Preprint of Shaun Gallagher, S

Although philosophical approaches to the self are diverse, several of them are relevant to cognitive science. First, the notion of a 'minimal self', a self devoid of temporal extension, is clarified by distinguishing between a sense of agency and a sense of ownership for action. To the extent that these senses are subject to failure in pathologies like schizophrenia, a neuropsychological model of schizophrenia may help to clarify the nature of the minimal self and its neurological underpinnings. Second, there is good evidence to suggest that although certain aspects of the minimal self are primitive and embodied, other aspects may be accessed only in reflective consciousness. Employing a modified concept of the minimal self, it may be possible to construct a robotic form of non-conscious self-reference that depends on an interaction between the robotic body and its environment. In contrast to the minimal self, the narrative self involves continuity over time and is directly relevant to discussions of memory and personal identity. There is growing consensus among philosophers and cognitive scientists about the importance of narrative and its relation to episodic memory and lefthemisphere functions. There are, however, at least two different views of how the narrative self is structured. On one model it is nothing more than an abstract point. On a more extended view, proposed here, the self is a rich amalgam of narratives that allows for the equivocations, contradictions, and self-deceptions of personal life. Even in this case, however, neurocognitive models contribute to our understanding of how narrative identity is structured. Ever since William James (1890) provided a catalogue of different senses of the self, philosophers and psychologists have been hard at work refining and expanding the possible variations of this concept. Supplementing James' inventory of physical self, mental self, spiritual self, and the ego, Neisser (1988), for example, suggested important distinctions between ecological, interpersonal, extended, private, and conceptual aspects of self. More recently, reviewing a contentious collection of essays from various disciplines, Strawson (1999) found an overabundance of delineations between cognitive, embodied, fictional, and narrative selves, among others. It would be impossible to review all of these diverse notions of self in this short paper, so I have focused on several recently developed approaches that promise the best exchange between philosophy of mind and the other cognitive sciences. Because these approaches move in divergent theoretical directions they should help to convey the breadth of philosophical analysis on this topic. They can be divided into two groups that are focused, respectively, on two important aspects of self. A first approach involves various attempts to account for a 'minimal' sense of self. If we strip away all of the unessential features of self, the intuition is that there is a basic, immediate, or primitive something that we are still willing to call a self. This approach leaves aside questions about the degree to which the self is extended beyond the shortterm or 'specious' present to include past thoughts and actions. Although identity over time is a major issue in the philosophical definition of personal identity, the concept of the minimal self is limited to that which is accessible to immediate and present selfconsciousness. Non-philosophers have found that certain aspects of the minimal self are relevant to current research in robotics. Furthermore, aspects of the minimal self that involve senses of ownership and agency in the context of both motor action and cognition can be clarified by neurocognitive models (developed to explain pathologies such as schizophrenia) that suggest the involvement of specific brain systems (including prefrontal cortex, SMA, and cerebellum). A second approach involves conceiving of the self in terms of narrative, a concept imported into the cognitive-science context by Dennett (1991) , but one which may have a more complex significance than indicated in Dennett's account. The narrative self is extended in time to include memories of the past and intentions toward the future. It is what Neisser refers to as the extended self, and what Dennett calls a 'nonminimal selfy' self. Neuropsychological accounts of episodic memory or loss of memory can help to circumscribe the neurological underpinnings of the narrative self. Self-reference and misidentification There are a number of ways to understand the notion of a minimal sense of self. In this section I approach the problem by discussing how we use the first-person pronoun in a self-referring way that never permits a mistake. This kind of self-reference has a feature that some philosophers call 'immunity to error through misidentification relative to the first-person pronoun' (Shoemaker, 1984). I will refer to it as the immunity principle. Once this principle is clarified we can ask whether it can ever fail. In the next section we will explore this possibility in relation to a neurocognitive model of schizophrenia that requires us to make a distinction between two aspects of the minimal sense of self: the sense of self-ownership and the sense of self-agency. Wittgenstein (1958) distinguished between two uses of the first-person pronoun in selfreference: 'as subject' and 'as object'. Use of the first-person pronoun as subject might best be discerned by understanding what a speaker could be wrong about, and the kinds of questions that one could sensibly ask her. For example, if someone says 'I think it is raining outside', she could be wrong about the rain. It may not be raining. But it seems that she could not be wrong about the 'I'. That is, she could not misidentify herself when she states that it is she who is thinking. So, according to Wittgenstein, the following question would be nonsensical: 'Are you sure that you are the one who thinks it is raining?' Such use of the first-person pronoun is immune to error through misidentification. In contrast, when we use the first-person pronoun 'as object' it is possible to misidentify ourselves. For example, in experimental situations a subject's arm may be deafferented (that is, the subject is deprived of normal proprioceptive feedback about the position of his limb and therefore cannot keep track of it without vision), and visual perception of arm movement manipulated through mirrors or videotape (Jeannerod, 1994; Deprati, et al., (1997). In such cases, the subject might be led to say, 'I am moving my arm to the left', when in fact the basis for his judgment is a videotape of someone else moving their arm (not the subject's arm) to the left. In that case, he makes a mistake about who is moving their arm to the left. To say 'I' in such a case involves an objective misidentification of oneself. Shoemaker (1984) suggests that the immunity principle applies to the use of 'I' as subject because when we use the first-person pronoun as subject we are not actually attempting to identify ourselves. In other words, when I self-refer in this way I do not go through a cognitive process in which I try to match up first-person experience with some known criterion in order to judge the experience to be my own. My access to myself (my self) in first-person experience is immediate and non-observational (that is, it doesn't involve a perceptual or reflective act of consciousness). In this regard, the immediate self that is referred to here is the pre-reflective point of origin for action, experience, and thought. Are there any exceptions to the immunity principle? Is there any instance of someone using a first-person pronoun as subject, and being wrong in their reference? Following suggestions made by Feinberg (1978) and Frith (1992) about certain schizophrenic experiences (including auditory hallucination, thought insertion, and delusions of control in which subjects report that their body is under the control of other people or things), Campbell (in press) has proposed that such experiences might be counterexamples to the immunity principle. A schizophrenic patient who suffers thought insertion, for example, may claim that she is not the one who is thinking a particular thought, when in fact she is the one who is thinking the thought. Frith gives the following example: 'Thoughts are put into my mind like "Kill God". It's just like my mind working, but it isn't. They come from this chap, Chris. They're his thoughts' (1992, p. 66). In such cases the schizophrenic patient misidentifies the source of the thought and seemingly violates the immunity principle. Now whether or not Campbell is correct in his claim that this is a counterexample to the immunity principle (see Gallagher, 2000) the implications of his analysis are quite productive. His argument implies that a scientific explanation of schizophrenic phenomena such as thought insertion might also count as a scientific explanation of how the immunity principle works. Frith's neurocognitive model of the breakdown of selfmonitoring in schizophrenia turns out to be a good candidate for explaining immunity to error through misidentification. If we can identify which mechanisms fail at the neurocognitive or neurological level when the schizophrenic patient suffers from thought insertion, then we also have a good indication of the mechanisms responsible for (or at least involved in) the normal immunity to error found in self-reference, and the immediate sense of self. This insight moves us from the conceptual and often abstract argumentation of philosophy to the more empirical inquiries of neuropsychology and neurophysiology. A neurocognitive model of immediate self-awareness A brief consideration of motor action will help to clarify two closely related aspects of minimal self-awareness: self-ownership (the sense that it is my body that is moving) and self-agency (the sense that I am the initiator or source of the action). In the normal phenomenology of voluntary or willed action, the sense of agency and the sense of ownership coincide and are indistinguishable. When I reach for a cup, I know this to be my action. This coincidence may be what leads us to think of ownership of action in terms of agency: that the owner of an action is the person who is, in a particular way, causally involved in the production of that action, and is thus the author of the action. In the case of involuntary action, however, it is quite possible to distinguish between sense of agency and sense of ownership. I may acknowledge ownership of a movement-for example, I have a sense that I am the one who is moving or is being moved-and I can self-ascribe it as my movement, but I may not have a sense of causing or controlling the movement, that is, no sense of agency. The agent of the movement is the person who pushed me from behind, or the physician who is manipulating my arm in a medical examination. My claim of ownership (my self-ascription that I am the one who is undergoing such experiences) can be consistent with my lack of a sense of agency. Phenomena such as delusions of control, auditory hallucinations, and thought insertion appear to involve problems with the sense of agency rather than the sense of ownership (see Stephens and Graham, 1994). There is good evidence to suggest that the sense of ownership for motor action can be explicated in terms of ecological self-awareness built into movement and perception (Neisser, 1988; Gallagher and Marcel, 1999). In contrast, experimental research on normal subjects suggests that the sense of agency for action is based on that which precedes action and translates intention into action (Marcel, In press; Fourneret and Jeannerod, 1998). In addition, research which correlates initial awareness of action with recordings of the lateralised readiness potential and with transcranial magnetic stimulation of the supplementary motor area, strongly indicates that one's initial awareness of a spontaneous voluntary action is underlain by the anticipatory or premovement motor commands relating to relevant effectors (Haggard and Eimer, 1999; Haggard and Magno, 1999). It turns out that some schizophrenic patients who suffer from thought insertion also make mistakes about the agency of various bodily movements. To explain this, Frith (1992) appeals to concepts of efference copy and comparator mechanisms originally used to explain motor control (Sperry, 1950; Holst and Mittelstaedt, 1950). According to the most recent version of this model, and consistent with the findings cited above, a comparator mechanism operates as part of a non-conscious premotor or "forward model" that compares efference copy of motor commands with motor intentions and allows for rapid, automatic error corrections (Frith et al., in press; Georgieff and Jeannerod, 1998). This mechanism, consistent with the findings cited above, anticipates the sensory feedback from movement and underpins an online sense of self-agency that complements the ecological sense of self-ownership based on actual sensory feedback (Gallagher, 2000) (See Fig. 1). If the forward model fails, or efference copy is not properly generated, sensory feedback may still produce a sense of ownership ('I am moving') but the sense of agency will be compromised ('I am not causing the movement'), even if the actual movement matches the intended movement (Spence, et al., 1997).