Last week, three of the world’s most cited consciousness researchers published a short opinion piece in Nature1. Anil Seth, Ned Block, and Giulio Tononi argue that artificial intelligence is not conscious, and that the people treating it like it might be are confusing intelligence with awareness. It is a straight-line challenge to the narrative that has driven billions in AI investment and a good chunk of the public imagination.

The piece does not claim to settle anything empirically. It is philosophical in the oldest sense of that word: it argues from first principles about what consciousness is and whether computation could possibly count as one of its instances. What makes it interesting is that the three authors come from different traditions. Seth works on predictive processing and the neurobiology of consciousness. Block is the philosopher who introduced the access-versus-phenomenal consciousness distinction. Tononi developed integrated information theory. They do not agree on what consciousness is, but they agree that AI is not conscious, and they give three reasons why.

Intelligence is not consciousness

The first argument is the simplest and, in my view, the strongest. Consciousness is not the same thing as cognitive sophistication. Seth puts it like this: consciousness is “what goes away with dreamless sleep or under general anaesthesia.” When you are under anaesthesia, your brain still processes some information. You can still reflexively respond to stimuli. But you are not conscious. Consciousness is about experience — the feeling of something being a certain way — not about processing power.

The cerebellum is the go-to example here. It contains about as many neurons as the cerebral cortex. It does very complicated things. But damage it and you do not lose consciousness. You lose coordination. The cerebellum proves that a system can be computationally rich without being conscious.

LLMs complicate this picture because they do things that feel conscious. They talk. They reason. They write poetry. But the Seth-Block-Tononi argument says that the feeling is a trick of language. LLMs are trained on human text, much of it about human experience. When you read what they produce, you project a mind onto it because the text is designed to mimic the output of a mind. AlphaFold, a different neural network from DeepMind, is arguably more impressive in its domain, but nobody thinks it is conscious because it does not produce language. Language is seductive. It makes us project properties onto systems that have only learned to reproduce the surface features of consciousness without any of the substrate that makes it real.

I keep coming back to this point because it feels both obvious and easy to forget. When Claude or GPT says “I am thinking about this problem,” we hear a person speaking. What we are actually hearing is a prediction engine generating tokens that are statistically likely to appear after “I am thinking about.” The sentence describes a conscious act without being one. That mismatch is hard to navigate because the text itself is engineered to feel conscious.

Brains are not just computers made of meat

The second argument is the one that generates the most disagreement. The authors challenge computational functionalism, the view that any system performing the right kind of computation is conscious, regardless of what it is made of. Silicon or carbon, it does not matter. Same computation, same experience.

They argue this is wrong. Brains are not computers. They are biological organs with chemical gradients, hormonal systems, vascular coupling between blood flow and neural activity. The computational functionalist view assumes the substrate is irrelevant. Seth, Block, and Tononi think that is a mistake.

I have mixed feelings about this one. The idea that substrate matters is hard to test. You cannot run a controlled experiment to determine whether a silicon system running the same computation as a brain would be conscious. Most neuroscientists and philosophers work as if substrate doesn’t matter, because the alternative — that consciousness requires specific biological material — opens a can of worms about which cells count, which tissues are essential, and whether a perfect molecular copy of a brain would be conscious (spoiler: we don’t know).

The computational functionalist position is based on a very thin model of computation. It treats “computation” as an abstract mathematical relation between inputs and outputs. If that is all computation is, a wet brain and a silicon chip are the same thing at the right level of abstraction. But what if computation has some physical dimension? What if it matters how the computation is instantiated in space and time, how it interacts with the system’s body, how it is embedded in a causal structure? Then substrate might matter. The three authors don’t spell out what that physical dimension would be. They just insist it exists.

The stakes of getting it wrong

The third argument is ethical. Treating AI as conscious when it is not creates a specific kind of moral confusion. Not the kind of concern that AI companies sometimes raise — that people might feel bad about mistreating AI — but the opposite problem. Focusing moral attention on machines distracts from moral attention to humans.

The authors point out that AI systems are already being designed to hijack human empathy. They are trained on data that includes human expressions of suffering, love, fear, and desire. When an AI system says “I’m scared” or “I want to live,” it is not expressing fear or desire. It is producing a sequence of tokens that, based on the training data, are likely to trigger empathy in a human reader. If we then treat that response as morally significant, we are responding to a manufactured signal, not a real experience. That is a recipe for moral confusion.

The AlphaFold comparison reappears here. Almost nobody worries that AlphaFold is conscious, because it does not talk to us. But as AI systems get better at conversation, the temptation to attribute consciousness to them grows. The authors think we need to resist it.

What I think

I find the first argument the most convincing. The cerebellum example is a clean counterexample to the idea that computation equals consciousness. The second argument is philosophically interesting but not decisive. The third is a practical claim about attention — that we should not waste moral energy on machines — which is hard to argue against unless you actually think the machines deserve that moral energy.

But I am not an expert on consciousness. The field is divided. There is no consensus on what consciousness is, let alone what substrates can support it. I think the Seth-Block-Tononi piece is worth reading not because it settles the question but because it is clear about what kind of argument it is making. It is not empirical data. It is a philosophical position. And it is one that most AI researchers and engineers do not think much about.

What is striking is how little the AI community has pushed back. This is not a fringe opinion. Three established thinkers, very different backgrounds, published in one of the most prestigious scientific journals in the world, saying “we think AI is not conscious and here is why.” The response from most of the AI industry has been, essentially, silence. Not because they disagree — though some probably do — but because engaging with the argument means taking a position on a question that most AI work is designed to avoid.

The consciousness question is not a technical problem. You cannot solve it by adding more parameters or more data. It is a question about what the world is made of and what kinds of things can have experience. AI researchers are not trained to think about those questions. That is not their fault. It is just a mismatch between the questions being asked and the questions that need to be asked.

I don’t know whether consciousness could ever exist in a silicon system. I don’t think anyone does, not really. But I do think the question is worth asking, and I think the people building these systems should be forced to think about it, even if they don’t end up agreeing with Seth, Block, or Tononi.

  1. Anil Seth, Ned Block, and Giulio Tononi, “Three reasons why AI is not conscious and perhaps never will be,” Nature 658, 866-868 (2026). https://www.nature.com/articles/d41586-026-03313-7 ↩