with our intuitions in a way that the “hypotheses” of Descartes and Russell discussed briefly on p 89 do not! 

90 One might try to appeal to “semantic considerations” regarding how ‘know’ is used/how it means, or

one might try to appeal to the knower’s specific situation to specify specific regularities for distinguishing idle and relevant cases, but Goldman contends that

91 “…I assume that radical or unusual alternatives are not ordinarily entertained or taken seriously….” 

91-92 He wants his analysis to grow, organically, out of our normal perceptual discriminations (rather than out of semantic, linguistic, or pragmatic considerations)!  Thus he turns to the “Judy and Trudy Twin Case.”  The case calls for an analysis in terms of Sam’s ability to “discriminate” between the twins, and leads to a consideration of counterfactual situations (e.g., were he presented with Judy, rather than Trudy, would he have not have said “It’s Trudy”?).  This yields Goldman’s “first stab” at an analysis:

92 S (noninferentially) perceptually knows that p iff

(1) S (noninferentially) perceptually believes that p,

(2) p is true, and

(3) there is no relevant contrary q of p such that, if q were true (rather than p), then S would (still) believe that p. 

He will find that this analysis is too restrictive, and in a very detailed discussion spanning pp. 92-101 Goldman will improve upon it by discussing first “perceptual equivalence” and using this to develop an improved characterization of such knowledge.  We will not focus on all the elements of his development here, but instead highlight a number of the most important points. 

92-93 the discussion of Oscar, Dack the dachshund, German shepherds, and wolves helps us see that when Oscar (who knows dachshunds and German shepherds are dogs, recognizes the difference between them, but has trouble discriminating German shepherds from wolves) sees Dack in the field, a relevant alternative would not be a wolf standing in the field.  This discussion leads him to claim:

93 if Oscar believes that a dog is present because of a certain way he is ‘appeared to,’ then this true belief fails to be knowledge if there is an alternative situation in which a non-dog produces the same belief by means of the same, or a very similar, appearance.  But the wolf situation is not such an alternative: although it would produce in him the same belief, it would not be by means of the same (or a similar) appearance. 

So, for noninferential perceptual knowledge, the notion of a “relevant alternative” must be one which uses the same (or a very similar) “appearance!” 

Since Goldman is, effectively, building upon his earlier causal theory, he wants to speak of the appearances and the “relevant alternatives” in terms of a causal discussion—whether or not Oscar or Sam would believe p if presented with a relevantly similar appearance which would produce the belief that p where p is not the case.  On p. 93 he notes that his discussion here will have to be “relativized to:

persons (or organisms),

times, and

the properties of objects

This leads to his discussion of “perceptual equivalence” [pp. 94-96] we are introduced to:

“Distance-orientation-environment” [DOE] relations,

“Ordered-triple” specifications of “perceptual states of affairs” which specify an object, a set of non-relational properties, and DOE relationship, and

The de re, relational, or transparent sense of ‘perceptually believes’:

--to understand this let me start with the distinction between it and ‘de dicto’ by using a poker example: some games are played in a manner which requires that the player “declare” her hand, others are played where the “cards speak.”  The former are de dicto games, the latter are de re ones.  If one mischaracterizes a winning hand in a de dicto game, one might lose!  In our world we often are tolerant in our interactions, and recognize that individuals meant to say something different from what they actually say. 

This, finally, leads to the more careful analysis of noninferential perceptual knowledge [pp. 97-101.].  Rather than deal with the technical details of his analysis, I want to focus on its core theoretical elements.  First, as he notes:

98 a perceptual cognizer may be thought of as a two-part mechanism.  The first part constructs percepts (a special case of internal states) from receptor stimulation.  The second part operates on percepts to produce beliefs.  Now, in order for the conditions of the analysans [his analysis on pp. 97-98] to be satisfied, each part of the mechanism must be sufficiently discriminating, or ‘finely tuned.’  If the first part is not sufficiently discriminating, patters of receptor stimulation from quite different sources would result in the same (or very similar) percepts, percepts that would generate the same beliefs.  If the second part is not sufficiently discriminating, then even if different percepts are constructed by the first part, the same beliefs will be generated by the second part.  To be sure, even an undiscriminating bipartite mechanism may produce a belief that, luckily, is true; but there will be other counterfactual, situations in which such a belief would be false.  In this sense, such a mechanism is unreliable.  What our analysis says is that S has perceptual knowledge if and only if not only does his perceptual mechanism produce true belief, but there are no relevant counterfactual situations in which the same belief would be produced via an equivalent percept and in which the belief would be false. 

On pp. 98-101 he “tests” his analysis against the “barn facsimile” case; the dachshund-wolf case; and a case suggested by Marshall Swain involving a dark room, mirrors, and a candle.  He goes on to discuss what might occur if percepts came from electrodes implanted in the brain rather than via the normal perceptual mechanisms, and possible moles on the twins (p. 100).  Together these lead him to add an additional clause to his analysis:

(4) S's propensity to form an F-belief of percept P has an appropriate genesis. 

 He knows this is "altogether too vague," but it points to what will also need to be spelled out if we are to have an adequate account of a "reliable process." 

101-102 Overall Summary:

101  His account is to avoid the "Cartesian" [internalist] view, which he holds to be too demanding--or too strong (they exclude clear exemplars of knowledge--indeed they seem to make it impossible).  Instead his view emphasizes that the beliefs which are to be known be "suitably caused" by a "reliable process" which would not only cause the beliefs in actual situations and not produce false beliefs in "relevant" counterfactual situations.  He contends that while his view does not address global skepticism ("Cartesian skepticism," it does not preclude skepticism of a more particular variety (skepticism about whether the factors involved are "reliable," "relevant," etc.). 

102 He contends that the attractiveness of his view is that it explains why it is tempting to say that the automatic doors "know someone is there," (that they can "tell" this, and that it clearly allows for animals and children to have perceptual beliefs which are clearly knowledge.  He contends that the core advantage of his view is that it:

...focuses on more primitive and pervasive aspects of cognitive life, in connection with which, I believe, the term 'know' gets its application.  A fundamental facet of animate life, both human and infra-human, is telling things apart, distinguishing predator from prey, for example, or a protective habitat from a threatening one.  The concept of knowledge has its roots in this kind of cognitive activity. 

(end)

III. Criticisms of Goldman’s Reliabilism and of Reliabilism In General:

1. In his “There Is No Good Reason To Be An Academic Skeptic,” Peter Klein maintains that:

…merely having a reliably produced belief without any reason for thinking that it is reliably produced seems  like having money in the bank without knowing that it is there.8

2. In his The Nature of Rationality, Robert Nozick notes that:

questions have been raised about the context in which reliability is supposed to obtain.  Is a procedure to be reliable only the times it actually is used, or reliable all the times it could be used in the world as it is, or reliable also in worlds similar to the actual one, or reliable in worlds that in general are as we believe the world to be (whether or not that belief is correct)?  There are also questions about the degree of reliability that a rational procedure must have.  Must it be the most reliable procedure available, or can a belief be rational when it arises from a quite reliable procedure that is not the very best?  Must the procedure be more reliable than not, or can a belief about the explanation of a phenomenon be rational when it arises thorough the most reliable procedure available for arriving at correct explanations, even though that reliable is less than 50 percent?  And in assessing a procedure, mustn’t we look not simply at the percentage of times the procedure gets things right but also at what happens when the procedure gets things wrong and how bad the resulting situation is?9

3. In her Considered Judgment, Catherine Elgin maintains that:

given the number and variety of our beliefs, it is reasonable to think that some are reliably correlated with their evidence.  So, arguably, such de facto reliabilism entitles us to confidence that we know something.  But since it does not enable us to identify the beliefs in question, it does not enable us to recognize our knowledge.  We remain hostage to epistemic fortune.  Being unable to differentiate between reliable and unreliable correlations, we cannot avoid error without luck.  Unrecognizable knowledge is cognitively and practically valueless, contributing nothing of use to deliberation or action.  At best, de facto reliabilism achieves a pyrrhic victory over skepticism.  The knowledge it provides is not obviously preferable to ignorance.10

4. In his An Introduction to Contemporary Epistemology, Matthias Steup maintains that:

there are three main objections to process reliabilism.  According to the first, there is the possibility of an evil-demon world in which perceptual beliefs are justified although they are unreliably produced.  This objection raises the issue of whether reliability is necessary for justification.  The second objection presents the challenge of a reliably clairvoyant subject who does not know that he is reliably clairvoyant; his beliefs are supposed to be unjustified even though they are reliably produced.  This objection raises the issue of whether reliability is sufficient for justification.  And according to the third objection, when we apply process reliabilism to particular cases, it is not possible to specify the relevant cognitive processes in a way that is neither too broad nor too narrow.  This objection raises what is commonly referred to as the “generality problem.”11

Process reliabilism is supposed to be a naturalistic theory—a theory that allows us to evaluate beliefs without applying evidentialist considerations.  However, when process reliabilism is applied to particular cases, evidentialist considerations make their entry through the backdoor.  In order to handle cases in which a reliably produced belief is defeated by contrary evidence, process reliabilists must introduce a process such as taking into account contrary evidence.  And in order to handle clairvoyance cases in which supporting evidence is missing, they must appeal to processes such as taking into account the absence of supporting evidence.  It could therefore be argued that, to get the right results, process reliabilists must ultimately employ evidentialist considerations.12 

Back to PHI 3300 Home Page

File revised on 09/28/2013.