Outline of Goldman's "A Casual Theory of Knowing"

SYNOPSIS:

Alvin Goldman argues that S knows that p, where p is an empirical proposition, if and only if the fact that p is causally connected in an “appropriate” way with S’s believing p.

OUTLINE:

Step 1: Goldman hypothesises that what goes wrong in Gettier examples—i.e., cases in which S allegedly has a justified true belief that p but does not know that p—is  that there is a lack of a causal connection between S’s believing that p and the fact that p.

Argument: In Gettier’s second example, in which Smith comes to believe the disjunction “Either Jones owns a Ford or Brown is in Barcelona” based on the false belief that “Jones owns a Ford”, the fact that makes the disjunction true (viz., Brown is in Barcelona) is not causally connected to Smith’s believing the disjunction.

Upshot: We must add the requirement of causal connection between the fact that p and S’s believing that p to the traditional analysis to avoid Gettier’s objections.

Step 2: Goldman begins an examination of different cases of empirical knowledge with knowledge based on perception.

Argument: Suppose that there is a vase before S which is being obscured from S’s view by a laser photograph of a different vase and that S comes to believe that there is a vase before him based on the photograph.  In such a case, according to Goldman, S does not count as seeing the vase since there is no causal connection between the real vase and S’s belief that there is a vase. 

Upshot: Our ordinary concept of sight (i.e., knowledge acquired by sight) includes a causal requirement.

                          

Step 3: Goldman continues his examination of different cases of empirical knowledge with knowledge based on memory.

Argument: Neither (i) believing (or knowing) a fact as some time t0 and then believing (or knowing) it at a later time t1, nor (ii) having the impression of remembering (e.g., having one’s brain artificially stimulated to produce a memory impression) are sufficient for remembering (or knowledge based on memory). We must add a causal connection between the earlier and later case of believing (or knowing) in order for it to count as a case or remembering.

Upshot: A causal connection between an earlier belief (or knowledge) that p and a later belief (or knowledge) that p is a necessary ingredient in memory.

Step 4: Goldman continues his examination of different cases of empirical knowledge with knowledge based on inference.

Argument: Goldman contrasts the case in which an eruption leaves lava on the countryside which remains in place until someone, S, perceives it and the case in which the lava is removed and then (centuries later) someone else (unaware of the real volcano) places lava on the country side to make it look as if there had been an eruption.  If, in the latter case, S comes to perceive the latter, and then infers that there was an eruption, S would not count as knowing that there is an eruption.  This is because the eruption is not causally connected to S’s belief that there was an eruption.

Upshot: A necessary condition of S having inferential knowledge that p is that his believing that p be connected with the fact that p by a causal chain.

Step 5: Goldman concludes his examination of different cases of empirical knowledge with knowledge based on testimony.

Argument: Cases of knowledge based on testimony begin with the fact that p causing a subject, T, to believe that p and subsequently assert that p, which eventually leads to another subject, S, believing that p based on T’s assertion. Here, as before, there must be a causal chain connecting the fact that p to S’s believing that p in order for the belief to count as knowledge.

Upshot: There is a causal requirement for knowledge based on testimony.

Step 6: Goldman concedes that the requirement that S knows that p only if the fact that p caused S’s belief that p is too strong.

Argument: Assuming that we can know facts about the future, and precluding the possibility of backward causation, S’s belief that p cannot always be caused by the fact that p in order for S to know that p.  Moreover, I may come to know that there was smoke coming out of my chimney last night based on an inference from the fact that there was a fire in my fireplace last night, even though the smoke is not the cause of the fire in my fireplace.

Upshot: We must weaken the requirement that there be a causal connection between the fact that p and S’s belief that p to include cases in which the fact that p and S’s belief that p share a common cause.

Step 7: Goldman considers John Saunder’s and Narayan Champawat’s counterexample to Michael Clark’s analysis of knowledge.

Argument: Suppose that Smith forms the belief that “Jones owns a Ford” based on the testimony of Brown, who is generally reliable.  However, (unknown to Brown or Smith) Jones actually sold his Ford the day before and bought a Volkswagon.  Jones subsequently wins a Ford in raffle, making Smith’s belief true.  Although Smith’s belief is true and fully grounded (i.e., all his grounds are true) he still does not know that Jones owns a Ford.

Upshot: Clark’s requirement that a belief be fully grounded fails to provide a sufficient condition for knowledge due to the lack of a causal connection between Smith’s belief and the fact believed.

Step 8: Goldman considers Keith Lehrer’s counterexample to Clark’s analysis of knowledge.

Argument: Suppose that Smith forms the belief that “Someone in the office owns a Ford” based on an inference from his belief that Jones owns a Ford and his belief that Brown owns a Ford, both of whom are in his office.  However, since Brown does not own a Ford, one of Smith’s grounding beliefs is false. However, Smith still knows that someone in the office owns a Ford even though his grounds include a false belief.

Upshot: Clark’s requirement that a belief be fully grounded is not necessary for knowledge.

Step 9: Goldman maintains that “causal chains” may include, not only admixtures of causes  and inferences, but also causes and logical connections.

Argument: Smith forms the belief that (p), “Someone in the office owns a Ford”, based on the conjunction of the belief that (q) “Jones owns a Ford” and  (r) “Jones works in his office”.  Given that (q) and (r) logically entail (p), and given that (q) and (r) combine to cause Smith’s belief that (p),  it follows that there is a causal connection between the fact that (p) and Smith’s believing that (p).

Upshot: If X and Y are logically related, and Y is a cause of Z, then X also counts as a cause of Z.

Step 10: Goldman argues that since his causal analysis of knowledge does not require that a knower be able to justify or provide evidence in favour of the proposition known it is able to account for cases of knowledge that the traditional analysis excludes.

Argument: S may know that Abraham Lincoln was born in 1809 based on a warranted inference from an Encyclopaedia entry even after S has forgotten how he came to learn this fact, so long as there is a causal connection between S’s belief and the fact that Lincoln was born in 1809.

Upshot: The justification requirement of the traditional analysis is not necessary for knowledge.

CONCLUSION:

Goldman concludes that S knows that p just in case the fact that p is casually connected in an “appropriate” way with S’s believing that p, where “appropriate” causal processes include: perception, memory, inferences (in which a proposition counts as warrant-providing only if it is true), or some combination of the aforementioned. Goldman concedes that the preceding causal analysis is not part of the meaning of the expression “S knows p”.  However, he maintains that it specifies the necessary and sufficient conditions for knowledge.

DISCUSSION QUESTION:

1. Do you agree that the agent in Goldman’s volcano example does not have knowledge? Why or why not?

2. What does Goldman mean when he says that S knows that p only if S’s believing that p is causally connected to the fact that p? What are the implications of this view for knowledge about the future?

3. Do you agree that S knows that p only if S’s believing that p is causally connected (in Goldman’s sense) to the fact that p? Why or why not?

4. Does Goldman’s analysis of knowledge successfully avoid Gettier-style counterexamples? Why or why not?

Outline of Clark's "Knowledge and Grounds"

SYNOPSIS:

Michael Clark argues that the necessary and sufficient conditions for S knowing that p are:

(1) p is true,

(2) S believes that p,

(3) S is justified in believing that p, and

(4) S’s belief that p is fully grounded.

OUTLINE:

STEP 1: Clark observes that Gettier’s examples are stronger than they need to be since he presents cases in which an agent’s grounds for a belief deductively entail the proposition believed, whereas there may also be Gettier-type cases involving non-deductive grounds.

Example: Suppose that Smith forms the true belief that Jones owns a Ford based on the testimony of Brown, who is typically reliable, but who (on the present occasion) has mixed up Jones with someone else. Although Smith’s belief (based on Brown’s testimony) is true and justified, it fails to constitute knowledge because it rests on false grounds. Insofar as Brown’s testimony constitutes non-deductive grounds for believing that Jones owns a Ford, this is a Gettier-style example involving non-deductive grounds.

Upshots: (i) Insofar as Brown’s testimony constitutes non-deductive grounds for Smith’s belief, Gettier-style examples are effective within a non-deductive context. (ii) S knows that p only if her belief that p is not based on false grounds.

STEP 2: Clark argues that adding the fourth condition that an agent’s belief be based on true grounds is still insufficient for knowledge.

Example: Suppose that Smith forms the true belief that Jones owns a Ford based on the testimony of Brown, who is typically reliable, but who (on the present occasion) is simply making up a claim that (unknown to Brown) happens to be true. Although Smith’s belief is true, justified, and rests on true grounds (i.e., Brown’s claim that Jones owns a Ford), it still fails to constitute knowledge. Thus, we have a Gettier-style example involving true grounds.

Upshot: Although Smith’s belief is based on true grounds, Smith lacks knowledge because the grounds on which he accepts these true grounds, viz. that Brown knows them, are false.

STEP 3: Clark claims that if we keep inquiring about someone’s grounds for accepting certain grounds, and the grounds for accepting those grounds, and so on, we eventually come to a place where it would be inappropriate to ask for further grounds.

Example: A series of questions posed to Smith about why he believes Jones owns a Ford will eventually culminate in an answer, which we would deem out of order to question further.

Upshot: There is always some finite list of grounds that can be offered for a belief.

STEP 4: Clark posits that if one of the grounds in the chain of supporting grounds for a certain belief is false, then that belief fails to constitute knowledge.

Example: If any of the beliefs in the chain of Smith’s grounds for believing Jones owns a Ford are false, we may deny that Smith has knowledge. If all the beliefs in the chain are true, then Smith’s belief that Jones owns a Ford is fully grounded.

Upshot: Being fully grounded is a necessary condition for a belief to constitute knowledge.

CONCLUSION:

Clark concludes that if we add the requirement that S’s belief that p be fully grounded to the requirement that p be true, S believe that p, and S’s belief that p be justified, we arrive at conditions that are jointly sufficient for knowledge.

DISCUSSION QUESTION:

1. Is Clark’s analysis of knowledge a successful response to Gettier-style counterexamples? If not, can you recommend any changes that would improve Clark's account?

Gettierising Nozick

Over at The Web of Belief, I lay out a Gettier case against local reliabilism (which is itself based on an earlier post of mine, Undescriminating Reliabilism). Let's call my proposed Gettier case against local reliabilism GLR. The version of local reliability I took as my target was admittedly quite simple, so here I will attempt to apply GLR to a much more sophisticated version of local reliability—namely, Nozick’s counterfactual account. Nozick [1981, p. 179] argues that the true belief that p is knowledge only if:

(N1) If p were true, then S would believe that p
(N2) If p were not true, then S would not believe that p¹

Following David Lewis [1973] we may say that (N1) and (N2) are true, as counterfactual statements, iff in possible worlds near to the actual world, if p is true, S believes that p, and if p is false S does not believe that p. Nozick recommends that we assess (N1) and (N2) by reference to what is the case in all nearby possible worlds. Roughly, a world may be described as ‘nearby’ if it is only slightly different from the actual world and ‘distant’ if it is radically different.

The following reply to GLR seems available to Nozick. Let us suppose that the set of nearby possible worlds include ones in which the computer is running a different program or is missing altogether. In such nearby worlds, although there is a red cube in the box, there is no hologram of a red cube. Since in such a world the subject would not believe that (a) although (a) is true, (N1) has not been satisfied.

One initial difficulty with this reply is that it is not immediately clear that worlds in which the computer is running a different program or is missing altogether should be considered nearby. But let us, for the sake of argument, assume that such worlds are in fact nearby.

Even so, Nozick’s strategy for responding to GLR proves too much, since it also impugns cases in which the subject intuitively has knowledge.

The ‘Jesse James’ Counterexample:
Consider the case of the Jesse James Bank Robbery, as described by Craig [1990]:

Jesse James, the reader will recall, is riding away from the scene of the crime with his scarf tied round his face just below the eyes in the approved manner. The mask slips, and a bystander, who has studied the ‘wanted’ posters, recognises him. The bystander now knows, surely, that it was James who robbed the bank. But Nozick has a problem: there is a possible world, and a ‘close’ one, in which James’ mask didn’t slip, or didn’t slip until he was already past the bystander; and in that world the bystander wouldn’t believe that James robbed the bank, although it would still be true that he did. So Nozick’s condition [N1] is not satisfied, and he is threatened with having to say that the bystander doesn’t know that it was James, even though the mask did slip. So his analysis looks like ruling out something which is as good a case of knowledge as one could wish for [p. 22].

Nozick is already equipped with a reply to the ‘Jesse James’ counterexample, akin to that employed in the Grandma case [See endnote 1]. In brief, he may simply argue that in worlds in which the mask did not slip, the method employed by the bystander would be different. Thus, given the version of Nozick’s counterfactuals revised to include the subject’s method, worlds in which the mask did not slip would not be included in the relevant nearby possible worlds.

However, to the extant that this reply is effective in preserving the bystanders knowledge in the Jesse James case, it is also effective at preserving the GLR subject’s ‘knowledge’ that (a). In GLR, the computer program, as the locus of reliability, constitutes part of the method by which S arrives at her belief that (a). Thus, by (N1*), all worlds in which the computer program is different or the computer is missing, a different method is being used and that world eo ipso fails to count as nearby.


Endnote:

1 Counter examples such as the Grandma case has prompted Nozick [1981, p. 179] to revise (N1) and (N2), limiting them to the same method (i.e., vision):

(N1*) If p were true, S (using M) would believe that p

(N2*) If p were not true, then S (using M) would not believe that p

References:

Craig, E. [1990], Knowledge and the State of Nature. Oxford: Clarendon Press.

Lewis, D. [1973], Counterfactuals. Oxford: Basil Blackwell.

Nozick, R., [1981], Philosophical Explanations, Cambridge: Cambridge University Press.

Un-discriminating Reliabilism (Part 2)

In [Goldman 1976], Goldman rejects the idea that knowledge requires evidence or justification, and instead conceives of knowledge as true belief that satisfies a reliability condition. On this view, reliability is what you add to true beliefs, instead of evidence or justification, in order to get knowledge. However, in [Goldman 1979], Goldman came to hold the view that instead of replacing the notion of justification in an account of knowledge, reliability is what makes a belief justified. This shift in Goldman’s thought is indeed significant, since it represents his move towards a full-fledged J-externalism. But there is an equally important (earlier) shift that takes place in [Goldman 1976]; namely, Goldman’s rejection of the causal account advocated in [Goldman 1967] and his embracing of the idea that knowledge requires certain discriminative abilities, a view he motivates with his now famous barn case (pp. 772-773). An analysis of the barn case leads Goldman to conclude:

A person knows that p, I suggest, only if the actual state of affairs in which p is true is distinguishable or discriminable by him from a relevant possible state of affairs in which p is false. If there is a relevant possible state of affairs in which p is false and which is indistinguishable by him from the actual state of affairs, then he fails to know that p. (Ibid, 774)

Notice that the above discrimination requirement avoids the unfavourable consequences of the Gettier case described in my previous post in at least two ways. First, it omits all talk of justification and so (in at least one sense) traditional Gettier problems never arise. Second, since according to the aforementioned Gettier case S is unable to discriminate between the actual red cube and the hologram, her true belief does not constitute knowledge (thereby preserving our common-sense intuitions regarding such cases). But, and this is crucial to the point I’m getting at, what is doing the actual work in [Goldman 1976]’s reply to Gettier is the discrimination requirement, and not the notion of reliability a la (J-Rel). Admittedly, in [Goldman 1976], Goldman defines reliability in terms of a subject being able to discriminate between relevant alternatives and thus the former is taken to entail the latter. However, in later formulations of the reliability requirement, explicit appeals to discriminative abilities are increasingly omitted, and Goldman eventually comes to define a reliable process as more or less one that tends to produce true beliefs. Moreover, I do not believe this shift in Goldman’s definition of reliability is by accident. Presumably, talk of discriminative abilities falls out of Goldman’s analysis (now taken to be unnecessary) once he shifts to an explicit J-externalism, where what makes a belief justified need not be internally available to the subject.

The objection may be raised that even with the J-externalist turn in Goldman’s thinking, the discrimination requirement remains an implicit part of his account. I will not deny that this may be the case. Even so, it has long been recognised that the discrimination requirement can stand on its own, independent of any appeals to reliability (see for example [McGinn 1984]). Thus, if my claim that what is doing the work vis-a-vis Gettier is actually the discrimination requirement, rather than reliability qua reliability, then this is at least sufficient to establish that when it comes to the aforementioned Gettier case, reliability a la (J-Rel) is at best explanatorily redundant and at worst explanatorily inert. (The second more unpropitious option holds if you consider my arguments in (Part 1) persuasive.) Moreover, even if one were to reject the account of the evolution of Goldman’s thought outlined above, my more general conclusions still stand: (1) that though reliability and discrimination are classically taken to go together this need not be the case and (2) if (DR) is doing the heavy lifting in classic reliabilist replies to Gettier then the widely held assumption that (J-Rel) constitutes a successful reply to Gettier may be unwarranted.

References:

Goldman, A. (1967), 'A Causal Theory of Knowing', Journal of Philosophy 64: 355-372.

Goldman, A. (1976), 'Discrimination and Perceptual Knowledge', Journal of Philosophy 73: 771-791.

Goldman, A. (1979), 'What is Justified Belief?' Justification and Knowledge, Dordrecht: Reidel.

Un-discriminating Reliabilism (Part 1)

J-reliabilism is perhaps the most well-known, if not most widely held, response to Gettier’s criticism of the JTB account of knowledge. However, in this post (as well as in the one to follow), I argue that J-reliabilism is insufficient for responding to Gettier. Furthermore, I argue that what is really doing the work in classical formulations of reliabilism, such as that found in [Goldman 1976], is the discrimination requirement, or (DR):

(DR) For any subject S, if S knows that p then S can distinguish the actual situation in which p from all relevant or nearby alternative situations in which ~p.

Recall, J-reliabilism amounts to following claim:

(J-Rel) For any subject S, S’s belief that p is justified IFF it was formed via a reliable process (i.e., a process that tends to produce true beliefs).

According to (J-Rel), reliability is what makes a belief justified. (see [Goldman 1979]). By defining justification in terms of reliability, J-reliabilists hope to eliminate that element of luck that allows our beliefs, under the traditional JTB account, to be Gettiered. However, it is not clear that (J-Rel) is sufficient for resisting Gettier type cases. For example, consider the following perceptual Gettier case:

Suppose S has strong perceptual evidence for, and comes to believe, the proposition:

(a) There is a red cube in the box on the table.

Now, it so happens that there is in fact a red cube in the box on the table, though the cube is being obscured from S’s visual field by some sort of barrier. Furthermore, the box is rigged up to a computer which projects a visual hologram of a red cube in the box. However, the computer is programmed to only project the hologram of the red cube in the box when there is a real red cube in the box. Moreover, S lacks any of this background information, and forms her belief that (a) purely on the basis of the hologram of the red cube. All of the following seem true in the above case:

(i) (a) is true
(ii) S believes (a) is true
(iii) S’s belief that (a) is formed via a reliable process

Ex hypothesi, (iii) is true since the computer is programmed to only project the hologram of a red cube when there is an actual red cube present (one may build in whatever stipulations one likes, such as that the computer is eternal and infallible in its operation etc.). Thus, S’s belief that there is a red cube in the box is reliable since the process by which the belief was formed would, given the computer’s programming, tend to produce true beliefs. However, I believe this represents a bona fide Gettier case since, though S has a justified (i.e., reliably formed) true belief, we wouldn’t say that she has knowledge.

I take the above Gettier case to show that mere reliability is insufficient for eliminating the element of luck from S’s belief that (a). To wit, (J-Rel) fails to eliminate the element of luck associated with S’s belief that (a) since the fact that her belief is reliable is (from the subject’s perspective) itself merely a matter of luck. Thus, we may say that it is lucky that S’s belief was formed via a reliable process. Significantly, the particular species of epistemic luck here described is easily eliminated by making the reliability of the belief forming process internally available to S. Thus, if S knew that what she was seeing was merely a hologram, but that the hologram was itself reliably linked to the presence of an actual red cube in the box, then our intuitions would grant that S does in fact know that there is a red cube in the box. Thus, I see the above Gettier case as corroborating the internalist intuition that that which justifies S’s beliefs should be internally available to her.

Notice that I haven’t made any references to S being able to discriminate between actual red cubes and holograms of red cubes, nor to definitions of reliability that appeal to relevant alternatives (See [Goldman 1986]). This failure to invoke (DR) is in keeping with my ultimate goal of showing that what is doing the real work in classic reliabilist accounts is not reliability qua reliability, but rather an implicit commitment to (DR). Above, I have argued that reliability alone [i.e., (J-Rel)] is insufficient for responding to Gettier. In my next post, I argue that the mistaken assumption that (J-Rel) and (DR) must go together has misled reliability advocates to impute to J-reliabilism an explanatory power that it actually lacks.

References:

Goldman, A. (1967), 'A Causal Theory of Knowing', Journal of Philosophy 64: 355-372.

Goldman, A. (1976), 'Discrimination and Perceptual Knowledge', Journal of Philosophy 73: 771-791.

Goldman, A. (1979), 'What is Justified Belief?' Justification and Knowledge, Dordrecht: Reidel.

Looking for a Causal Criterion for Vision

Below, I attempt to derive (what may be described as) the standard causal criterion for vision [see (3)*** below] from a critical examination of a non-causal definition of vision. By so doing I hope to make explicit some of the motivating intuitions that lie behind the causal account. The non-causal definition of vision with which I will like to begin may be schematically expressed as follows:

S sees an object O before him only if

(1) S has an experience E such that it looks just as if O is before him
(2) There is an object O before S

However, this definition hardly seems sufficient for saying that S sees O before him. For example, we may construct something akin to a Gettier-type counterexample to (1) and (2). (I say that the following example is merely ‘akin’ to a Gettier case because unlike the genuine Gettier counter-example mine is not concerned with the question of justification.) Imagine S has his eyes closed and is hallucinating that there is an object O before him. Suppose further that there actually is an object O before him. In such a case, the requirements of both (1) and (2) have been met, and yet we would not wish to say that S sees O before him since S would continue to have the experience E even if O was not present. This observation has motivated a causal account of perception that stipulates that S only counts as seeing O if E is caused by O. Thus we arrive at:

(3) S’s experience E is caused by O.

As it stands, the stipulation that E is caused by O does not seem sufficient since we can imagine instances when such causal dependence holds but in which we would still not count S as seeing O. Grice presents us with two such scenarios [Grice 1961, 141-5]. The first case is where S has his eyes closed and another person, P, describes the object O to S, which then prompts S to have a hallucination of O. In such a case, S’s experience E is caused by O, albeit indirectly via P’s description. Since E is ultimately causally dependent on O, the demands of (3) have been met. Yet, we would still be unwilling to say that in such a scenario S sees O. Thus, as it stands, the addition of (3) is still not sufficient for defining a case of vision.

One possible reply to Grice’s first example would be to point out that it is highly unlikely that P’s description of O could give rise to a hallucination in S that perfectly matched O. Thus, if we added a clause to (3) requiring that E perfectly match O we would (it is supposed) exclude cases of hallucinations based on the description of an object. Thus we arrive at something like:

(3)* S’s experience E is caused by O and E perfectly matches O.

At least three points can be made with regards to the above reply to Grice’s first example. First, it is not clear that it is impossible for P to describe O in enough detail for S to have a hallucination of O that fulfilled the requirement of (3)*. Admittedly, this may be difficult to do depending on how complex O happens to be, but it certainly isn’t (strictly speaking) impossible. In fact, it may even turn out that the perfect match between E and O may simply be a matter of luck, this would still be sufficient to fulfill the requirement of (3)*, so long as E is (indirectly) caused by O. Moreover, we could imagine a case in which O is a simple geometric shape such as a triangle or square. In such a case it wouldn’t seem all that difficult for P to describe O in such a way that S’s hallucination of O perfectly matched O.

Second, (3)* does not seem like a plausible necessary condition for vision since there are many cases in which we may be willing to say that S sees O although E does not perfectly match O. For example, let us suppose that S sees a white shaggy dog, which he mistakes for a sheep. In such a case, S’s experience E may not perfectly match O and yet it is not clear that we would wish to deny that S sees O. Other examples include misreading the letters (or words) on a street sign and certain cases of blurred vision. Moreover, there is at least one reading of ‘perfectly matches’ in which it wouldn’t make sense to say any experience E perfectly matches (or in any way matches) O. On such a reading experiences may be taken to be a very different kind of thing from the objects that give rise to them. While I’m not particularly keen on pushing for such an interpretation of experience vis-à-vis objects, it does seem to be a viable position one may hold.

Our third and final objection to (3)* comes from Grice’s second counterexample. Grice imagines a scientist who looks at an object O and stimulates S’s brain in exactly the way necessary for S to have an experience E that matches O. If instead of a scientist we substituted a sophisticated computer, then the possibility of E perfectly matching O perfectly (at least as perfect as we could reasonably expect in a normal case of vision) becomes quite plausible. Given this fact, along with the fact that E remains causally dependent on O, the demands of (3)* have once again been met. Nevertheless, we would still be unwilling to say that in such a scenario S sees O. Thus, the addition of (3)* still leaves us with an insufficient causal criterion for vision.

Another way one may object to Grice’s examples is by pointing out that in both the causal dependence of E on O is indirect. If we stipulate that the causal dependence must be direct we arrive at:

(3)** S’s experience E is caused by O directly.

The word ‘directly’, as used in (3)** suggests that there cannot be any mediating factors between O and E. However, this does not seem like a plausible requirement for vision since there are cases where O may cause E indirectly where we would nevertheless still wish to insist that S sees O. For example, imagine that S is looking at bacteria on a slide with the aid of a powerful microscope. In such a case, E’s causal dependence on O is mediated (i.e., by the microscope) and yet we may still wish to insist that S sees O.

Although the requirement that O directly cause E seems problematic, it does hint at a valuable intuition: namely, that there are a number of possible causal scenarios linking E to O that would be unacceptable as far as our putative understanding of vision is concerned. According to this line of reasoning our concept of vision includes the notion that there should be a certain type of causal relationship between perceptual experiences and their objects. Thus, for S to see O, his perceptual experiences must be caused in the right sort of way, where this is understood as excluding all causal chains that deviate from ‘the norm’. What constitutes ‘the norm’ can be arrived at easily enough by examining unproblematic instances (i.e., paradigm cases) where the locution ‘S sees O’ correctly applies. Examples, such as Grice’s discussed above, represent causal chains that deviate from the norm and therefore fail to be causally relevant in the right way. If we add the requirement excluding deviant causal chains to (3) we arrive at:

(3)*** S’s experience E is caused by O in the right sort of way.

Admittedly, what the expression ‘right sort of way’ entails needs to be spelt on in greater detail. Nevertheless, (3)*** is regarded by many to reflect the canonical statement of the causal requirement of vision and is taken, along with (1) and (2) to be sufficient for establishing that S sees O.

The causal criterion we have derived represents merely the first step, rather than the end, of the debate regarding whether the causal description of vision is the best available. For example, many believe disjunctivism represents an attractive alternative to the causal picture, though it remains unclear whether the two approaches are necessarily incompatible. Moreover, while (1), (2) and (3)*** are here taken to be sufficient for vision, another question worth considering is whether they also represent a necessary condition.