Brie Gertler

The Mechanics of Self-Knowledge

Philosophical Topics 28 (2002): 125-46

It is often said that we can know our own thoughts more directly or with more certainty than anyone else can know them. And this disparity is usually taken to be principled, in that we would not be the rational, reflective beings that we are without it. My aim is to trace the consequences of a principled disparity between self-knowledge and other-knowledge for what may be termed the “mechanics” of self-knowledge”. I use a new thought experiment to show that if introspective states are merely causally related to introspected thoughts, the disparity between self-knowledge and other-knowledge is not truly principled. An account of self-knowledge adequate to a truly principled disparity will allow that thought tokens can be included in self-attributing judgments about them.

I. The Cartesian Intuition

My argument takes seriously the following intuition about self-knowledge. At least sometimes, while you reflect on an occurrent thought, you know what it is that you are thinking with ~~a greater degree of~~more certainty than anyone else could ~~possibly know it~~have regarding your thought. Moreover, at times you know this without using evidence about the so-called “external” world; and your knowledge does not depend on contingent facts about that world. This claim is qualified in important ways. First, it applies only to occurrent thoughts, thoughts that are actually entertained at the moment they are self-ascribed, and not to standing (dispositional) states. Second, it concerns only thought contents and not attitude modes— (such as believing, desiring, or intending—) towards those contents. (This fits with the previous ~~point~~ qualification in that attitude modes are often construed dispositionally.) Third, the claim applies only to creatures that are sufficiently rational and reflective. Finally, the claim allows that one can make false self-attributions, and that one can remain unaware of one’s own mental states. Despite the fact that some attribute weightier claims to Descartes, I will call the intuition at hand ‘the Cartesian intuition’. It has three components.

1. Knowledge of one’s own occurrent thoughts can at times achieve a level of certainty higher than that which knowledge of others’ thoughts (or, equivalently, others’ knowledge of one’s own thoughts) could achieve.

2. One can at times know one’s own occurrent thoughts without investigating the external world.

3. (1) and (2) are true of rational, reflective creatures in all possible worlds.[1]

While most philosophers—including content externalists, who are thought to face a particular challenge regarding privileged access—accept components (1) and (2) of the Cartesian intuition, some who accept (1) and (2) reject (3). For instance, Warfield claims that externalism and privileged access are incompatible ~~only if~~unless “every possible world in which externalism is true is a world in which individuals do not have privileged self-knowledge”.[2] ~~(Warfield 1997, 233)~~ Others assume that a view accommodates privileged access so long as it allows that subjects enjoy privileged access in the ~~neighborhood~~ actual world and nearbyof possible worlds ~~centered on the actual world~~. By contrast, component (3) of the Cartesian intuition entails that the epistemic disparity separating self-knowledge from other-knowledge holds across all possible worlds. To put this point in a particularly Cartesian way: I enjoy privileged access even in a world where “some malicious demon of the utmost power and cunning has employed all his energies in order to deceive me”.[3] ~~(Descartes 1984, 15; AT 22)~~ I defend this intuition further below.

Note that a methodological difference between self-knowledge and other-knowledge is neither necessary nor sufficient for the epistemic disparity expressed in (1)-(3). An A methodological difference is not required by the epistemic disparity ~~does not itself require a difference in methods~~: perhaps a single method is used in detecting mental states, but this method is more reliable when applied to the person using the method than it is when applied to others. And the claim that I can determine my own states by using a method which you cannot use (to determine my states) is consistent with the epistemic parity between these methods, and even with the epistemic superiority of yours.

I will argue that the epistemic disparity expressed in the Cartesian intuition is, in fact, due to a methodological difference. Specifically, I will argue that methods of introspection compatible with causal accounts of self-knowledge cannot do justice to the ~~intuition~~disparity. If the Cartesian intuition is correct, the relation between a thought content and an introspective state isn’t merely a causal relation. And I will show that one alternative to a causal account, the inclusion account popularized by Burge ~~(1988)~~ and Heil ~~(1988)~~, holds promise as a way to capture the principled disparity between self-knowledge and other-knowledge.[4]

II. Causal Accounts

Many philosophers maintain that a subject’s privileged access to her mental states reflects the distinctive causal (or functional) relations between those states and her beliefs about them. I will call such views, which take subjective privilege to derive purely from causal relations between introspected and introspective states, “causal accounts” of self-knowledge. (Since functional properties are second-order causal properties, as I use “causal property” and “causal account” these include functional properties and accounts which exploit them.) Causal accounts hold special appeal for those who individuate mental states by their functional roles, for functionalists usually define introspective beliefs partially by their tendency to be appropriately caused by the lower-order states they concern.

Those who individuate mental states by their functional roles or other relational features typically reject epistemically internalist standards of justification for self-knowledge. According to epistemically internalist standards, a belief is justified only if one grasps one’s evidence for that belief. As Boghossian points out, a state’s functional role is a relational feature of it, and “you cannot tell by mere inspection of an object that it has a given relational or extrinsic property.”[5] ~~(Boghossian 1989, 15)~~ Grasping a relationally individuated state on the basis of non-relational evidence one grasps requires inference. So if self-knowledge required that one grasp one’s evidence, then self-knowledge of relationally-defined state tokens would be inferential. And, Boghossian concludes, if all self-knowledge is inferential then the internalist requirement that we know our evidence traps us in a regress: for any evidence one has that one is in a particular state is itself mental, and hence (on these accounts) relational, and hence known only inferentially. Relational accounts of intentional content thus appear to preclude internalistically justified self-knowledge.

Boghossian’s point applies to causal accounts of self-knowledge generally, regardless of whether such an account is combined with relational individuation conditions for thoughts. For causal accounts take self-attributing beliefs to be justified by the causal processes from which they result. Hence the internalist would require, for knowledge, that one grasp that the self-attributing belief ~~was a result of~~resulted from appropriate causal processes. Because this grasping would involve inference, the internalist requirement would yield the inferential regress Boghossian describes. Those who endorse causal accounts of self-knowledge thus favor reliabilist standards for self-knowledge. To enjoy privileged access, the subject must be in a better position to detect her own thoughts than others are. But, the reliabilist holds, even one who enjoys privileged access may lack internalist justification, for she need not recognize the distinctive causal relations which underwrite this access.

Let us grant that the reliabilist route to accommodating self-knowledge allows introspective beliefs to be non-inferential (and thereby avoids Boghossian’s regress). Still, to accommodate the Cartesian intuition it is not enough to secure non-inferential self-knowledge. We must also show that there is a principled epistemic disparity between self-knowledge and other-knowledge. To illustrate the problem here, I offer a new thought experiment featuring a familiar character, the neuroscientist of the future. The thought experiment shows that, if purely causal relations between introspective and introspected states suffice for non-inferential self-knowledge of relationally-defined states, the process of inference which others currently use to determine one’s thoughts is at most practically necessary. That is, the causal account which provides for reliable non-inferential self-knowledge also provides for equally reliable non-inferential other-knowledge, in principle. The causal account thus places self-knowledge epistemically on a par with other-knowledge, in principle, and so cannot accommodate the Cartesian intuition.

Nora is a neuroscientist of the future. She is also Nick’s sponsor in Alcoholics Anonymous. To ensure that she can offer moral support to Nick the moment he begins to crave his drink of choice, a gin martini, Nora takes some radical steps. She conducts a thorough analysis of Nick’s brain and identifies the brain states which realize his “gin” thoughts. Nora may pick out Nick’s gin thoughts by their functional role, their representational content, their type-identity to a neurophysiological state, their syntax in a LOT, etc. The thought experiment requires only that gin thoughts have some physical realizers which an advanced neuroscience could identify. Nora also conducts a similar analysis of her own brain and identifies the states which realize her “Nick is thinking about gin” beliefs. She then programs this information into a monitoring device which she attaches to both Nick’s brain and her own.

The device performs two operations. First, it constantly scans Nick’s brain for the “gin” realizer states. (To detect gin thoughts with wide content, Nora can monitor Nick’s environment as well; this may be unnecessary since Nick’s “twin gin” thoughts would presumably be of equal concern.) Whenever Nick occurrently tokens the state associated with the content “gin”, the monitor sends a signal to Nora’s brain which causes it to realize the occurrent belief “Nick is now thinking about gin”. (If being occurrent is a matter of degree, suppose that the device sends its signal whenever his standing disposition to engage in gin-seeking behavior is sufficiently close to being actualized.) Second, to avoid false alarms, the monitoring device constantly scans Nora’s brain for the “Nick is now thinking about gin” state. When it detects that Nora is on the verge of realizing this state, it blocks the realization until it confirms that Nick is, at that time, thinking about gin.[6] We may suppose that the device operates very quickly.

Nora eventually trusts her “Nick is now thinking about gin” beliefs completely. When it occurs to her that Nick is thinking about gin, she acts on this belief without hesitating or subjecting the belief to any special scrutiny. After the device has operated for a while, the way that she comes to occurrently believe that Nick is thinking about gin appears epistemically similar to the way she comes to have other occurrent beliefs. Compare Nora’s coming to believe that Nick is thinking about gin, on the one hand, with her suddenly noticing that the room is dark, on the other. Reliabilist standards apply to each belief (“Nick is thinking about gin”, and “the room is dark”). Roughly, each qualifies as knowledge if and only if it is the result of an appropriate causal chain. In the former case, the relevant causal chain is the proper functioning of the device; in the latter, it is the proper functioning of her perceptual system (perhaps the room is bright but ~~she is losing~~ her eyesight is failing). And Nora typically forms beliefs of these types without reflecting on her evidence; she typically makes no attempt to reconstruct the causal chain (the stages in the device’s operation, or the perceptual process) which led to ita belief.

Now according to causal accounts of self-knowledge, Nick’s access to his gin thought is exhaustively explained by the causal process which resulted in his self-attributing belief. But Nora’s causal access to Nick’s gin thought appears to parallel Nick’s causal access to it, epistemically speaking. Both Nora’s Nick-attributing belief and Nick’s self-attributing belief are products of causal chains embodied in properly functioning systems. And neither belief requires, for its warrant, that the believer investigate the causal process leading up to it. Of course, Nick’s belief system is natural, whereas Nora’s device is artificial. But this difference in naturalness isn’t itself an epistemic difference.~~; and t~~ The artificial quality of Nora’s device seems not to affect Nora’s reliabilist warrant for believing that Nick is now thinking about gin:~~. For~~ Nora’s belief that Nick is thinking about gin appears to meet the epistemic requirements of reliabilism, since the device establishes an appropriate counterfactual-supporting link between Nick’s gin thoughts and Nora’s belief. And Nora’s method of detecting Nick’s gin thoughts is no more prone to error ~~about Nick’s gin thoughts~~ than is Nick’s own method ~~himself~~. Adding a bit of detail to the story, Nora’s method may in fact be more reliable on this score than Nick’s. Her interest in Nick’s abstinence gives her a motivation to be sensitive to his gin thoughts, while Nick may have developed a habit of repressing his gin thoughts, in an attempt to avoid acting on them.[7]

Nora’s belief that Nick is now thinking about gin appears to be as warranted as, and to require no more inference than, Nick’s self-attributing belief, according to the causal account. So if that account is correct, it seems that one can, in principle, grasp someone else’s occurrent thoughts with no more inference than one uses in grasping one’s own, and just as reliably. But then self-knowledge is not distinguished from other-knowledge by being less (or non-)inferential or by ~~being~~ resulting from a more reliable method. The causal account of self-knowledge thus conflicts with the Cartesian intuition with which we began.

I will consider three main avenues of response to this argument: (A) Nick’s method is more reliable than Nora’s after all; (B) the argument overlooks the role of intensions; (C) the case conflicts with facts about personal identity.

Response A: Nick’s method is more reliable than Nora’s after all.

Probably the most attractive response on behalf of the causal account is this.

Even if Nora is right about Nick’s gin thoughts more often than Nick is, Nick’s method of detecting his own gin thoughts is more reliable than Nora’s method of detecting them. Reliability depends not just on what is actual but also on counterfactual possibilities. In worlds nearby the world in which Nora’s device operates as described, Nick and Nora exist but Nora’s device is not in place. In those nearby device-less worlds, Nick is correct about his gin thoughts more often than Nora is; hence Nick’s method is more reliable ~~about his gin thoughts~~ even in the world where the device does operate. So Nick does enjoy privileged access, for—by virtue of these counterfactual truths—at least some of his self-attributing beliefs have greater warrant than any of Nora’s Nick-attributing beliefs.

This defense of the causal account safeguards privileged access only for some possible worlds, namely, worlds not surrounded by worlds in which Nora’s device operates. It therefore conflicts with component (3) of the Cartesian intuition, according to which the sort of epistemic privilege involved in self-knowledge doesn’t depend on which world is actual. An account of privileged access accommodates (3) only if it ensures that Nick enjoys privileged access in every world, including the world which is at the center of a large neighborhood of worlds in which Nora’s device operates as described. This response will not ensure privileged access in that world.

Of course, the appeal of the causal account may override, for some, the appeal of component (3). Let me then further motivate component (3). Identify an occurrent thought—perhaps a thought about ice cream. With proper self-reflection, you can feel quite sure that you are thinking about ice cream. Could you even conceivably be equally sure that someone else was thinking about ice cream? The intuition that you could not, even by using Nora’s device, is the basis for component (3). It suggests that you needn’t be free of intrusive Nora’s device in order to enjoy privileged access to your ice cream thought. Perhaps most importantly, it suggests that regardless of whether Nora is using her device to monitor your ice cream thoughts, nearby worlds needn’t be device-less. Even at the center of a large neighborhood of worlds in which Nora is using the device, you still enjoy privileged access to the ice cream thought.

The best tack for one who accepts the causal account is simply to reject (3). As explained above, one can consistently reject (3) while maintaining (1) and (2). That is, one can accept that, since Nora’s device is not operating in the actual world or in worlds nearby, self-attributing beliefs are formed by a process that is more reliable than the process of forming other-attributing beliefs; yet facts about the external world can block privileged access, and so privileged access is not present in all possible worlds. This position is consistent with maintaining that, in those worlds in which subjects do enjoy privileged access, self-knowledge does not require investigating the external world.

While this is the most promising position for a proponent of the causal account, it comes at a price. Ifirst, it entails that ~~even the most basic self-attributing beliefs such as those involved in Descartes’~~ ~~cogito~~ fall short of absolute certainty, since certainty would require ruling out that one occupies a world in which privileged access is lacking. Denying the possibility of certainty will be a problem for anyone who finds that, at the moment at which she engages in introspective reflection, she cannot doubt that she is entertaining a given content. (At least, it will be a problem if she takes this inability to doubt as reflecting epistemic certainty rather than as a merely psychological inability.) Certainty aside, denying component (3) entails that the epistemic disparity between self-knowledge and other-knowledge is not principled: we do enjoy a limited first-person privilege, but only as a matter of luck. That is, it entails that even when you self-attribute a thought content under ideal conditions, at the moment at which you are reflecting on an occurrent thought, your self-attributing belief is no more certain than another person’s belief that you are thinking a particular thought (or your belief that another person is thinking a particular thought) could be. ~~In addition~~Second, rejecting (3) limits the criteria available to individuate persons. If, as many philosophers maintain, personal identity is constant across worlds, then rejecting (3) means that particular epistemic relations between states do not determine personal identity; any such relations can hold between states of distinct persons.[8]

Accepting component (3) of the Cartesian intuition is not as costly as it may appear. Most notably, it does not require accepting a “K-K” principle—the epistemic principle which says that one knows that p only if one knows that one knows that p—, for it allows that some persons with self-knowledge may not know that they have self-knowledge. It requires only the much more limited claim that we can, in principle, know our own thoughts with more certainty than anyone else could possibly know them.

The causal account has the consequence that, in a world at the center of a neighborhood of worlds in which Nora’s device operates, Nick’s access to his gin thought is epistemically on a par with Nora’s access to it. So the causal account entails that there is no principled epistemic disparity between self-knowledge and other-knowledge.

Response B: The argument overlooks the role of intensions.

The Nick and Nora case shows that, if the causal account is correct, Nora’s belief about Nick’s gin thought can be result from a process which is precisely as reliable and (non-)inferential as the best process available to Nick~~’s own belief about his gin thought~~. Given the modality imposed by component (3) of the Cartesian intuition, a principled disparity between self-knowledge and other-knowledge will invoke a relation which one can bear only to one’s own states. Intensional relations are obvious candidates for such a relation.[9] Here is my own version of this strategy for reconciling privileged access with the causal account.

True introspection requires that one picks out the introspected state (or its bearer) via a reflexive indexical. Just as you are singularly able to refer to yourself using the reflexive indexical “I”, subjects are singularly able to pick out their own states using reflexively indexical concepts (or terms in a LOT). Clearly, this is due to the intensional force of reflexive indexicals; there is no possible world in which a reflexive indexical “I” picks out someone distinct from the person using the indexical. So the intensional features of Nora’s “Nick is thinking about gin” belief distinguish it from self-attributing beliefs. And they do so in principle, since Nora’s belief necessarily lacks the reflexivity, present in Nick’s “I am thinking about gin” belief, which true introspection requires.

If this strategy is to work, the intensional difference which reflexivity underwrites must be epistemically significant. Otherwise, it cannot explain a principled epistemic disparity between self-knowledge and other-knowledge. And this is a serious obstacle: the claim that purely indexical differences between beliefs are epistemically inert lies behind Boghossian’s worry that the sort of self-knowledge compatible with externalism is as “cognitively insubstantial” as the purely indexical knowledge of location expressed by “I am here now”.[10] ~~(Boghossian 1989, 17-20)~~

I believe that the causal account fails to allow for an epistemically significant indexical difference between Nora’s “Nick is now thinking about gin” belief and Nick’s “I am now thinking about gin” (or “there is an occurrent gin thought which is mine”) belief. According to the causal account, any epistemically significant feature of the relation between introspective and introspected states is exhausted by the causal features of these states. This means that if the intensional features of Nick’s self-attributing belief are epistemically salient, his belief refers to his gin thought by virtue of standing in a particular causal relation to that thought. To be sure, only Nick can accurately formulate the belief in a way which marks that this belief belongs to the very subject the belief concerns (“I am now thinking about gin”). But no epistemic consequences follow from the fact that only Nick can accurately express his gin thought with “I am now thinking about gin”. For on the causal account, others’ beliefs can have an epistemically equivalent, albeit intensionally different, relationship to Nick’s gin thought.

To see this, first recall a familiar point about indexical reference. The referent of an indexical term—such as “now”, “here”, or “I” —is partially determined by the context in which the term is used. For instance, my current utterance of “tomorrow” has a determinate referent—Friday, say—only because it is anchored to the day of utterance, Thursday. As ~~Perry~~ ~~(1979)~~ ~~and~~ Lewis ~~(1979)~~ and Perry express this point, all indexical reference trades on reflexive relations: my utterance “tomorrow” is reflexively related to Thursday and thereby refers to Friday.[11] (Of course, the utterance doesn’t refer to Thursday; the reflexive relation is not, in this case, a referential relation.) The Lewis-Perry point pertains to any sort of indexical reference, regardless of whether the referential vehicle is linguistic or mental.

Applied to our case, the fact that indexical reference trades on reflexive relations means that, if Nick’s introspective belief indexically refers to his gin thought, this reference depends on a reflexive element in the introspective belief. The causal account maintains that any epistemically salient relation between introspective belief and introspected thought must be exhausted by these states’ causal features. Assuming that reflexive reference is epistemically salient, the reflexive element in the introspective belief effects indexical reference to the gin thought by virtue of a causal relation to the gin thought. So Nick’s indexical reference to his gin thought results from a reflexive element in his self-attributing belief, together with a causal link to the gin thought, loosely as follows. Nick indexically refers to his gin thought by virtue of that thought’s being “the cause of this”, where “this” picks out the self-attributing belief. (Nick needn’t think of his gin thought as “the cause of this”, of course. That phrase merely expresses the causal and reflexive relations responsible for effecting indexical reference to the gin thought.)

Now once Nora’s device is in place, her belief can also indexically refer to Nick’s gin thought, via a reflexive element in her own belief and the belief’s causal link to Nick’s thought. That is, Nora’s reference to Nick’s gin thought can equally result from that thought’s being “the cause of this”, where “this” picks out Nora’s belief that Nick is thinking about gin.

The intensional difference between Nick and Nora lies not in how they secure Nick’s gin thought as a referent but in Nick’s additional entitlement to think of the gin thought as a state of his (as he would put it, as “mine”). On the purely causal account, this entitlement is epistemically irrelevant. Compare: when you and I both view my reflection in a mirror, the reflection causes each of us to have a visual image of a person who is 5’5”. The fact that only I am entitled to think of the shared cause as a (bodily) state of mine is epistemically irrelevant: there is no principled disparity between my knowledge of my height and your knowledge of it.[12]

If my argument is sound, the causal account specifically denies the epistemic salience of the intensional difference between Nick’s self-attributing belief and Nora’s Nick-attributing belief. For the causal account limits the epistemically salient features of the relation between Nick’s gin thought and his self-attributing belief to causal features of that relation. And the thought experiment shows that these causal features can be paralleled by the relation between Nick’s gin thought and Nora’s belief that he is thinking about gin. It appears, then, that response B fails: no epistemically salient intensional difference distinguishes the connection between Nick’s “I am thinking about gin” belief and his gin thought, on the one hand, from Nora’s “Nick is thinking about gin” belief and Nick’s gin thought, on the other.

Response C: The case conflicts with facts about personal identity

The third response is as follows.

Non-inferential access, or reliability above a certain threshold, partially determines personal identity. So the portrayal of Nora as having a non-inferential, highly reliable ~~access to~~method of detecting one of Nick’s states is incoherent.

This response would rule out the Nick and Nora case in principle. But in order to do so, it must presuppose that the scenario is impossible. The scenario appears possible, however; we can conceive of Nora’s device operating as described.

If Nora’s device is conceivable, then causal theorists must concede that at least component (3) of the Cartesian intuition is mistaken. Otherwise, they are committed to a choice between two unpalatable options. If anyone who is merely capable of non-inferentially knowing Nick’s states qualifies as (identical with) Nick, then the possibility of Nora’s device implies that everyone capable of using it is identical with everyone upon whom it could be used. This might well mean that all persons together constitute a single subject. If, on the other hand, one is identical with Nick just in case one actually knows one or more of his states with as little inference or as much reliability as he does, then employing Nora’s device fuses these previously distinct characters, rendering them identical. Neither option seems acceptable.

While the causal account of introspection can accommodate non-inferential, highly reliable self-knowledge, the Nick and Nora case shows that the highest epistemic standard for self-knowledge which the causal account can accommodate is one which, on that account, other-knowledge can also meet. So the causal account is unable to capture the principled disparity between self-knowledge and other-knowledge.

III. The Inclusion Account

An alternative to the causal account of self-knowledge has recently achieved prominence, as a ~~tool~~ ~~for demonstrating~~ means of illustrating that content externalism is compatible with privileged access. According to content externalism’s definitive thesis, a state’s intentional content logically depends upon the subject’s causal relations to things in the environment external to her. Externalism does not presuppose that mental states are individuated by causal roles, for externalist content need not supervene on causal relationships among internal states and external facts. Instead, it may be only partially determined by such causal relationships. Perhaps for this reason, some externalists have rejected the causal account of self-knowledge. The alternative they have embraced construes the relationship between introspected and introspective thoughts as not merely causal but, rather, as a relationship of inclusion: introspective thoughts embed or contain introspected thoughts. In the current section I explain the inclusion account and illustrate its promise as a way to accommodate the Cartesian intuition. I close by briefly considering whether externalists can exploit the account.

The Inclusion Account Described

Burge has argued, influentially, that externalism can accommodate at least “basic self-knowledge”, since introspective judgments embed or include the wide content of introspected thoughts. He describes the inclusion relationship as follows.

One is thinking that p in the very event of thinking knowledgeably that one is thinking it. It is thought and thought about in the same mental act.[13] ~~(Burge 1988, 116; the last sentence is repeated nearly verbatim in Burge 1996, 244)~~

The content of the first-order (contained) thought is fixed by nonindividualistic background conditions. And by its reflexive, self-referential character, the content of the second-order judgment is locked (self-referentially) onto the first-order content which it both contains and takes as its subject matter.[14] ~~(Burge 1988, 122)~~

On Burge’s view, the introspected thought content is included in (“contained” by) the content of the introspective judgment, and hence the external factors which determine the former also partially determine the latter. Burge’s account differs from the causal account—which, recall, is a merely causal account—by employing the inclusion relation to explain privileged access. The inclusion ~~relation~~ account identifies one relatum with a proper part of another and claims that privileged access depends upon this inclusion relation, whereas the causal account maintains that a causal relation between introspected and introspective states ~~stand in~~ ~~only~~ ~~a causal relation, and that this causal relation~~ suffices to explain privileged access. So the inclusion account is strictly inconsistent with the causal account.

The inclusion account can certainly allow that introspected states causally influence introspective states. But it denies that the introspected-introspective relationship is exhausted by such causal influence. Compare: we might say that the instability of a single table leg causes the instability of the table as a whole. Yet the leg is part of the table, and so the relationship between the table leg and the table isn’t merely causal. This analogy does not commit the inclusion account to construing thoughts as objects in any robust sense; as is clear from the first passage quoted, for Burge the inclusion relation holds between events. And events are also divisible into proper parts which, in a sense, causally contribute to the larger event. Consider a shortstop catching a line drive. This event may causally contribute to a double play, a larger event of which the catch was a proper part. As in the table case, the shortstop’s catch isn’t merely causally related to the double play.

Did Burge intend the inclusion strategy as a competitor to the causal account? A competing interpretation of the above-quoted passages sees the relation between higher-order judgment and lower-order thought as purely causal, and as underwriting a proper-part relation only between content types. This alternative reading construes the relation between the introspected token and the introspective token as analogous to the relation between an original document and a photocopy of one of its pages. The photocopy causally inherits types of marks, and—supposing that the copier creates no additional marks in the copying process—the types of marks on the photocopy are a proper part of the types on the original multi-page document. In this case, the proper-part relationship holds between types of marks, but ~~a merely causal relationship holds~~not between tokens: the presence of token marks on the ~~original~~ photocopy causally depends on the presence of ~~partially~~ ~~causes~~ ~~there to be~~ numerically distinct yet type-identical marks on the ~~photocopy~~original. To be sure, since these tokens are type-identical they share logical features such as semantic content. But they are logically independent in that no change in one logically necessitates a change in the other. There is at most a causal dependence between them. By positing a merely causal ~~relation~~ dependence between tokens, this interpretation assimilates Burge’s position to the causal account.

Interpreting Burge’s position as a version of the causal account blocks it from accommodating the Cartesian intuition, given my arguments above. Since my primary interest here is not exegetical, this is reason enough to put the causal interpretation aside. Still, it’s worth noting that the token-inclusion interpretation is more loyal to Burge’s statements than the ~~merely causal,~~ type-inclusion interpretation is. First, while sharing a content-type with an introspected state might allow an introspective state to refer to the introspected content in a loosely “self-reflexive” way, this story does not explain how the introspective state thereby becomes, as Burge describes it, “second-order”. A difference in orders rests on the embedding of content tokens and not merely the overlap of content types. Second, his commitment to a not-merely-causal ~~relation~~ dependence between introspected and introspective states is reflected in his claim that “in the case of cogito-like judgments, the object, or subject matter, of one’s thoughts is not contingently related to the thoughts one thinks about it.”[15] ~~(1988, 120)~~ This logical and not merely nomological ~~link between the~~dependence of the judgment ~~and~~ on its object is required to explain why “[n]o errors at all are possible in strict cogito judgments; they are self-verifying.”[16] (~~ibid~~~~., 121)~~ The token inclusion interpretation is a more accurate construal of Burge’s view than the causal interpretation, and—given my arguments above—holds more promise as a way to capture the Cartesian intuition.

The Inclusion Account Defended

If the inclusion relation obtains between content tokens and not merely between content types, the inclusion account can secure a truly principled disparity between self-knowledge and other-knowledge. I will argue for this conclusion in two steps. First, I will show that the metaphysics of the inclusion relation ~~safeguards~~ shields the inclusion account from the modal difficulties which beset the causal account. Unlike the causal account, the inclusion account yields a metaphysical disparity between Nick and Nora significant enough to accommodate all three components of the Cartesian intuition. My second task is to show that the inclusion relation between tokens can underwrite a genuinely epistemic and not merely metaphysical disparity between self-knowledge and other-knowledge. My treatment of the epistemic consequences of the inclusion account will fall short of an exhaustive epistemology of inclusion. I will, however, provide grounds for believing that the inclusion account has the epistemic resources to avoid the problems which the causal account faced.

(a) Metaphysics

Simply put, the metaphysical disparity between Nick and Nora derives from Nick’s singular ability to be in a state which embeds his “gin” token. Nora’s monitoring device may result in her reliably instantiating a state which embeds a “gin” token, one type-identical to Nick’s, whenever Nick thinks about gin. But only Nick himself can have a belief which embeds his “gin” token. This follows from two claims: whoever instantiates an embedding (higher-order) token also instantiates the (lower-order) token it embeds; and no occurrent mental token belongs to more than one person. The former claim is part of the metaphysics of inclusion: tokening a belief entails tokening anything it embeds. The latter claim is a highly plausible assumption.[17] And even if (as some proposals for individuating persons maintain) an occurrent token could belong to an entity that qualifies as two persons in some sense (perhaps by virtue of a prior fusion or a future fission), sharing a token would render these persons partially identical. The Cartesian intuition requires only that subjects have access to their states that is privileged relative to other—i.e., distinct—persons. Since Nick and Nora are wholly distinct, Nora cannot have a belief which embeds Nick’s thought token.

On the inclusion account, the definitive introspection relation is one which can obtain only between the subject’s own states, in principle. There is no possible world in which Nora’s belief (that Nick is now thinking about gin) embeds Nick’s gin thought, whereas there are possible worlds, and neighborhoods of them, in which Nora’s relevant belief-forming processes are as causally sensitive to Nick’s gin thought as Nick’s are. Unlike the causal account, the inclusion account makes for a metaphysical disparity between self-attributing beliefs and other-attributing beliefs which is constant across worlds. The inclusion account thus avoids the difficulty with component (3) of the Cartesian intuition on which the causal account foundered.

Finally, this principled difference can ground a referential difference. Nick can refer to his gin thought with a logically direct reflexive, whereas Nora cannot. My response to the objection from intensional differences showed that the causal account precludes Nick’s introspective belief from referring to his thought content with a logically direct reflexive; it can refer to that content only via a causal indexical of a type which Nora can also use. By contrast, the “embedding” relation allows for reference to the content token that is logically and not merely causally direct, since the referring state contains (embeds) its own referent. This is in line with Burge’s claim that the second-order thought is “locked onto” the first-order thought “self-referentially”. So the metaphysical difference between Nick and Nora grounds a principled difference in how they can refer to Nick’s gin thought. I now turn to the epistemic consequences of these metaphysical and referential differences.

(b) Epistemology

In considering the prospects for satisfying the Cartesian intuition, it is important not to overestimate what that intuition implies. It does not imply that Nick is always aware of his gin thoughts, or that he is infallible about them. He may repress, and so fail to notice, some gin thoughts. And he may falsely believe that he’s occurrently craving gin, for instance when his therapist mistakenly tells him that he is. Because of these possible shortcomings, Nora may be correct about whether Nick is thinking about gin more often than Nick himself is. The Cartesian intuition implies simply that Nick has available a means of self-knowledge which can yield greater certainty than any means of knowing Nick’s mental states which Nora could have.

I have argued elsewhere ~~(forthcoming)~~ that the proper-part relation between an introspective belief and a phenomenal token explains the directness and authority of phenomenal self-knowledge.[18] In brief, the proper-part relation allows the embedding state to constitute direct demonstrative attention to the embedded content. This demonstrative attention serves a dual role. It is a semantic demonstrative, securing its referent (the phenomenal content) for the subject without any mediating description, by means of the basic judgment “it is thus [here, now]”. And the attention justifies the belief, since it renders the subject aware of the phenomenal content.

Because that account employs the qualitative features of the embedded content as the objects of demonstrative attention, it cannot be applied to states without essential, defining qualitative features. If propositional attitude contents do have essential, defining qualitative features—rather than, as Lormand ~~(1996)~~ contends,[19] merely incidental qualitative accompaniments—then direct demonstrative attention to qualitative features of propositional attitude contents, underwritten by the “proper part” relation between tokens, holds promise as an explanation of privileged access to occurrent propositional attitudes. Most philosophers deny that occurrent propositional attitude contents have essential, defining qualitative features. (However, this denial has recently been challenged~~; see especially Siewert 1998~~.[20]) Obviously, I cannot address this issue here. I will not assume that propositional contents have essential qualitative features, let alone defining ones.

Without invoking qualitative features of propositional attitude contents, how can we accommodate the Cartesian intuition? Recall Burge’s claim that, in introspection, a thought content is “thought and thought about in the same mental act”. The inclusion account succeeds only if the very event of thinking a thought with a certain content can be part of the event of knowing that one is occurrently thinking a thought with that content. This suggests that the inclusion account fits with a model of introspective awareness as attending to an event, where the event of attending subsumes the event—the occurrent thought—attended to. (Attention is selective, of course, and it may be beyond one’s abilities to attend to all of one’s current thoughts. In addition, attention may alter the contents of one’s thought, so that the thought one attends to may differ from the thoughts one had prior to exercising attention. Partially for this reason, the Cartesian intuition does not claim that we can ever achieve certainty about past thoughts.)

On this model, introspective attention is crucially dissimilar from perceptual attention. In attending to an occurrent thought content, the state which constitutes introspective awareness of the content subsumes the target of the attention, viz., the propositional attitude content token. By contrast, perception is a causal process, involving perceptual states which are logically independent of their targets. Introspection may involve causation in various ways: you may be caused to attend to your occurrent thoughts, and you may be caused to have some thoughts rather than others. More to the point, the basic relation between introspective awareness and introspected content may be causal in the sense that catching a baseball may partially cause a double play, while the catch partially constitutes the double play. But unlike perceptual states, introspective states embed or include their targets, introspected thoughts.

Of course, merely attending to an event doesn’t yield knowledge of it; knowledge further requires that the target is conceptualized in some substantive way. As Boghossian ~~(1989)~~ observes, if self-knowledge doesn’t involve any robust descriptive concepts it is no more substantial than the knowledge that “I am here now”. One can know this without knowing, in any robust sense, where one is.[21] This is the toughest nut to crack in accommodating the Cartesian intuition: identifying concepts which are sufficiently descriptive to yield substantial beliefs, but which are also sufficiently transparent to yield certainty when applied in appropriate ways by rational, self-reflective creatures. My reason for thinking that there are such concepts is the Cartesian intuition itself: my knowledge that I’m now thinking that writing requires concentration (to borrow an example from Burge) is more certain than any knowledge I could possibly have of others’ thoughts, regardless of what sort of science fiction scenario we posit.

I can only gesture towards the sort of concepts which could play the role required by the epistemology of inclusion. First, skepticism about whether there are concepts which satisfy the above constraints is somewhat mitigated by the fact that the epistemic disparity is claimed only for rational, self-reflective creatures. This qualification indicates that the concepts exercised in introspecting propositional attitude contents may be those implicated in the capacity for rational self-reflection.[22]

Since I’m not making any assumptions about the relationship between propositional attitude contents and phenomenal contents, I cannot assume that the concepts exercised in introspecting propositional attitude contents (call these “intentional concepts”) will be the same as, or similar to, phenomenal concepts. In fact, I believe that intentional concepts are importantly different from phenomenal concepts. Still, we will make progress towards solving the present problem if we can show that, just as fully grasping pain requires reflecting on some of one’s own occurrent phenomenal contents, fully grasping intentional concepts requires reflecting on some of one’s own occurrent propositional attitude contents. Arguably, a grasp of phenomenal concepts such as pain is constitutively tied to the ability to correctly apply it under certain circumstances. (Could one who truly had the concept pain fail to classify an excruciating headache as painful, instead counting it as a tickle?)

The parallel with intentional concepts would be this: a grasp of intentional concepts requires first-person reflection on thoughts. Reflecting on one’s own occurrent thoughts yields an adequate intentional concept (such as a concept of the content “gin”) only if it puts one in a position to apply this concept correctly under particular, highly circumscribed (ideal) conditions. That is, intentional concepts may be “epistemically individuated” in Peacocke’s sense: “individuated, partly or wholly, in terms of the conditions for a thinker’s knowing certain contents containing those concepts”. ~~(Peacocke 1999, 13)~~[23] So if intentional concepts are epistemically individuated, they are individuated in such a way that, if certain conditions obtain, one has privileged self-knowledge. Given the argument of Section III, the conditions adequate for self-knowledge aren’t those specified by the causal account. My suggestion is that the conditions for privileged self-knowledge are, instead, those partially specified by the inclusion account. Intentional concepts would then be individuated in a way which entailed that one who possessed an intentional concept would apply it correctly in ideal conditions, where one such condition is that the subject is in a belief state which embeds that thought. (The other conditions would include rationality, sufficient attention, etc.)

If an account of intentional concepts along these lines can be satisfactorily developed—and I have not demonstrated that it can—then the inclusion account will yield a principled epistemic disparity between self-knowledge and other-knowledge. While Nora’s method of attributing gin thoughts to Nick involves causal factors which vary across possible worlds, Nick has available a method of self-attributing gin thoughts which, when used, yields true beliefs in every possible world. This is because the following conditional is necessarily true: the presence of an embedding token entails the presence of any token it embeds. The process Nick~~’s distinctive access~~ uses to detect his “gin” token, when in fact he introspects it, is perfectly reliable, for when he uses that process the content token is itself a component of the self-attributing belief. Nick may, in fact, never introspect any of his gin thoughts. But if he were to have a truly introspective belief, that is a belief which conformed to the inclusion account, he would also have the relevant content token.

This way of using the inclusion account to distinguish self-knowledge from other-knowledge may seem to involve some sleight of hand. For the presence of the self-attributed state is built into the inclusion relation, and so the account seems to provide ~~merely~~no more than a trivial guarantee of the truth of beliefs which exemplify this relation. To see the threat of triviality here, note that it is possible to specify a method M whereby Nora can know Nick’s states with perfect reliability. One need simply stipulate that Nora’s Nick-attributing belief is formed via M only when Nick is, in fact, thinking about gin. The perfect reliability of M follows trivially from this stipulation.

The problem here is obvious. Given how method M is defined, the fact that Nora’s belief is formed via M is outside her ken. A genuine epistemic difference between self-knowledge and other-knowledge appears to requires that the inclusion relation is, by contrast, within the subject’s ken. Now if introspective self-knowledge requires an awareness of the inclusion relation, it will lead to a familiar skeptical regress. Suppose that awareness that token a embeds token b involved a belief, Cc, to the effect that a does embed b. Then, if Cc is to ensure certainty, the subject must also be aware of her grounds for Cc: this in turn involves a further belief, Dd, to the effect that Cc is justified. Thus begins the e~~se steps spark a~~ ~~familiar~~ skeptical regress.

The inclusion account fits nicely with one traditional way of blocking the regress of justification, viz., claiming that some beliefs are justified without reference to other beliefs. The inclusion account says that some self-attributing beliefs are justified by the presence of the content they embed. Since what does the justifying is part of the justified belief itself, the justificatory regress is blocked. Burge seems to invoke this idea when he describes “basic, cogito-like judgments” as “self-verifying”. Again, the claim is not that having a thought ensures that one is aware of it; obstacles to self-awareness may be pervasive. The claim is rather that under certain conditions, one’s knowledge that one is thinking a particular thought can attain greater certainty than anyone else’s knowledge of that fact could ever attain. If any beliefs can serve as epistemic bedrock, beliefs about one’s own mental states are prime candidates.[24] And the inclusion account explains how introspective beliefs can block the regress of justifying states: for the reflexivity in the introspective belief effectively collapses the justifying state and the content of the belief it justifies.

Another epistemic consequence of the inclusion account deserves notice, namely, that the inclusion account also avoids the inferential regress described by Boghossian. That regress is not the more general skeptical regress, which is yielded by the idea that every belief must be justified by another belief. Instead, Boghossian’s regress is generated by the special need to infer a relational fact (such as: I am appropriately causally related to H₂O rather than to XYZ) from non-relational evidence (one’s “inner” state). The inclusion strategy avoids the need for inference on this basis: if the self-attributed thought is embedded in the self-attributing judgment, and the judgment can qualify as a grasp of its component content (the “gin” thought) by virtue of reflexive reference to it, then the justifying state is precisely as relational as the state which it justifies belief in. For the justifying state ~~just~~ is one and the same as the attributed state.

Finally, we must ask: does the inclusion account resolve content externalism’s apparent problem with privileged access? Externalists have focused on the metaphysics of the inclusion account, in an effort to show that higher-order states inherit the wide content of lower-order states. While some externalists combine the metaphysics of inclusion with a reliabilist epistemology, Burge instead takes introspective judgments to be logically immune from error. Perhaps for this reason, he limits the robustness of introspective knowledge, claiming that such knowledge does not secure “an ability to explicate correctly one’s thoughts or concepts via other thoughts and concepts”. It ensures merely that one “understand[s] what one is thinking well enough to think it.”[25] ~~(Burge 1988,125)~~ Those who believe that externalism is incompatible with privileged access claim that we don’t, in fact, understand any wide thought content “well enough to think it”. For this reason, we are not able to truly use them in thinking.

In the end, whether the inclusion account is compatible with externalism will depend on whether we do understand any wide thought content “well enough to think it”. The availability of the inclusion account, whereby higher-order thoughts logically inherit the content of lower-order thoughts, relocates the issue about whether externalism can accommodate privileged access. The issue no longer centers on the relation between introspective and introspected states, but on content at a single level. Is wide content truly the subject’s own thought, a content which she truly grasps? This is now the crucial question regarding externalism and privileged access.

Many epistemological issues remain unresolved, and the inclusion account must be filled out by a thorough epistemology of introspection. An epistemology of introspection would explain, most importantly, how we introspectively conceptualize propositional attitude contents. I have provided only a sketch of the metaphysical and epistemic grounding for privileged self-knowledge which the inclusion strategy makes available.

IV. Conclusion

I have argued that the causal account of introspection is at odds with the Cartesian intuition, specifically with the claim that the epistemic disparity between self-knowledge and other-knowledge is constant across worlds. The alternative I favor, the inclusion account, has the metaphysical resources to accommodate the principled disparity between self-knowledge and other-knowledge. The inclusion account succeeds only if an intentional (propositional attitude) content can justify a belief which embeds it. I ~~believe~~ have sketched reasons for thinking that it can. But the development and defense of what may be called the epistemology of inclusion must await another occasion.

Alternatively, one’s allegiance to the causal account, and pessimism about the prospects for working out the epistemic details of the inclusion account, may trump one’s commitment to the Cartesian intuition. I have shown that in accepting a causal account one incurs a high cost. One must concede that ~~even the most basic self-attributing beliefs fall short of absolute certainty; that, certainty aside,~~ the epistemic disparity between self-knowledge and other-knowledge is not principled;, and that privileged access to states cannot anchor individuation conditions for persons. This price seems to me intolerably steep. There is thus a pressing need for a comprehensive account of how we know our own occurrent propositional attitude contents. If my reasoning here is sound, this account cannot be a version of the causal account, but must instead exploit the metaphysics of inclusion.[26]

~~WORKS CITED~~

~~Bermudez, Eilan, and Marcel, eds. (1998)~~ ~~The Body and the Self~~~~. Cambridge, MA: MIT Press.~~

~~Boghossian, P. (1989) “Content and Self-Knowledge”.~~ ~~Philosophical Topics~~ ~~17:5-26.~~

~~Burge, T. (1988) “Individualism and Self‑Knowledge”.~~ ~~The Journal of Philosophy~~ ~~85:649‑63. Reprinted in Ludlow and Martin (1998), 111-27.~~

~~Burge, T. (1996) “Our Entitlement to Self-Knowledge”.~~ ~~Proceedings of the Aristotelian Society~~ ~~96:91-116. Reprinted in Ludlow and Martin (1998), 239-63.~~

~~Clark, A. and Chalmers, D. (1998) “The Extended Mind”.~~ ~~Analysis~~ ~~58:7-19.~~

~~Descartes, R. (1984) “Meditations”. In Cottingham, Stoothoff, and Murdoch, trans.,~~ ~~The Philosophical Writings of Descartes~~~~, vol. II. Cambridge: Cambridge University Press.~~

~~Gertler, B. (forthcoming) “Introspecting Phenomenal States”.~~ ~~Philosophy and Phenomenological Research~~.

~~Heil, J. (1988) “Privileged Access”.~~ ~~Mind~~ ~~97:238-51. Reprinted in Ludlow and Martin (1998), 129-45.~~

~~Lewis, D. (1979) “Attitudes~~ ~~De Dicto~~ ~~and~~ ~~De Se~~”. ~~The Philosophical Review~~ ~~88:513‑43.~~

~~Lormand, E. (1996) “Nonphenomenal Consciousness”.~~ ~~Noûs~~ ~~30:242-61.~~

~~Ludlow and Martin, eds. (1998)~~ ~~Externalism and Self-Knowledge~~~~. Stanford, CA: CSLI Publications.~~

~~Lycan, W.G. (1996)~~ ~~Consciousness and Experience~~~~. Cambridge, MA: MIT (Bradford).~~

~~Peacocke, C. (1999)~~ ~~Being Known~~~~. Oxford: Oxford University Press.~~

~~Perry, J. (1979) “The Problem of the Essential Indexical”.~~ ~~Noûs~~ ~~13:3-21.~~

~~Shoemaker, S. (1988) “On Knowing One’s Own Mind”,~~ ~~Philosophical Perspectives~~ ~~2:183-209. Reprinted in Shoemaker (1996), 25-49.~~

~~Shoemaker, S. (1996)~~ ~~The First-Person Perspective and Other Essays~~~~. Cambridge: Cambridge University Press.~~

~~Siewert, C. (1998)~~ ~~The Significance of Consciousness~~~~. Princeton: Princeton University Press.~~

~~Warfield, T. (1997) “Externalism, Self-Knowledge and the Irrelevance of Slow Switching”.~~ ~~Analysis~~ ~~57:282-4. Reprinted in Ludlow and Martin (1998), 231-4.~~

~~Wittgenstein, L. (1953)~~ ~~Philosophical Investigations~~~~. New York: Macmillan Publishing Co.~~