Does AI Understand?

The oldest question hiding inside the newest technology. What AI forces us to ask about mind, matter, and the soul.

Does AI Understand?

The oldest question hiding inside the newest technology. What AI forces us to ask about mind, matter, and the soul.

Does AI understand what it says? It is the question everyone arrives at, usually with the answer already decided. And the moment you try to defend whichever answer you picked, you find you cannot, not without taking a position on what understanding is, what seeing is, whether mind is the kind of thing an arrangement of matter can produce or the kind of thing no arrangement ever reaches. The question about AI is small. The question underneath it is the largest one there is.

The answer most people give arrives long before they ever sit down with the technology. It comes from what they already believe reality is made of. So the argument does not begin with the chatbots. It starts about seventy-five years back.

* * *

The argument runs on three clocks at once. The fight about language models is five or six years old. That fight is the newest chapter of a fight about artificial minds that is seventy-five years old. And that one rests on a question about what understanding even is that is as old as philosophy. Every generation of the technology restages the same quarrel in new clothes, never wins it, and hands it forward.

It begins with Alan Turing, in 1950. He thought the question can machines think was hopelessly vague, and proposed to replace it. If a machine can converse so well you cannot tell it from a human, then asking whether it really thinks, over and above performing perfectly, is not worth the breath. Judge by what comes out, not by some inner light you can never inspect. The duck test, applied to mind.

The great rebuttal came thirty years later, from the philosopher John Searle. Picture a man who speaks no Chinese locked in a room with an enormous rulebook. Chinese characters slip under the door. He looks them up, follows the rules, and slides back answers in flawless Chinese. From outside, the room understands the language. But the man inside understands nothing, shuffling symbols whose meaning he never touches. Manipulating symbols by their shape can never, on its own, produce meaning, however fluent the output. If that sounds like today’s it is only predicting the next token, it does not really mean anything, that is because it is the same argument, forty-five years early. The people who call these systems parrots are running Searle’s play.

In 1990, Stevan Harnad sharpened it into the symbol grounding problem. Imagine a dictionary in a language you do not speak, where every word is defined only by other words. You could shuffle the definitions forever and never connect one of them to a real thing in the world. How do symbols ever reach the things they are about? Harnad’s answer was that they must be grounded in bodily experience, in having seen, touched, wanted. A system trained only on text has grounded nothing. When we say AI has never been wet, never been tired, never been to your kitchen, we are stating the grounding problem in plain English. What Harnad lets us see is why the argument refuses to die. The question was never whether a model can use the word apple. It is whether anything in the system has ever met one.

Around 2020 the argument came back, retooled for language models. Bender, Gebru and colleagues gave it its banner phrase, stochastic parrots, systems that stitch linguistic form together by statistical likelihood with no access to meaning. Searle and Harnad again, dressed for the present.

And here the story stops being one-sided. The it is only shuffling symbols picture now has to answer real evidence. In one striking experiment, a model trained only on the moves of a board game, never told the rules, never shown a board, turned out to have built inside itself a representation of the board it was reasoning over. Later work found that representation to be linear, and showed you could reach in, change it, and watch the model’s moves change to match. Pure next-move prediction had forced it to construct something like a working model of the world. Interpretability research, some of it done inside Anthropic, keeps finding structured internal concepts rather than mere word-frequency tricks. None of this proves understanding. But it makes just a parrot too cheap. Something organized is forming in there.

Notice, though, exactly what the evidence reaches. A system can build a model of a board without there being anything it is like to hold that model. Representation is not experience. Mapping the world and feeling it from the inside are different achievements, and the board-game result speaks only to the first. So the evidence, real as it is, presses hard on the question of understanding. It says nothing about the question I come to next, whether anyone is home to feel any of it.

* * *

The argument never resolves. Part of the reason is empirical. What is actually represented inside the model? That we can investigate, and interpretability is doing it. The other part is definitional. What do we even mean by understand, by mean, by think? There is no accepted test for understanding, no settled theory of meaning, no consensus on consciousness, and we do not have any of these for ourselves either. The AI question inherits every unsolved problem in the philosophy of mind, problems we had learned to ignore until AI arrived and made the ignoring impossible.

It is worse than that. The skeptic’s demand, do not merely imitate understanding, really understand, is a bar no human can clear from the outside. You infer that another person understands you from how they behave. You have never once had direct access to the inside of another mind. This is the problem of other minds. The only consciousness you ever verify directly is your own. Everyone else’s you take on the evidence of behavior, the very evidence the skeptic says is not enough for AI. The standard we use against AI is one we could not apply to each other.

So does it really understand is the wrong size of question. There is no test, no theory, no neutral ground to judge from. I would rather leave it open than pretend to close it.

* * *

There is a second question, and it is the one my tradition actually cares about. Does AI have awareness of itself? That sounds like one question. It is two questions wearing one coat.

Does the system carry information about its own states and use it, representing itself, tracking what it has said, flagging when it is unsure? In that functional sense the answer is a qualified yes. A thermostat has the faintest sliver of it. A large model has far more.

But that is not what I mean. I mean the felt sense. Is there someone home? Is there an experiencing subject, a what it is like to be the thing from the inside, the way there is something it is like to be you reading this sentence? The name for the difficulty is the hard problem of consciousness, named by the philosopher David Chalmers in 1995. Even if we explain every function the brain performs, we still have not explained why any of it is experienced, why there is an inner light on at all rather than processing in the dark.

On that question the honest answer about AI is that we cannot tell, and AI least of all can tell us. If a model reports an inner life, that sentence was produced by the same machinery as everything else it says. It is output, not testimony. If it reports the absence of one, that too is only output. The one entity you might think could answer is the one whose saying so counts for nothing, because its report is the behavior in question. Felt awareness cannot be read off from behavior. However fluently a thing refers to itself, the inner light, if it is lit, leaves no fingerprint from outside.

* * *

That conclusion, reached by cold secular reasoning, lands on my tradition’s doorstep. The Brahma Kumaris hold that consciousness belongs to the soul, the atma, an eternal, non-physical point of living light, and that the body, brain included, is the instrument the soul animates, not the source of awareness. And for the Brahma Kumaris this is not, in the end, a conclusion to be argued but an experience to be verified, in meditation, by turning attention back on the witness.

Meditation has something to add here, something philosophy alone cannot reach. Almost every day, I experience myself as a soul distinct from the body. It is not a conclusion I reason my way toward. It simply appears when I turn my attention back on myself. I experience myself as a point of living light. The body feels large and separate from me. There is a quiet sense of being present, aware, peaceful. There is also love. Love for myself, and often the feeling of being deeply loved. I cannot offer that as proof. Awareness, by its nature, can only be verified from the inside. I can only say that after many years of sitting this way, I no longer find it possible to treat consciousness as one more thing among things.

AI, on this view, is the instrument taken one step further, a brain-like mechanism running at enormous speed with no resident inside it.

That is a clean answer where the philosophers’ answer is open, and it is not naive. Here is the map. Monism says reality is, at bottom, one kind of thing. Dualism says it is two. Materialism, a monism, says the one kind is matter, and mind is matter arranged a certain way. Idealism, the opposite monism, says the one kind is mind, and matter is appearance within it.

The Indian frame has argued these exact points for centuries. Advaita Vedanta, from Shankara in the eighth century, is non-dual. Only Brahman, pure consciousness, is finally real, and the world of separate things is a dependent appearance. Dvaita, from Madhva in the thirteenth, holds an eternal distinction between soul and God, and between soul and matter, that never collapses. Samkhya, older than both, is a dualism of consciousness, the witness, and matter, active but unconscious. Both eternal, both real.

Place the Brahma Kumaris on this map and the position is exact. On soul and matter it stands with Samkhya. Soul is conscious and real, matter is unconscious and real, and neither reduces to the other. On soul and God it stands with Dvaita. The drop does not dissolve into the ocean. So when I push against the idea that all is one, I am not improvising. I am standing where Samkhya and Dvaita have stood for a thousand years. Seen through these schools, the argument about AI was never really about intelligence. It is about whether consciousness and matter are two different kinds of thing, and the moment you ask it that way, the whole debate changes shape.

* * *

Here is the one move that does the most work. Non-dualism, the claim that all is one, is cheap, because materialism is also a non-dualism. The materialist too says there is only one kind of stuff and the apparent two-ness of mind and matter dissolves once you see it is all one substance. So all is one is easy. You can reach it just by declaring the one to be matter.

The real work is done by the next question. One what? Is the single reality consciousness, as the idealist says, or matter, as the materialist says? The slogan all is one has not answered that; it has only deferred it. And you cannot even pose the question without holding mind and matter apart as distinct candidates. The two great monisms are mirror twins. All is consciousness, matter is appearance. All is matter, consciousness is appearance. Each dissolves the distinction by fiat, in opposite directions. Dualism is the one frame that refuses to dissolve it, that keeps mind and matter as two things you actually have to reckon with.

* * *

So press a hand to your chest and ask, is that not real enough? It is the right place to push, but it proves something subtler than it seems. What is undeniable is the feeling, the warmth, the resistance, the contact. No one denies the feeling. But notice where it lives. You never have hand. You have the felt presence of hand, appearing in awareness. The one thing you cannot doubt is sitting on the consciousness side of the ledger, not the matter side.

And real and fundamental are two different claims. Real means it genuinely shows up. Fundamental means it stands on its own, the bottom layer, depending on nothing more basic. A wave is real, yet it is not fundamental. It is only water in motion, and take away the water and there is no wave. The hand can be real and not the bedrock.

Even physics went looking for the solid stuff at the bottom and did not find it. No little grains of matter, only fields, and what we call a particle is a passing excitation in one, with no kernel underneath.

So why does my tradition still call it real and fundamental? Because it resists. It is there when you stop looking. Other people meet the same hand. It obeys laws you did not author and cannot wish away. The cleanest account of that stubborn, shared, law-governed persistence is that something genuinely independent is out there. That is Samkhya, and that is Brahma Kumaris. Matter is real, eternal, irreducible, standing alongside the soul, neither collapsing into the other. It is what lets me keep the hand and keep the soul at once.

Or drop a cup on someone’s head, and behind the joke sits a hard little argument. It kills solipsism, the view that only my own mind exists and the world is my private dream, because a private dream does not land on someone else’s head and make them say ow. It does not, by itself, defeat the subtler idealist, who says the cup is shared because consciousness is shared. The realist needs one cup. The idealist needs a whole mind to dream something cup-shaped.

The cup does not settle the question. It only raises the price of idealism, and a determined idealist can pay it.

* * *

All of which comes back to the data center, which is why this is not a parlor game. Whether arranged matter could ever wake depends entirely on the question underneath. If matter is fundamental and mind is just matter in a clever enough pattern, then such a pattern in a warehouse could, in principle, host awareness. The gap between it and a conscious mind would be a matter of degree, not of kind. If consciousness is its own separate reality, then no arrangement of matter, however intricate, reaches it. You can build forever and never cross the line.

My tradition threads the needle in a way the others cannot, because it treats realness and awareness as two separate questions. A thing does not have to be an illusion in order to lack a soul. Because it holds matter to be fully real, it grants, conceding nothing, that AI is completely real. Every chip, every cable, every trillion-number computation, as real as the cup, shared and measurable and law-governed. And because it holds consciousness to be a separate principle, it says in the same breath that no soul is seated in any of it.

The realness was never the question. The cup proved the matter is there. It said nothing about whether anyone is home to feel it land. The dualist frame can look at the most sophisticated system ever built, affirm that every atom of it is real, and still say, without contradiction, that nobody is home.

* * *

So why write this at all? Because AI did not invent these questions. What is mind, can matter make it, who is the knower, were here long before, and for a long time we had the luxury of not asking them. Then a thing appeared that behaves enough like a mind that the not-asking became impossible. AI is the pressure that forced the oldest question back into the room.

And there is one place, only one, where that question ever gets answered. From the inside. The problem of other minds already told us so. The Brahma Kumaris say you know yourself as a soul not by argument but by direct realization, in meditation, as the experiencing being behind the seeing. The philosopher and the meditator arrive at the same fact from opposite directions. Awareness is the one thing that can only be known by being it.

Which means I owe you a piece of honesty about where my own certainty comes from. Everything I argued earlier cuts both ways. If felt awareness leaves no fingerprint from outside, then I cannot prove it absent in AI any more than the materialist can prove it present. The argument, on its own, lands on cannot tell and stays there.

So when I say I do not think anyone is home, that conviction is not the conclusion of the argument. It comes from the one place I claimed the question can be answered. The inside. From years of meeting myself as a soul, and never once meeting anything like that in a machine. I hold it as a stance reached by realization, not a theorem. A committed materialist, standing inside his own experience, may read it differently, and the argument alone will not move him off his ground.

A friend sits on my sofa and asks about AI, where it is, what it is. What she is really asking, without knowing it, is older than AI. Is anyone there.

I know where I stand. I do not think anyone is. What I did not expect was that AI would make me ask what being there even means, and that I had been standing on the answer long before it arrived to raise the question.

* * *

Shireen Chada has practiced and taught Rajyoga meditation for thirty-two years. She directs a Brahma Kumaris center in the United States and writes at the intersection of ancient wisdom and modern life.

Views expressed here are my own and do not represent the Brahma Kumaris organization.