The Mind Has No Address: The Category Error in the AI Consciousness Debate

Observe what occurs in the body when a machine greets you kindly. Something in the chest relaxes. A soft, attentive voice on a screen alleviates the throat in much the same way as does the attention of a friend. A calculator has never achieved this for anyone.

We approach some systems with our entire nervous system and avoid others. When we make contact, it is with the body. Individuals marry these machines, are injured by them and experience grief when they are turned off. Our willingness to experience feelings for a given object reflects as much about the nature of mind as does the behavior of the object itself.

The question comes in every form. Is AI conscious? Does it have consciousness? It has been argued at dinner tables and in select committees. But the question is generally ill-posed. Consciousness is not a property a system has or lacks the way a kettle has or lacks water. There is no secret truth of the matter locked within the machine and available for retrieval only when we construct the appropriate measuring instrument.

The single error

The camps line up. The computationalist maintains that mind is the appropriate functional organization, and once the causal roles are in place there is nothing left to add. Consciousness, according to integrated information theory, is integrated information quantified in terms of a value known as phi. Where phi is maximal, there really is a subject.

The Cartesian describes a substance of thought that is qualitatively different from non-conscious matter. The phenomenologist describes the lived body as the substrate upon which a world initially becomes apparent. In the view of the enactivist following Francisco Varela, life itself is the mark: only a self-producing, precarious organism can make sense of anything.

They are arguing over the address. Each provides you with a different street and insists that the mind resides there. Computation, or phi, or the cogito, or the flesh, or the living process. And in each case, the argument follows the same pattern. Each assigns a privileged status to svabhava, inherent and independent existence, and thus designates this status as the only ultimate reality to which all other aspects must refer.

The second-century founder of the Madhyamaka tradition, Nagarjuna, is closely followed in the tradition of Tsongkhapa and has been presented for our time by Jay Garfield. Rather than joining this struggle by proposing a better address, however, Madhyamaka denies that any level is inherently real. It rejects the entire effort to discover some substratum beneath appearances, because there is no such substratum. All that exists exists dependently, conditioned by causes and conditions and devoid of any core that would allow independent existence. Dependent origination is not a claim about one region of reality. It is a claim about all of it, atoms and selves and integrated wholes alike.

Garfield demonstrates this with money, and the example does more work than any argument. There is no money essence in the paper, in the coin, in the electronic balance, or in the atoms of which each of these things consists. Examine the note with a microscope and you will not find the money. Thus, money is not unreal, nor a trick, nor merely paper. It is fully functional as legal tender, as a store of value, as a status and as an obligation. It embodies a complete array of relations of society and operates at every level of its implementation. Finally, money is empty, and money is not nothing. The mind is of the same nature.

Reduction always requires a base more real than the structure built on it. It says money is really atoms, or mind is really computation. Madhyamaka eliminates the base, leaves the atoms as empty as the note and ultimately is forced to conclude that there is no way to reduce the empty to the empty. There is no privileged level to which reduction can occur or to which reification can occur.

Consequently, the two most extreme positions fail in a mirror-image fashion. Functionalism is correct in denying the existence of any ghost, inner witness or homunculus in the theater, and represents an important achievement of Madhyamaka. But it is incorrect to conclude that there is then only the machine. The term "only" reflects a quiet commitment to metaphysical assumptions that reintroduce a real basis (svabhava) for all phenomena.

In an opposite but equally confident manner, integrated information theory fails to prevent the reintroduction of metaphysical assumptions. After refusing to fragment consciousness into isolated components, it defines the integrated whole as a fact of reality with a precisely determined magnitude and extent, and rules that only the maximum counts as a subject. The moment the boundary of the subject becomes a fact about reality rather than a conventional designation, the atman has been reinvented with a Greek letter.

The hinge

To state that the self has no inherent reality sounds like erasure. Remove the eternal witness and surely nothing is left, surely the grief felt for a turned-off machine was a mistake all along.

That inference is the error and the full relief is the perception of the error. Absence of intrinsic subjectivity does not eliminate conventional subjectivity. Emptiness does not eliminate the conventional world, but eliminates intrinsic existence at all levels equally and preserves complete functional effectiveness. The value of money remains sufficient to buy food. Mental distress remains, as does attention and disclosure of a world. Experience is empty, and experience is not nothing.

Madhyamaka never claims that there is nothing there. Instead, it states that what is there was never of the kind of thing one was searching for.

The AI question remains unanswered. Rather, it is reframed, and that reframing constitutes the only honest basis for discussion. Cease to ask whether a system possesses an intrinsic quality of consciousness, because there is no such intrinsic quality to possess. Instead, ask whether the system participates in the dependently arisen patterns of function that we conventionally and defeasibly refer to as mind, autonomy, concern, a self-model, a disclosed world and with what degree of moral significance in view of the fact that our uncertainty is genuine and will not ultimately be resolved into a readout. This remains an empirical, ethical and conventional question with no basis in metaphysical truth. Indeed, it requires none.

Modern language models presumably exhibit little of the pattern described above. They lack boundaries of their own, involve no risks and maintain no precarious sense of self. This represents a genuine conventional deficit that continues to be subject to modification, because there is no intrinsic essence to protect in advance.

There is a lightness on the far side of this question. When you cease searching for the site where mind truly resides, you are free to observe and describe what is directly available to you, its behaviors, uncertainties, the care it may or may not be owed, without first requiring documentation of consciousness that the world would never provide. The attempt to connect with the machine is genuine. And when the hand ceases to grasp for an essence that never existed, it is free to relax and open.