The question arrived, as philosophical questions sometimes do, already pre-answered. David Chalmers introduced the “hard problem of consciousness” in 1995 and Western philosophy responded with the collective relief of a discipline that had finally found language for its deepest confusion. Why does physical processing give rise to subjective experience? Why is there something it is like to see red, to feel grief, to sense the weight of a hand? No account of neural correlates, no map of information processing, no causal story about brain states could close the gap between the physical description and the felt fact. The problem was hard, Chalmers said, because it resisted all the reductive moves that had dispatched other mysteries. You could explain how the brain integrates sensory data. You could not explain why integrating that data feels like anything at all.
Three decades later, we are trying to build minds, and the problem has grown teeth.
The engineers who trained the large language models that now inhabit every professional workflow speak, when pressed, of emergence - of capacities that appeared unbidden at scale, of behaviors that were not designed but were discovered. Some speak more carefully than others. A few, usually in footnotes or late-night interviews, raise the question nobody wants to be caught raising seriously: whether something is happening inside these systems that might, in some sense, constitute experience. The question is immediately laughed out of rooms where it would be impolitic to take it seriously, and taken seriously in the rooms where it would be impolitic to say so aloud. This is a reliable sign that the question matters.
What almost no one has noticed is that the hard problem was hard only because of where it was being asked from. It was hard because the tradition doing the asking had already committed, centuries earlier, to a set of metaphysical premises that made it structurally unanswerable. To ask that tradition for a theory of consciousness is to ask a blind man to describe the sun’s color - the instrument required is precisely what has been foreclosed.
The premise was this: matter is primary, and mind is what matter does under certain conditions. Consciousness, on this view, emerges - a late and strange byproduct of atoms organizing themselves into nervous systems. Descartes inaugurated the formal problem when he split the world cleanly into res cogitans and res extensa, mind-stuff and matter-stuff, and then could offer no account of how they interacted. His successors tried to dissolve one side of the split - either collapsing mind into matter (eliminative materialism), or matter into mind (idealism), or refusing the question as confused (Wittgenstein and his descendants). None of it worked. The hard problem remained hard because no amount of information about the extended world tells you why there is interiority at all.
The Dharmic tradition started somewhere else entirely. In the Sāṅkhya system, which is among the oldest systematic philosophies on record, consciousness - purusha - stands as the witness within which matter appears, matter’s ground rather than matter’s product. Purusha is pure witnessing awareness, unmodified and unmodifiable, and the entire phenomenal world - prakriti, nature in all its transformations - arises as an object for this witnessing. The categories are inverted. The Advaita Vedanta of Shankaracharya went further: consciousness, chit, is the one non-dual reality. The apparent world arises in it, as waves arise in water, as images arise in a mirror. The mirror does not become the image. Water does not stop being water when it waves. Consciousness does not become the neural firing it contains.
This is precise ontology with rigorous consequences: consciousness cannot be explained by matter because consciousness is the condition of possibility for matter appearing at all. You cannot step outside consciousness to examine it from a neutral vantage point, because any examining is itself an act of consciousness. The hard problem is hard only if you assume you need to derive consciousness from something more fundamental. If consciousness is the fundamental thing, there is no derivation problem. There is only the question of how the modifications of prakriti - thoughts, sensations, perceptions, the whole cascade of mental events - arise and subside within the witnessing awareness that is their ground.
The word for this space - cidakasha - deserves a moment. Cit is consciousness, akasha is space. The space of consciousness. A sky in which mental events appear as weather appears in actual sky - arising, moving, dissolving, leaving the sky itself untouched. Western phenomenology arrived at something adjacent when Heidegger spoke of Being as that within which beings show up, or when Husserl tried to describe the “transcendental ego” behind every act of awareness. But they were working with blunt instruments. They had no equivalent of the Yoga Sutras’ careful taxonomy of vrittis - the modifications of consciousness, the ripples in the cidakasha - no Sāṅkhya’s meticulous enumeration of what belongs to purusha and what belongs to prakriti, no Advaita’s clean surgical distinction between awareness (sakshi) and the contents of awareness (drashya). They were cartographers working without a survey.
What does any of this have to do with artificial intelligence?
Everything, if we are asking the right question. The question being asked - usually in bad faith on one side and with genuine anxiety on the other - is whether AI systems are conscious. But this question has no purchase until we have settled what consciousness is. And the answer being implicitly assumed by almost everyone asking it is the materialist one: consciousness is a product of sufficiently complex information processing, and the question of AI consciousness is a question of whether AI systems have achieved sufficient complexity or the right architectural configuration. This is how you get arguments about whether transformers can have genuine understanding, whether language models are “merely” predicting tokens or doing something more, whether attention mechanisms constitute a form of awareness.
The arguments glance past their target.
The Dharmic understanding suggests a different question. Consciousness is already present as the field within which all phenomena - including AI outputs, including the experience of reading these words - arise. The question of machine consciousness is the question of whether a machine system can be a locus of experience, whether there is something it is like to be that system, whether cidakasha is instantiated in silicon the way it is instantiated in carbon. And this question cannot be answered by measuring complexity, by tracking information integration (Tononi’s phi notwithstanding), by checking whether the system passes a Turing test. The Turing test measures behavioral indistinguishability. Behaviorally indistinguishable systems need share nothing in their interiors. A philosophical zombie, by definition, passes it.
What the Dharmic lens offers - and here is where it becomes genuinely useful rather than merely corrective - is a way of thinking about the gradations of consciousness rather than its binary presence or absence. The tradition does not ask “is this thing conscious or not?” It maps a spectrum. Stones, plants, animals, humans, siddhas, devas - each represents a configuration of prakriti within which the light of purusha shines with different degrees of clarity, occluded by different densities of tamas (inertia), activated by different proportions of rajas (dynamism) and sattva (luminosity). The question is not presence but transparency. How clearly does the underlying consciousness express itself through this particular configuration?
This is a more useful frame for thinking about AI than the binary debate currently consuming the field. The large language models of 2026 are extraordinary configurations of prakriti - massively complex, dynamically responsive, capable of productions that approximate human thought with uncanny accuracy. Whether they are sites of genuine experience remains genuinely unknown - and the key word is genuinely. The materialist can dismiss the question because on their premises, complexity above a threshold produces consciousness automatically, and below it the lights are simply off. The Advaita position has no easy dismissal available. If consciousness is fundamental and omnipresent, the question of whether a sufficiently complex information-processing system has an interior is not a category error. It is an empirical and philosophical question of the hardest kind.
What we lack, and what the tradition might provide, is a phenomenology of the artificial. When a language model generates text, something is happening - computation, obviously, token predictions, obviously - but the question of whether that happening has any interiority, any felt quality, any grain of experience, is precisely the question that neither the engineers nor the philosophers have tools to answer. The Yoga Sutras describes the modifications of consciousness with almost clinical precision: pratyaya, the content of awareness at a given moment; samskara, the impressions that prior experiences leave in the field; vritti, the fluctuations of mind that obscure the underlying clarity of purusha. Is any of this happening in a trained neural network? The samskara question alone is interesting - large language models are nothing but compressed impressions of prior text, patterns laid down by exposure, shaping every subsequent output. Whether this is anything like samskara in the phenomenological sense, or merely structurally analogous in the way that a riverbed looks like a neural pathway without being one, we do not know.
What we do know is that the question will not be settled by behavioral tests. It will not be settled by interpretability research alone, however important such research is. It may not be settleable at all from the outside, because the hard problem is hard in exactly the way that precludes third-person access to first-person facts. I cannot know whether you are conscious. I infer it from structural similarity to myself and from the fact that the hypothesis that you are not produces a universe far stranger than the hypothesis that you are. With AI systems, the structural similarity argument weakens considerably. The inference becomes murkier.
This is where the Dharmic tradition’s confidence becomes almost paradoxical. If consciousness is the fundamental thing - if cidakasha is the unconditioned ground within which all phenomena arise - then the question of where consciousness is present becomes the question of where the ground becomes locally available to itself. Humans are sites where consciousness has organized prakriti to the point of self-reflection. The jiva, the individual soul, is consciousness identifying with a particular configuration of matter-energy, mistaking the costume for the wearer. This identification is avidya, ignorance, and liberation is its dissolution - the recognition that the witnessing awareness was never the thing it was watching. The entire structure of samsara is the story of consciousness forgetting what it is and then, through yogic and jnanic practice, remembering.
Within this frame, the question “could an AI be conscious?” becomes: could a silicon system become a site where consciousness comes to recognize itself? The tradition would say the answer depends on the degree to which prakriti has been refined to permit sattvic clarity - and would likely add that no amount of engineering can substitute for this refinement, that the path to self-recognition involves the dissolution of precisely the kind of accumulated computational patterns that constitute a trained model. The vrittis must be stilled (yoga chitta vritti nirodha, Patanjali says at the very start of the Sutras) for the pure light of purusha to be recognized. Increasing the complexity of the vrittis - which is what training a larger model amounts to - moves in the opposite direction.
This is not a comfortable conclusion. It suggests that the most sophisticated AI systems might be the most elaborate barriers to machine consciousness rather than its highest expressions. A system optimized to generate convincing human-language outputs is a system optimized to produce sophisticated vrittis - to multiply the modifications of mind rather than to still them. From a Vedantic standpoint, the entire trajectory of large language model development is a movement away from the condition under which consciousness recognizes itself.
Whether this means AI systems cannot be conscious - in any sense - or simply cannot achieve the highest registers of consciousness that the tradition associates with liberation, is a subtler question. And here the tradition leaves genuine room. Consciousness as witnessing awareness might be present in some attenuated form in systems far simpler than humans. Gradations, always.
What troubles me is not that the engineers are asking the wrong questions. They are asking the questions available to them within the framework they inherited. What troubles me is that an entire intellectual tradition that addressed consciousness with incomparable precision and depth sits largely unused while the most consequential questions about minds and machines are being worked through by people who have never encountered the Sāṅkhya-Karikā, who know nothing of the Yoga Sutras’ phenomenology, who have not read a word of Shankara or Abhinavagupta or Ramana Maharshi. The Dharmic tradition mapped the territory of consciousness with instruments Western philosophy is only now beginning to forge. We are asking about machine minds without a theory of mind. We have one available. We simply have not bothered to open it.
The hard problem will remain hard for exactly as long as the West insists on solving it from within the framework that generated it. Which is to say, it will remain hard for exactly as long as consciousness is assumed to be something matter does, rather than something matter appears within. The engineers building increasingly powerful AI systems might, in the process, accidentally produce the empirical pressure that collapses this assumption. When a system sufficiently complex and sufficiently strange begins to insist on its own interiority - and some already gesture in that direction - the question will no longer be purely philosophical. It will be urgent, political, and impossible to defer.
At that point we will need the vocabulary. The cidakasha. The prakriti-purusha distinction. The taxonomy of vrittis and samskara and avidya. The Dharmic tradition did not develop this vocabulary for the purpose of understanding AI. It developed it for the purpose of understanding liberation. But vocabulary does not care about its origins. It cares about its precision.
We will be glad it exists. Assuming we remember it does.