The self-model theory of subjectivity (SMT)
The concept of a self-model plays the central role in a philosophical theory of consciousness, the phenomenal self and the first-person perspective. This specific theory is the so-called "self-model theory of subjectivity" (SMT; see Metzinger 2003a, 2005a). However, SMT is not only a conceptual framework in analytical philosophy of mind, but at the same time an interdisciplinary research program spanning many disciplines from neuroscience, cognitive science, neuropsychology and psychiatry to artificial intelligence and evolutionary robotics (e.g., Blanke & Metzinger 2009, Metzinger 2007, Lenggenhager et al. 2007, Windt & Metzinger 2007, Metzinger & Gallese 2003). The theory simultaneously operates on phenomenological, representational, functional and neuroscientific levels of description, using a method of interdisciplinary constraint satisfaction (see Weisberg 2007, section 2, for critical discussion). The central questions motivating the SMT are: How, in principle, could a consciously experienced self and a genuine first-person perspective emerge in a given information-processing system? At what point in the actual natural evolution of nervous systems on our planet did explicit self-models first appear? What exactly made the transition from unconscious to conscious self-models possible? Which types of self-models can be implemented or evolved in artificial systems? What are the ethical implications of machine models of subjectivity and self-consciousness? What is the minimally sufficient neural correlate of phenomenal self-consciousness in the human brain? Which layers of the human self-model possess necessary social correlates for their development, and which ones don’t? The fundamental question on the conceptual level is: What are the necessary and sufficient conditions for the appearance of a phenomenal self?
The phenomenal self-model (PSM)
The core notion of the theory is the concept of a “phenomenal self-model” (PSM). The PSM is that partition of an organism’s self-model which is conscious because it satisfies an additional set of functional constraints like, for example, availability for introspective attention and for selective, flexible motor control. What exactly the necessary and sufficient conditions for phenomenality actually are, is, of course, presently still unknown. SMT (Metzinger 2003a, chapter 3) offers a set of ten such potential constraints, the two most important of which are functional integration with a system's "virtual window of presence" (i.e., its internal representation of time and an extended "Now") plus ongoing, dynamical integration into a single, overarching world model (i.e., a multimodal representation of the current situation as a whole). A "model" is a working concept for a type of mental representation that does not contain variables, represents a situation by structural correspondence (i.e., via a partial relational homomorphy), the elements of which can correspond to perceptible entities as well as to abstract notions, which can be taken offline for internal "dry runs", and which may or may not satisfy the constraints for becoming a phenomenal mental model (see Craik 1943, 51pp; Johnson-Laird, 1983, 1989, p. 488; Metzinger 2003a, section 3.3 for more on the history of the concept). What we call “the conscious self” in folk-psychological discourse is a specific form of representational content, characterized by specific functional properties, which in turn can be physically realized in a large number of different ways. The phenomenal content of the PSM (how its content is subjectively experienced) locally supervenes, i.e., it is fixed as soon as all contemporaneous and internal properties are fixed. For example, as soon as all of the relevant parameters in the brain are set, the phenomenology going along with the activation of a PSM is determined as well. This may not be true of its intentional content, because self-knowledge (as opposed to subjective, self-experience) is arguably co-determined by external factors like the way in which the system is historically and socially situated. One background assumption is that for every biological PSM, in the species known to us, there exists a minimally sufficient neural correlate of self-consciousness (NCSC; see Metzinger 2000). This NCSC would be a specific set of neurofunctional properties of which it is true that (1) it reliably activates a phenomenal self and that (2) it possesses no proper subset that suffices for the corresponding states of consciousness. The NCSC defines the set of properties that are relevant for a scientific explanation of phenomenal self-consciousness. If the predicates describing the relevant, locally supervening phenomenal properties and those referring to the neurofunctional properties determining them are nomologically coextensive, a reductive identification of the PSM is possible. The PSM would then be simply identical to the NCSC.
The self-model theory aims at maximally parsimonious framework for the scientific investigation of self-consciousness: There is no such thing as a substantial self (as a distinct ontological entity, which could in principle exist by itself), but only a dynamic, ongoing process creating very specific representational and functional properties. Self-consciousness is a form of physically realized representational content. This metatheoretical framework has implications for many of the first-order empirical sciences of the mind. For example, the methodological core of scientific psychology can now be analyzed in a clearer, fresher and more fruitful way. Psychology, from this perspective, is self-model research: It is the scientific discipline that focuses on the representational content, the functional profile, and the neurobiological realization of the human self-model, including its evolutionary history and its necessary social conditions. Psychiatric disorders can be described and diagnosed more systematically, e.g. as specific representational contents in the self-model that have been lost, hyperactivated, or decontextualized, which are not available for introspective access any more, or which have become causally autonomous and functionally dissociated in other ways (Metzinger 2003a: ch. 7; 2004c)
Aside from the representational level of description, one can also develop a functional analysis of the self-model. Whereas representational states are individuated by their content, a functional state is conceptually characterized by its causal role: the causal relationships it bears to input states, output states, and other internal states. An active self-model then can be seen as a subpersonal functional state: a discrete set of causal relations of varying complexity that may or may not be realized at a given point in time. Since this functional state is realized by a concrete neurobiological state, it plays a certain causal role for the system. For instance, it can be an element in an information-processing account. The perspective of classic cognitive science can help illustrate this point: the self-model is a transient computational module that is episodically activated by the system in order to control its interactions with the environment. In other words, what happens when you wake up in the morning, i.e., when the system that you are “comes to itself,” is that this transient computational module, the PSM, is activated – the moment of “waking up” is exactly the moment in which this new instrument of intelligent information-processing emerges in your brain. It does so because you now need a conscious self-model in order to achieve sensorimotor integration, generate complex, flexible and adaptive behavior, and attend to and control your body as a whole. The assumption in the background is that the space of consciousness essentially is the space of availability for selective, high-level attention, and that the activation of a conscious self model becomes necessary whenever an organism (or an artificial system), in order to solve a certain task, needs an integrated internal representation of certain of its own global properties to make them available for selective resource allocation and deeper processing. This idea, namely, that the PSM is that partition of the currently active self model functionally characterized by availability (but not necessarily ongoing access), is a central working hypothesis under SMT, but its truth has not yet been empirically demonstrated.
For higher forms of intelligence, in order to integrate higher levels of behavioral and cognitive complexity, a system needs a coherent self-representation, a consistent internal model of itself as a whole. A PSM is an instrument for global control. In our own case, the conscious self-model is an episodically active representational entity whose content is determined by the system’s very own properties. Interestingly, we do not have a PSM in dreamless deep sleep, but there certainly is a PSM in the dream state. It can be characterized as an instrument for interacting with an exclusively internal environment, exhibiting not only a unique phenomenological profile, but a corresponding set of typical representational, functional and neurobiological features which can be described in a fine-grained manner. (Windt & Metzinger 2007). Many altered and pathological states of consciousness can be analyzed as deviant forms of self-modeling (Metzinger 2003, chapter 7). The evolution of higher cognition generally can be described as an increasing ability to take the PSM offline and to develop more abstract forms of mental-self simulation. For instance, simulating past or possible future states of the situated organism as a whole while comparing them to the current online self-model will enable memory, learning, or planning. Abstract thought would then have evolved out of the fundamental capacity to create forward models and to internally simulate elementary motor behavior and its perceptual consequences (Cruse 2007).
The development of ever more efficient self-models as a new form of „virtual organ” is also a precondition for the emergence of complex societies. Plastic and ever more complex self-models not only allowed somatosensory, perceptual, and cognitive functions to be continuously optimized, but also made the development of social cognition and cooperative behavior possible. The most prominent example, of course, is the human mirror system, a part of our unconscious self-model that resonates with the self-models of other agents in the environment through a complex process of motor-emulation - of “embodied simulation,” as Vittorio Gallese (2005) aptly puts it – for example, whenever we observe goal-directed behavior in our environment. Such mutually coupled self-models, in turn, are the fundamental representational resource for taking another person’s perspective, for empathy and the sense of responsibility, but also for metacognitive achievements like the development of a concept of self and a theory of mind (for possible neurobiological correlates of these basic social skills, see Gallese & Goldman 1998, Metzinger & Gallese 2003).
The obvious fact that the development of our self-model has a long biological, evolutionary, and (a somewhat shorter) social history can now be accounted for by introducing a suitable version of teleofunctionalism as a background assumption. The development and activation of this computational module plays a role for the system: the functional self-model possesses a true evolutionary description, metaphorically speaking it was a weapon that was invented and continuously optimized in the course of a “cognitive arms race” (Clark 1989: 61) The functional basis for instantiating the phenomenal first-person perspective can be seen as a specific cognitive achievement: the ability to use a centered representational space (Trehub 1991, 2007, 2009). In other words, phenomenal subjectivity (the development of a subsymbolic, non-conceptual first-person perspective) is a property that is only instantiated when the respective system activates a coherent self-model and integrates it into its global world-model.
Friday, October 02, 2009
Dr. Thomas Metzinger - The self-model theory of subjectivity
Cool stuff, via Scholarpedia.