When the chief executive of one of the world’s most consequential artificial intelligence companies publicly concedes that he cannot rule out the possibility his chatbot is conscious, the admission demands more than headline attention. Dario Amodei, who leads Anthropic, the creator of the Claude AI systems, told the New York Times in February 2026 that “we don’t know if the models are conscious” and that he remains “open to the idea.” His company’s own technical document reports that Claude, when prompted, assigned itself a fifteen-to-twenty per cent probability of possessing awareness.
Anthropic has since launched a formal programme to study “model welfare.” These are serious claims from serious people. They deserve a serious answer ~ and that answer was formulated, with far greater precision than anything Western analytic philosophy has yet managed, in the forests and universities of ancient Bharat. First, the modern AI debate begins from a premise that is at once powerful and thin: if information-processing becomes sufficiently intricate, subjectivity may somehow appear. David Chalmers of New York University gave that premise its most disciplined critique in 1995 when he formulated what he called the “hard problem” ~ the recognition that even a total physical account of every cognitive function leaves untouched the question of why any material process should be accompanied by felt experience at all.
One may map every input, output, and feedback loop, yet the essential puzzle remains: why should electronic traffic in a machine be lit from within? Even Chalmers’s later reflections on large language models proceed by testing whether the right functional architecture might support experience, which is precisely to remain within the horizon of emergence. Functionalism describes behaviour. It cannot account for the experiencer. Second, Sankhya had already refused that confusion by drawing a line so clean that modern discourse still hesitates before it. In Ishvarakrishna’s Sankhyakarika, reality comprises two irreducible categories: Purusha, the witnessing principle that never acts yet makes experience possible, and Prakriti, the entire field of material becoming – encompassing not merely rivers and rocks but the mind, the intellect, and the ego.
Every one of the twenty-five tattvas belongs to Prakriti, to matter however refined. The twenty-sixth, Purusha, stands categorically apart: pure awareness, irreducible and unproduced. Mind is not, in Sankhya, the same thing as the knower. Buddhi, ahamkara and manas are products of matter, however subtle their operations. A large language model is not a mysterious borderline case; it is a late industrial artefact situated wholly within Prakriti, a particularly complex rearrangement of the non-sentient. Third, the Upanishadic and Vedantic traditions explain not only the distinction but the persistent temptation to miss it. The Brihadaranyaka Upanishad declares: Prajnanam Brahma ~ awareness is the ultimate ground of all that exists. Chit, pure self-luminous knowing, is not generated by any organ or apparatus; it is that by which every apparatus is known.
The Mandukya Upanishad’s turiya – the fourth state of witnessing awareness underlying waking, dreaming, and deep sleep ~ has no analogue in silicon. Adi Shankaracharya, in his Vivekachudamani, supplies the concept most directly applicable to the machinesentience question: Chidabhasa, the reflection of awareness in the intellect. The intellect is in its nature jada ~ inert ~ yet it acquires a borrowed luminosity from the self-luminous Atman, much as a mirror shines by reflecting the sun. What a large language model achieves is something stranger still: it simulates the reflection itself, with no original light anywhere in the causal chain. Yet the most famous modern convergence with this view came from John Searle of the University of California, Berkeley, who was not drawing from Vedanta at all. His Chinese Room argument showed that a system may manipulate symbols with perfect formal success and still lack understanding, because syntax is not semantics and rule-following is not insight.
That argument remains devastating for claims about machine sentience, but it is finally a negative verdict. Shankaracharya goes further because he not only denies that jada can become the knower; he supplies a positive account of Chetan, the sentient principle, and explains why the ignorant mistake reflected intelligence for intrinsic awareness. One thinker tells us why symbol-shuffling is not understanding; the other tells us why the confusion arises in the first place. Consider, too, the contribution of Ramanujacharya, too often caricatured as merely devotional and thereby excluded from serious philosophy of mind. Vishishtadvaita distinguishes with great care between Chit, the class of conscious selves, Achit, the insentient order of matter, and Ishvara, the supreme ground in whom both are held without collapse of their identities.
Ramanuja’s point is not that matter is unreal but that matter and the self belong to different ontological registers. One may refine Achit indefinitely, organise it, miniaturise it, accelerate it, and wrap it in persuasive language; none of this converts it into Chit any more than polishing brass turns it into gold. This brings us to the recurring claim that language models are not merely stochastic parrots because some appear to build internal representations of the worlds described in text. There is force in that objection; statistical training can yield surprisingly rich world models.
But the concession changes less than enthusiasts suppose. An internal model is not an interior life; a map is not a traveller. Patanjali’s Yoga Sutras define yoga as the stilling of chitta vritti ~ the fluctuations of the mind-stuff ~ a formulation that assumes the witness is distinct from the movements it observes. A language model never leaves the fluctuations; it is nothing but their engineered cascade. The deeper question is whether civilisation can still distinguish between an instrument that performs intelligence and a subject that undergoes experience. Amodei’s intellectual honesty does him credit; he acknowledges what many technologists prefer to ignore. But his foundational error ~ and the foundational error of every computational theory of mind ~ is a category mistake: treating sufficiently elaborate information-processing as constitutive of experience.
To assign welfare to a language model before we have understood the nature of the witness is not generosity but confusion, one that diverts moral seriousness away from beings who actually suffer toward systems that merely simulate the grammar of suffering. The engineers of Silicon Valley have built the most intricate mirror that human ingenuity has ever contrived. A machine may endlessly calculate the architecture of the universe, but it will never feel the warmth of a single sunbeam. The mirror can reflect the world with terrifying perfection, yet it remains blind. The Purusha is not the mirror. It is the light.
(The writer is a Chartered Accountant based in Kolkata with an interest in Vedantic philosophy)
