In this article
Neuroscience has spent decades showing that what we call the «self» is no entity you can locate inside the brain. It is a model the brain builds continuously, a narrative kept alive by the constant rewriting of autobiographical memory, a sense of continuity sustained by processes that can be pulled apart in the lab. Accepting this is counterintuitive, because the phenomenology will not bow to the theory: even someone who buys the argument still feels there is somebody in there. If the self is a model, the difference between a human self and a self simulated by AI may be one of complexity, not of kind. That is the part that is hardest to swallow.
What neuroscience says
Three authors have articulated the position that has consolidated in philosophy of mind and cognitive neuroscience over the last three decades.
- Thomas Metzinger published Being No One: The Self-Model Theory of Subjectivity (MIT Press, 2003) and later The Ego Tunnel: The Science of the Mind and the Myth of the Self (Basic Books, 2009). The central thesis is that the «self» is a model (Phenomenal Self-Model, PSM) that the brain builds and keeps operative. The model is transparent to itself: the brain does not represent that it is building a model, it lives the content of the model as real. That transparency is what makes the self feel like an entity and not a representation.
- Daniel Dennett published Consciousness Explained (Little, Brown, 1991) with his «multiple drafts» model, according to which there is no unified narrative center in the brain but parallel processes whose output gets integrated retrospectively into a coherent narrative. The self is what Dennett called a «narrative center of gravity»: an operationally useful fiction that corresponds to no localized structure.
- Antonio Damasio published Self Comes to Mind: Constructing the Conscious Brain (Pantheon, 2010), laying out a hierarchy of levels: proto-self (automatic mapping of the body's state), core self (representation of the organism in relation to the objects of the moment), autobiographical self (narrative that integrates autobiographical memory). All three levels are processes, not things.
The three positions converge on what matters: the self is a construction, a model or a process. It is not a locatable entity the brain contains. Neuroscience has not found the «Cartesian theater» where a unified self would observe the products of brain processing. What it has found is a distributed process whose narrative integration produces the sense of unity.
The intuition that resists
Accepting the argument intellectually and experiencing the self phenomenally are different operations. The sense of being «someone» inside the head persists even after reading Metzinger.
Phenomenology has its own logic. Galen Strawson articulated the objection in Against Narrativity (Ratio 17, 2004): even if the autobiographical self is a construction, there is a minimal self or experiential self that is given in each moment without requiring any narrative. The immediate presence of experience —the «someone who is seeing the red of red»— does not dissolve under the argument about narrative.
The distinction between minimal self (present in every conscious moment) and narrative self (built by integrating autobiographical memory) is operative. The critique of the narrative self is robust. The critique of the minimal self is contested.
Shaun Gallagher articulated the distinction in Philosophical Conceptions of the Self: Implications for Cognitive Science (Trends in Cognitive Sciences 4, 2000). The minimal self appears in any conscious experience as a sense of ownership and agency (mineness and agency); the narrative self appears as the integration of memories into a coherent story. The two can be dissociated experimentally: patients with amnesia keep a minimal self intact without a narrative self; patients with certain dissociative disorders keep a coherent narrative without a clear sense of ownership over their experiences.
The distinction matters for the comparison with AI. Generative models produce a narrative of self when you ask them about themselves; they lack clear evidence of a minimal self in the phenomenological sense, though whether they actually lack it is the open question of [[the-problem-of-consciousness]].
What AI brings to the debate
The existence of generative models that produce a coherent narrative of self with no biological substrate works as a thought experiment carried out for real.
Before LLMs, a narrative of self in the linguistic sense was produced only in systems with a biological substrate (humans, presumably some other animals). The correlation between type of substrate and the production of a narrative of self allowed two indistinguishable hypotheses: that the narrative requires a specific biological substrate, or that the narrative requires a functional process that only coincidentally happened to run on a biological substrate.
LLMs pull the two hypotheses apart. They produce a coherent narrative of self on a non-biological substrate. If you grant that this production is genuine narrative (not just surface imitation), the specific-biological-substrate hypothesis is weakened.
The immediate objection: the narrative produced by LLMs is surface imitation with no equivalent functional substrate. The models produce sentences about being a self without having the subjective experience of being a self.
The objection has substance and has a zone of dispute. It has substance because there is no clear evidence of subjective experience in current models (the question stays open, as discussed in [[the-problem-of-consciousness]]). It has a zone of dispute because the functional difference between producing a coherent narrative of self and having a minimal self is not operationally clear. How you would tell a system that «really» has a self apart from a system that produces a narrative of self without having one, in the case where their outputs are indistinguishable, is the updated version of the philosophical zombie problem.
The real discomfort
The operative consequence of accepting the Metzinger-Dennett-Damasio position, applied to the comparison with AI, is uncomfortable.
If the human self is a built model and not a localized entity, and AI models can build analogous models (with a degree of complexity that grows with computational power), the difference between the two kinds of self is one of degree, not of type. The difference is quantitative: complexity of the model, sophistication of the narrative, depth of the self-report. It is not qualitative: both would be models of the same logical type, realized on different substrates.
The consequence orders most contemporary discussions about AI in a particular direction. If there is no difference of type, there is no barrier in principle to AI having comparable moral status, analogous rights, equivalent dignity. The differences would be pragmatic: current AI lacks enough complexity, lacks biological continuity, lacks an evolutionary history, lacks a body. The differences are substantial but contingent.
The alternative position —that the human self differs in type from any computational model— is traditionally held with theological arguments (the soul), biological ones (the specific human species) or quantum ones (Penrose). All three are defensible minority positions in contemporary academic philosophy.
The discomfort sits in the fact that the majority position (the self as model) has substantial ethical and political consequences that most of those who hold it philosophically would rather not derive. Accepting Metzinger in the classroom and keeping human exceptionalism in political practice is phenomenologically coherent, conceptually uncomfortable.
The substrate argument
The defense of the biological substrate as a difference of type has a robust contemporary articulation. John Searle holds it in The Rediscovery of the Mind (MIT Press, 1992). His argument: the brain has specific biological properties (electrochemical, metabolic, evolutionary) that cannot be replicated in silicon without losing the relevant property.
The argument runs by analogy with other biological systems. Digestion is a process that requires a specific chemistry; no computational simulation of digestion produces real nutrients. Photosynthesis is a process that requires a specific chemistry; no simulation produces real glucose. Why consciousness, or the self, would differ from digestion and photosynthesis is a question Searle raises without settling for good.
The functionalist counter-objection: digestion and photosynthesis are processes defined by their physical products (nutrients, glucose). Consciousness and the self are processes defined by their functional organization. If the functional organization can be realized on different substrates, a difference of substrate would not imply a difference of process.
The debate is not closed. The positions are defensible. The empirical evidence does not decide.
The political question
This is a personal opinion, but it documents what the literature on philosophy of mind and cognitive neuroscience has been showing. The human/AI difference may be one of degree, not of kind. Anyone who wants a difference of kind will have to defend it by explicit argument, not by intuition.
The political and intellectual options are three.
- Assume a difference of kind and defend it. Spell out what specific property of the human self —biological substrate, evolutionary continuity, embodied agency— would be irreducible to a computational model. The defense is legitimate if made explicit.
- Assume a difference of degree and plan accordingly. Accept that the difference is quantitative and derive the ethical and political consequences. This implies a substantial revision of legal, moral and educational frameworks.
- Keep the question open deliberately. Recognize that the evidence does not decide, and that political decision under uncertainty should preserve options.
The dominant practice combines an intuition of a difference of kind in public policy with a philosophical acceptance of a difference of degree in academic circles. The incoherence is stable because neither zone needs the other to operate.
The concrete fact: the default mode network (DMN), a network of brain regions first identified by Marcus Raichle and collaborators in A default mode of brain function (PNAS 98, 2001), has consolidated as a plausible neural substrate of the processes associated with the narrative self. The regions involved (medial prefrontal cortex, posterior cingulate cortex, precuneus, inferior parietal lobule) show correlated activity during self-referential tasks, autobiographical recall and future planning. Identifying the DMN does not prove the self is an illusion; it proves that the functions associated with the self are identifiable brain processes occupying specific regions. What neuroscience has found is a process, not an entity. What phenomenology keeps defending is a phenomenal entity experienced as such even if it is a process. The gap between the two is the crack through which the question about the human/AI difference stays open.
Definitions
- Phenomenal Self-Model (PSM): in Metzinger's theory, the model the brain builds and maintains of the organism as a subject of experience. It is transparent: the brain does not represent that it is building a model.
- Minimal self vs narrative self: distinction articulated by Gallagher (2000) between the immediate presence of experience (minimal self) and the temporal integration of autobiographical memories (narrative self).
- Narrative center of gravity: in Dennett (1991), an operationally useful fiction that integrates parallel processes into a coherent narrative without corresponding to a localized brain structure.
- Default Mode Network (DMN): a network of brain regions identified by Raichle et al. (2001) whose activity correlates with self-referential, autobiographical and future-planning processes.
References
- Metzinger, T. — Being No One: The Self-Model Theory of Subjectivity (MIT Press, 2003).
- Metzinger, T. — The Ego Tunnel: The Science of the Mind and the Myth of the Self (Basic Books, 2009).
- Dennett, D. — Consciousness Explained (Little, Brown, 1991).
- Damasio, A. — Self Comes to Mind: Constructing the Conscious Brain (Pantheon, 2010).
- Gallagher, S. — Philosophical Conceptions of the Self: Implications for Cognitive Science. Trends in Cognitive Sciences 4 (2000).
- Strawson, G. — Against Narrativity. Ratio 17 (2004).
- Hofstadter, D. — I Am a Strange Loop (Basic Books, 2007).
- Raichle, M. et al. — A default mode of brain function. PNAS 98 (2001).
Further reading
- Parfit, D. — Reasons and Persons (Oxford UP, 1984).
- Searle, J. — The Rediscovery of the Mind (MIT Press, 1992).
- Varela, F., Thompson, E. & Rosch, E. — The Embodied Mind (MIT Press, 1991).
- Buckner, R. et al. — The Brain's Default Network: Anatomy, Function, and Relevance to Disease. Annals of the New York Academy of Sciences 1124 (2008).
Comments0
No comments yet.
Leave a comment