Every theory that sets out to explain consciousness walks into a peculiar trap: it must specify who receives the result of the explanation. As soon as a model describes how neural processes generate an internal representation of the world, the question the model was supposed to answer resurfaces at the next level. Who contemplates the representation? Who experiences the model? A little man in the head — a homunculus — staring at a screen, inside whose head sits another, staring at another screen, ad infinitum. This regress bears the name homunculus problem. It is not merely a logical objection to particular theories of consciousness. It exposes the fundamental structure of a philosophical dead end.
#The Mill in Which Nobody Thinks
Leibniz captured the core of the problem in 1714 in the Monadology as a thought experiment. Imagine a machine built so that it thinks, feels, perceives. Enlarge it until you can walk inside it as you would a mill. What do you find? Parts pushing against each other. Wheels meshing together. Mechanical operations that can be described without perception or consciousness appearing anywhere (cf. Leibniz, Monadology, 1714, section 17).
Leibniz wrote against Descartes, who in 1641 in the Meditationes had carried through the split between thinking substance and extended substance. Descartes’ dualism created the problem by separating two worlds between which no bridge was any longer conceivable. Leibniz’s mill argument shows the consequence: if the physical world contains only extension and mechanics, then you can enlarge it, disassemble it, walk through it at will and never encounter consciousness. Not because the magnification is too coarse, but because consciousness belongs to a different order than mechanics.
#Who Observes the Self-Simulation?
In the current debate, the homunculus problem returns under new terminology. When Joscha Bach describes consciousness as an emergent control process — as a self-simulation that a system generates about itself — the question Leibniz’s mill raised remains: who observes this simulation? Bach treats consciousness as a software process, a model that maps itself. But a model that maps itself experiences nothing so long as there is no recipient for whom the mapping constitutes experience. The simulation of introspection is not introspection.
Gwendolin Kirchhoff addressed this point directly in the Everlast AI debate with Joscha Bach (2026). The homunculus, the self-simulation of which Bach speaks, is observed by someone — namely by a mind. Consciousness is something that lies beyond the phenomenal person we believe ourselves to be. Meditative practice and phenomenology converge in showing that there is an awareness that runs deeper and wider than any single content of consciousness (cf. Gwendolin Kirchhoff, Everlast AI Debate, 2026, 26:38).
In the same debate, Kirchhoff drew the connection to the boot problem: the mechanistic metaphysics that treats consciousness as a product of physical processes generates both the hard problem and the boot problem, and it can solve neither. Natural philosophy, which conceives the cosmos as a living, consciousness-pervaded whole, does not face these problems because it need not retroactively assemble consciousness from the non-conscious (cf. Gwendolin Kirchhoff, Everlast AI Debate, 2026, 76:24).
#The Regress as Symptom
In analytic philosophy, the homunculus problem is frequently treated as a technical flaw — as an indication that a particular theory has not cleanly closed its chain of explanation. The attempted solution then runs: eliminate the observer, explain consciousness as a pattern without a subject, a process without a recipient. Bach’s answer moves within this framework: there is no homunculus, there are only patterns, self-models, functional structures.
Kirchhoff’s objection is more fundamental. Eliminating the observer does not remove the problem — it conceals it. Whoever defines consciousness as a mere pattern has not answered the question the regress poses but forbidden it. The pattern exists for someone, or it does not exist as consciousness. The question of who experiences the pattern does not vanish because one declares it inadmissible.
Jochen Kirchhoff formulated the ontological dimension of this thought in Science Under Scrutiny (2021): science has failed at the phenomenon of consciousness — it had to pass. Consciousness is precisely what is primary. One must always begin from consciousness; it is the starting point of any serious philosophy (cf. Jochen Kirchhoff, Science Under Scrutiny, 2021, 40:07).
From the perspective of natural philosophy, the homunculus regress is not the deficiency of a particular theory but the inevitable result of a false approach: the attempt to derive consciousness from something that is not itself conscious. Whoever starts from mechanism and asks how experience arises from it will need a new observer at every level of explanation, because the mechanism itself is blind. The regress ceases only when consciousness is understood not as a product but as a presupposition.
#Goethe’s Homunculus as Philosophical Figure
Goethe gave the figure of the homunculus a form in the second part of Faust (1832) that translates the philosophical problem into poetry. Homunculus — the artificial being produced in a glass flask — can see, think, speak, but cannot enter the world. It remains in its vessel. It needs, as Goethe formulates, the becoming of creation in order to truly exist. Artificial life does not transform into real life. The homunculus cannot become a human being unless it surrenders to natural becoming.
The figure illustrates what the philosophical regress shows in abstract form: a model, a simulation, a construction can imitate consciousness, but the imitation has no inside. It is mill, not mind. The step from model to experience is not gradual, not quantitative, not surmountable through increasing complexity. It presupposes what it purports to generate.
If you ask about the connection between the homunculus regress and the question of artificial intelligence, the answer lies in the structure of the regress itself: what is missing is not computing power, not architecture, not data, but the subject that apprehends the computation as experience. The emergence thesis, which claims consciousness appears beyond a certain threshold of complexity, merely shifts the homunculus by one level. The epiphenomenon argument — that consciousness is a causally impotent by-product — eliminates the homunculus by eliminating consciousness itself. Both manoeuvres bypass the regress without dissolving it. Consciousness philosophy unfolds the terrain on which this question stands.