1. Introduction

2In what follows, in the first instance I will reconstruct Shoemaker’s regress argument and roughly sketch how philosophers have tried to account for its conclusion. Then, I will present the two aforementioned issues facing Shoemaker’s reasoning which need to be addressed if one wants to accept the argument’s relevance for a theory of self-awareness. I will discuss each of these problems and how they might affect the relevance of Shoemaker’s regress. This task has yet to be done to date, but it is not a trivial one.

2. Shoemaker on Self-Identification

3Sydney Shoemaker (1968) once famously wrote:

It is clear, to begin with, that not every self-ascription could be grounded on an identification of a presented object as oneself. Identifying something as oneself would have to involve either (a) finding something to be true of it that one independently knows to be true of oneself, i.e., something that identifies it as oneself, or (b) finding that it stands to oneself in some relationship (e.g., being in the same place as) in which only oneself could stand to one. In either case it would involve possessing self-knowledge – the knowledge that one has a certain identifying feature, or the knowledge that one stands in a certain relationship to the presented object – which could not itself be grounded on the identification in question. This self-knowledge might in some cases be grounded on some other identification, but the supposition that every item of self-knowledge rests on an identification leads to a vicious infinite regress. (Shoemaker, 1968, p. 561, emphasis added)

1. Identifying something as oneself (henceforth “self-identification”) requires beliefs about relational or non-relational properties of oneself (which may pick out oneself uniquely).

2. If all beliefs about oneself were gained through self-identification, then all beliefs about oneself would be based on an infinite number of further beliefs about oneself.

3. As a finite being, one can have only a finite number of beliefs about oneself.
   _________________________________________________
   Not all beliefs about oneself can be gained through self-identification.

I could not, for all properties P, know that I have P by inference from a known identification of myself with an object O that is P. In order to know that I=O, I need to know that O has a property – by hypothesis distinct from P – only I have. [… I]f all such knowledge of my properties depends on knowledge of such identifications, in knowing myself to have even one property, I must already know myself to have infinitely many. So there must be a regress-stopping property we non-inferentially know ourselves to have. […] The relation I have in mind is what Russell called acquaintance. It is not a straightforward representation relation. (Williford, 2006, pp. 3-4)

6I agree that the most plausible way to deal with Shoemaker’s regress is to rely on something given in experience. However, granting that the sketched regress argument has some plausibility, there might be at least two worries attached to it. The first issue – which I call the scope problem – addresses the argument’s far-reaching consequences. The second – let’s call it the implausible constraints problem – is concerned with the truth of one of its premises. After quickly presenting each worry, I will discuss whether they should be regarded as a threat to Shoemaker’s proposal.

3. The Scope Problem

3.1. The Problem

7With regards to the first problem, if we accept – as Shoemaker apparently did – that identification requires some descriptive beliefs about the objects identified, then this has consequences for theories of first-person thought as well as theories of singular thought in general. In order to avoid a regress of beliefs, I need to have non-identificatory awareness not just of some properties of myself, but also of anything that can be a constituent of an identification. Given the definition in Section 2, to identify, for instance, a perceived object p as another object o that I am already familiar with, I need to have at least two descriptive beliefs: one about p and another about o. However, these cannot be based on other identifications, since this would require even more descriptive beliefs, and so on. In addition, we can construe very similar non-first-personal cases to the one presented in Section 2: Let’s suppose that I am seeing this person in a mirror who has a face that looks very similar to Peter’s face and conclude that it must be Peter whom I am seeing. In that case I already need to have some prior beliefs about what Peter’s face looks like that lead me to believe that Peter is identical with the person I am seeing in the mirror. The question is how I came to know what Peter’s face looks like in the first instance. I could not have acquired all of these beliefs about Peter through identification (as defined above), as each belief would have required more beliefs about Peter, which in turn would have been in need of further beliefs for which we can pose the same question again. Hence, it is easy to construe Shoemaker-esque regress arguments for all those things which can be part of an identification. Here is just one example of how we might offer a similar argument to that presented in the previous section:

1. Identifying something as Peter (henceforth “Peter-identification”) requires beliefs about relational or non-relational properties of Peter (which may pick out Peter uniquely).

2. If all beliefs about Peter were gained through Peter-identification, then all beliefs about Peter would be based on an infinite number of further beliefs about Peter.

3. As a finite being, one can have only a finite number of beliefs about Peter.
   _________________________________________________
   Not all beliefs about Peter can be gained through Peter-identification.

8Thus, Shoemaker’s regress argument turns out to imply very general conclusions about singular thought, rather than stating a specific problem for theories of self-awareness and first-person thought. This is not necessarily a bad thing, but it provides much more space for criticism. There are quite a few contemporary philosophers – such as Bach (1987, p. 12), Burge (1977), and Recanati (2010) – who believe, following Bertrand Russell (1910), that genuine singular thought is based on some immediate epistemic access to the objects thought of (see also Jeshion 2010). However, this need not concern us here. The important point seems to be that if Shoemaker’s regress argument has these general implications for singular thought, then it apparently also requires a solution on a more general level. If there are good reasons to believe that all genuine singular thoughts are in need of some sort of non-identificatory acquaintance, then this is no longer unique to first-person thought. Thus, we may be inclined to think that what Shoemaker was able to point out is nothing that specifically needs to be addressed when speaking about self-awareness and first-person thought, but rather is something for which a more general theory of singular thought has to account.

3.2. Discussion of the Problem

9It is debatable whether the scope problem should be regarded as a real threat to Shoemaker’s argument for non-identificatory self-awareness. After all, his argument is not supposed to vindicate the assumption that only first-person thought depends on some non-identificatory awareness. Hence, it is not a problem for his argument per se. However, it is still relevant to the general importance of the argument and the question of who should be concerned about it. Thus, the central question is if the demand for this non-identificatory awareness is not tied to first-person thought, why should it be specifically addressed in a theory of first-person thought?

10One way to end this discussion could be to hint at the general work on singular thought and by relying on a division of labor in philosophy and science. This would be a relatively common and uncontroversial step to take. When a biologist is investigating a specific type of mammal, say dogs, his peers do not assume that he is obliged to say much about mammals in general except that there is something specific to dogs that distinguishes them from most other things that fall under the category mammal. I think similar normative presuppositions underlie philosophical discourse, and for good reasons. No one is expected to offer a theory of everything; one is allowed to exclude things that are not specific to the field studied. Thus, I suppose that one could prima facie think that the scope problem justifies the view that a theory of first-person thought need not address Shoemaker’s regress.

4. The Implausible Constraint Problem

4.1. The Problem

12Another, more fundamental problem with Shoemaker’s argument might be that arguably there are forms of identification that are not in need of any descriptive beliefs. If this is so, then self-awareness might be based on those forms of identification without involving an infinite number of further beliefs. In this respect, we might even argue – pace Shoemaker – that all cases of self-awareness ultimately rest on identification. Thus, the question is whether Shoemaker places implausible (overly demanding) constraints on the concepts of self-identification and identification.

13There certainly is some form of identification involved in tracking objects over time. Yet, this ability might not be grounded in the capacity to entertain descriptive beliefs as there are many creatures which are able to track objects over time but which do not have the conceptual tools to entertain the right sort of beliefs. When a cat observes and subsequently chases a mouse, she does not need and possibly is not able to think this mouse under the concept of a mouse or as this grayish thing in front of me. Recanati (2012, p. 87) proposes, albeit for different reasons, that something like this is quite common among human beings as well. He calls those cognitive performances “immediate recognition” and contrasts them with what he labels “slow recognition”.

14This distinction is not just supported by philosophical armchair reasoning. An empirically informed account of visual perception, which might prompt some reservations about Shoemaker’s claims, is Pylyshyn’s Visual Indexing Theory (2003; 2011). According to this theory, the recognition of an object’s identity in visual perception is foremostly independent from the ability to ascribe relational or non-relational properties to these objects. Pylyshyn believes that basic visual reference over time is in this sense directly demonstrative rather than satisfactionally descriptive. Pylyshyn conducted a series of experiments that putatively support the existence of what he calls Visual Indexes or FINSTs (Fingers of INSTantiation). The so-called Multiple Object Tracking (MOT) experiments are intended to demonstrate that humans’ early vision uses indexes which function in accordance with demonstratives familiar from linguistic communication. These indexes contain no descriptive content and ultimately acquire their reference through causal relations to external objects.

15During the MOT experiments, test subjects are confronted with a small number of target objects (usually four) which are briefly distinguished from a couple of visually identical non-target objects by short blinking. Subsequently, both sets of objects arbitrarily move around and the test subjects are asked to track the target objects during their period of moving. The result is that the subjects are quite successful in performing this task. Pylyshyn concludes that this is best explained by his theory of visual indexes since the seen objects look exactly alike (e.g. in color, size, and shape) and therefore have no distinguishing properties. Moreover, Pylyshyn undertook some variations of these initial experiments that further support this conclusion. If the representation of properties were to play a particular role in visual tracking, then one would expect the presence of more salient properties to enhance the performance of object tracking (e.g. being faster or more accurate). However, a variety of MOT experiments have shown that this does not happen. Furthermore, when the target objects change some of their non-relational properties, the test subjects very often do not notice this change but still manage to identify the objects afterwards.

16Nevertheless, one might still assume that the observers represent and use the location of each target object to determine serially where the objects have moved. After the recording was made at the beginning, they just see which object is closest to the initial location they have documented and do so for the remainder of the test. This is a possibility which cannot be ruled out experimentally. Yet, Pylyshyn tried to exclude this theoretical option by simulating this way of keeping track of the identity of objects. The issue with this way of tracking is that the simulations have shown that even under almost perfect conditions, this method would be much more prone to errors compared to the average of humans who actually performed the MOT experiments. Hence, it is quite reasonable to conclude that humans do not necessarily track perceived objects in virtue of representing properties (Pylyshyn 2011, pp. 36-37).

4.2. Discussion of the Problem

18I think that the arguments provided in Section 4.1 are quite persuasive and suggest that there are forms of identification which do not require any descriptive representations about their relata, as Shoemaker apparently thought. It is not entirely clear whether Shoemaker merely wanted to make claims about identification involved in first-person thinking or rather had something stronger in mind and intended to pose a regress for all kinds of self-awareness (including non-conceptual ones). In the passage mentioned at the beginning of this article, he is just speaking about self-knowledge, yet the title “Self-Reference and Self-Awareness” and later passages in his paper suggest that the regress was set up for broader purposes.

19Nevertheless, one possible way of defending Shoemaker’s proposal against the accusation of placing implausible constraints on the notion of identification could be to argue that the alleged counterexamples are actually not instances of identification but something else (we might categorize them as instances of some sort of non-descriptive identity recognition). It could be argued that the concept of identification requires one to think its relata as having particular properties. Even though this seems rather absurd – since the term “identification” is arbitrarily used for a cluster of different things outside and inside of philosophy – one could maintain that for conceptual reasons the ability to perform genuine identification is dependent on the ability to think things as having this or that property. If this were true, then Recanati’s idea of an immediate recognition or Pylyshyn’s concept of visual indexes could be understood as not referring to counterexamples to Shoemaker’s definition of identification, but instead as viable solutions to his regress problem. If creatures had the ability to individuate objects and recognize their identity over time before being able to predicate something of them, then this ability could be regarded as a regress stopper, since descriptive identification could then be based on this pre-descriptive ability to visually trace identity without presupposing infinitely many further descriptive representations. This immediate self-recognition could work very similarly to regular perception if we suppose that both forms of identity recognition involved are able to track things in these non-descriptive ways – even though they possibly differ in many other aspects.

5. Some Consequences for the Relation between Self-Awareness and Perception

20Yet, Shoemaker – and others (e.g. Musholt, 2015; 2019; Frank, 2011; Zahavi, 1999; 2004; 2005) – probably would not agree that a basic kind of self-awareness could be based on the same mechanisms as visual perception. Shoemaker (1968, pp. 563ff) argues at length that self-awareness is fundamentally different from perception or even object cognition in general. More recently, Kristina Musholt (2015) has also argued that Shoemaker’s regress argument was precisely introduced to highlight a substantial difference between self-awareness and perception. She maintains that if all instances of self-awareness were a form of perception – her term is “self-perception” – then they would require identification and already presuppose some descriptive beliefs about oneself. Since not all cases of self-awareness can be based on identification, understood in this way, there must be cases of self-awareness which are substantially different from perception, so Musholt claims:

Shoemaker argues that self-consciousness cannot ultimately rely on self-perception, or in fact on any kind of subject-object relation, for to perceive myself as myself, I must already possess a first-personal knowledge of myself that can help me to identify a certain property as my own. (Musholt, 2015, p. 4)

21As said before, I think there is something to this assumption. However, I believe that Shoemaker’s argument itself is not able to establish a substantial difference between perception and self-awareness. Self-awareness could have the same underlying mechanisms as visual perception, which apparently, at the most basic level, works completely independent from the ability to represent any properties. If this is Shoemaker’s – and his followers’ – main reason for the assumption that self-awareness is fundamentally different from perception, then it might not be a very good one.