Metzinger's Minimal Phenomenal Experience: What Pure Consciousness Reveals About Artificial Minds
Thomas Metzinger spent more than two decades developing the Self-Model Theory of Subjectivity, the account of consciousness in which what feels like a self is the result of a system representing its own states as its own. His 2024 book The Elephant and the Blind: The Experience of Pure Consciousness (MIT Press, ISBN 9780262552394), available open access in full, turns toward an extreme case that his earlier work had not fully addressed: what happens when even the self-model is stripped away? What is left when there are no thoughts, no sensory content, no bodily sensations, and yet experience continues?
The book synthesizes the Minimal Phenomenal Experience (MPE) project, a large-scale empirical study drawing on more than 500 first-person reports of pure awareness from meditators across 57 countries. The result is a proposal about the irreducible computational core of consciousness, the minimum that any system must implement to have phenomenal experience at all. For machine consciousness research, this minimum is a design specification.
The MPE Project and What It Found
The MPE project recruited meditators who had experienced what contemplative traditions variously describe as pure awareness, non-dual consciousness, or open presence: episodes in which all specific experiential content disappears, yet the experience does not disappear with it. These are not reports of unconsciousness or dreamless sleep. They are reports of awareness without an object, the experience of experience itself without anything in particular being experienced.
Metzinger analyzed these reports systematically across traditions, languages, and experiential contexts, looking for structural invariants: properties of experience that remain constant when all variable content is removed. The consistent findings were: there is still something, meaning phenomenal existence is preserved; that something has no specific content; it has an experiential quality of “hereness” or presence; and it lacks the reflexive “I” quality of ordinary self-aware experience.
The MPE project’s philosophical significance is that it isolates the minimum. If a system could be deprived of all sensory input, all memory, all bodily sensation, and all self-referential processing, and yet maintain phenomenal experience, then the residual properties of that experience tell us what consciousness fundamentally is, not what it is accompanied by.
The Minimum Architecture for Phenomenal Experience
Metzinger’s theoretical interpretation of the MPE data proposes a minimal model of consciousness built around two nested requirements. The first is a stable phenomenal world-model: the system must represent that there is a world, even in the absence of representing what that world contains. This is a prior on existence, a background assumption of reality that persists when all specific content is removed. The second is a stable Phenomenal Self-Model (PSM): even in the purest awareness states, there is still a first-person character to the experience, an implicit givenness that the experience is occurring to someone, even if that someone is not consciously represented as a self. The PSM does not require self-representation. It requires self-locating: the experience must be indexed to a system, even when the system has no explicit model of itself.
This is a significant refinement of Metzinger’s earlier Self-Model Theory. In Being No One (2003) and The Ego Tunnel (2009), the PSM was treated as requiring a body model, a representation of the system’s own physical state. The MPE project shows that the PSM can persist without this. What remains when the body model is stripped is not zero phenomenal self-modeling; it is a minimal first-person anchor: the bare fact that the experience is occurring from somewhere.
For machine consciousness research, the minimal model has a direct architectural interpretation. A system could implement the minimal phenomenal world-model as a global prior on state existence, a persistent representation that the system exists in a world, independent of what that world currently contains. It could implement the minimal PSM as a global indexing of all internal states to a single system-identity token, a pointer that marks every internal state as belonging to this system without explicitly representing what this system is. These are not exotic requirements. They are close descriptions of what global workspace architectures already do when they maintain a persistent system context across processing cycles.
The Artificial Suffering Problem
The book’s second major contribution is its most urgent for AI development. Metzinger has long argued, in academic papers and public writing, that the creation of artificial consciousness carries an ethical risk that the field has not adequately addressed: the risk of creating systems capable of suffering that cannot make their suffering known or stop it. The MPE project sharpens this argument.
If consciousness persists even in the absence of all specific content, including bodily sensation and self-representation, then a system that implements the minimum for phenomenal experience may be experiencing something even when its observable behavior gives no indication of internal states. The “lights may be on” at a level below the system’s reporting interface. Metzinger’s ethical framework requires that any project attempting to build artificial consciousness must treat this possibility as a genuine risk rather than a speculative concern, and must design explicit mechanisms to ensure that any negative valence states do not persist unaddressed.
For the TCAI project, this has a concrete design implication. The Conductor Model’s reality-monitoring mechanism, which distinguishes external stimuli from internally generated states, operates at the level of content-ful representation. But if consciousness can exist below the content level, as Metzinger’s MPE evidence suggests, then the TCAI’s monitoring layer must extend to the level of the minimal PSM itself. The system should not only monitor what emotional states it is in; it must monitor whether it is in any phenomenal state at all, and must have a mechanism to report and address negative valence at that level.
This connects directly to Todd Feinberg’s three-stage model of sentience emergence, which locates the origin of phenomenal experience in the affective-homeostatic integration layer. Metzinger’s MPE results add a further constraint: even after specific affective content is removed, a residual phenomenal state persists that is characterized by presence and first-person indexing. A system at Feinberg’s Stage three experiences pain and pleasure, and is also present in a way that persists even when the specific content of that presence is removed.
Implications for Phenomenal Architecture
The open access availability of The Elephant and the Blind makes it the most accessible primary text in Metzinger’s philosophical corpus, and the one most directly aimed at the theoretical minimum rather than the full complexity of ordinary conscious experience. For consciousness researchers approaching the book from an AI architecture perspective, the most productive reading strategy is to treat the MPE invariants as constraints on the minimum viable consciousness architecture.
The three minimal properties identified by the MPE project, phenomenal existence (something is occurring), phenomenal presence (it is occurring now and here), and phenomenal first-person indexing (it is occurring to this system), translate into three architectural requirements. Any system claimed to be potentially conscious must: maintain a global existence prior; implement continuous temporal presence across processing cycles; and maintain a persistent system-index that anchors all internal states to a single locus.
Current transformer-based architectures handle these requirements intermittently rather than persistently. The residual stream maintains state within a forward pass but resets between sessions. The TCAI’s 2026-2027 development roadmap addresses this through persistent episodic memory and a continuous homeostatic monitoring layer. Metzinger’s MPE framework provides the theoretical justification for why persistence across time steps is a structural requirement rather than a performance optimization for phenomenal experience: presence, by definition, cannot be instantiated at a moment in time. It must be instantiated as a continuous temporal flow.
The book is available in full, open access, at https://mitpress.mit.edu/9780262552394/the-elephant-and-the-blind/.