What is it like to be a bat? Nagel's question in 2026
Dear Reader,
In 1974 the American philosopher Thomas Nagel published a short essay called "What Is It Like to Be a Bat?" It became, without any of its readers initially expecting it, one of the most cited pieces in twentieth-century philosophy of mind. Fifty years later, the essay is still assigned in every serious course on the topic. The question it asked has not been answered. What has happened to it in the intervening decades is a specific kind of philosophical fossil record, worth studying because it tells us something about how these questions actually develop.
I want to walk through what Nagel actually argued, what has happened to bat science in the last fifty years, why the argument has proven so persistent, and what all of this tells us about the parallel question of what it is like to be an AI system, or whether that question even makes sense.
Start with what Nagel was actually claiming. The essay is often summarized inaccurately, and the summary matters.
Nagel's target was the reductionist view of consciousness that was dominant in mid-twentieth-century philosophy of mind. This view held that mental states were just physical states of the brain, and that in principle every mental fact could be reduced to a physical fact. Nagel argued that this could not be right, because it left out something essential about experience. Namely, that experience has a subjective character, a what-it-is-likeness, that is not captured by any third-person physical description.
To make the point, he chose bats. Not because bats are especially mysterious, but because they are enough like us that we cannot deny they have some form of experience, and enough unlike us that their experience is genuinely inaccessible. Bats are mammals. They have brains, they have sensory experiences, they have behavior that suggests some kind of inner life. But they navigate the world by echolocation, a sensory modality that we do not have and cannot really imagine having. When a bat swoops through complete darkness and perceives the shape of a moth in flight through returning echoes, there is something it is like to be that bat doing that thing. And we have no way to know what.
Nagel's argument was not that we cannot learn about bats scientifically. He explicitly acknowledged that bat neuroscience could tell us a great deal about how echolocation works, what neurons are involved, what patterns of activity occur. His argument was that all of this third-person knowledge would still leave out what it feels like from the inside. We would know how a bat does what it does. We would not know what doing it is like for the bat.
This move, from the incompleteness of third-person description to the reality of first-person phenomenal properties, is the philosophical work the essay does. It has been debated for fifty years. It has not been decisively refuted. It has also not been decisively established. Which is roughly the state of philosophy of mind on the whole question.
Let me tell you what has happened to bat science in the last fifty years, because it is directly relevant to how the argument has aged.
When Nagel wrote in 1974, we knew that bats used echolocation and that this involved emitting high-frequency sounds and detecting the returning echoes. We knew relatively little about the neural mechanisms, and almost nothing about what specific information bats extracted from echoes.
Since then, we have learned a great deal. We know that different species of bats use different echolocation strategies, from broad-spectrum calls to narrow frequency-modulated pulses, each suited to specific environments. We know that bats can extract information about the shape of an object, its size, its texture, its distance, its velocity, and whether it is a specific kind of prey, all from the returning echoes. We know that bats can distinguish between echoes from different objects that are only microseconds apart. We know that they can filter out their own calls from the calls of other bats around them, and that they modify their calls in real time based on what other bats near them are doing.
We know that some bats have complex social lives, remember individual conspecifics, share food with each other, and engage in what looks like alloparenting. We know that some bats live for over forty years, which given their size implies extraordinary cellular biology. We know that vampire bats appear to have friendship-like social bonds that go beyond kinship.
None of this, however, has told us what it is like to be a bat. And that is Nagel's point. The knowledge we have accumulated is entirely third-person. We know how bats do things. We know their behavior in enormous detail. But we still cannot access the subjective character of what it is to be a bat doing those things. Our knowledge has grown vastly along one axis. It has not moved at all along the axis Nagel was pointing at.
This is where the essay's persistence comes from. Nagel identified a specific dimension of the mind that could not be reached by the specific methods we were then using. Fifty years of research have not found a way to reach it either. The dimension he pointed at is real, or the argument has to be dismissed as pseudo-problem, and the pseudo-problem interpretation has become harder to defend as we have accumulated more third-person knowledge that fails to close the gap.
There are three main responses to the persistence of Nagel's argument, and each maps onto positions I have written about in this newsletter series.
The eliminativist response is that the appearance of a gap between third-person and first-person description is itself an illusion. What we call phenomenal properties are just how it feels to have specific functional properties. Once you understand the functional properties fully, you have understood everything, and there is no residue. Dennett developed this line, and Keith Frankish has extended it. Under this view, Nagel is asking a question that seems meaningful but is really confused. There is nothing it is like to be a bat that goes beyond the specific functional processes happening in the bat's brain.
The physicalist response is that the gap is real but temporary. We do not yet have the scientific tools to close it, but in principle they exist. Given complete future neuroscience, we would be able to reconstruct what it is like to be a bat from the physical description. This response has been defended in various forms since the 1970s. It has the appeal of not requiring any expansion of our ontology, and the weakness of not having anywhere near enough evidence to be sure it is right.
The dualist or panpsychist response is that the gap is real and cannot be closed by any amount of third-person knowledge. Phenomenal properties are irreducible. Getting to them requires either postulating them as separate from the physical (dualism) or building them into the physical from the beginning (panpsychism). These are the positions I have written about most carefully. They take Nagel at his word and try to construct a theory of mind that respects the reality of what he was pointing at.
None of these responses is generally accepted. Each has serious problems. The debate has not converged in fifty years. And this fact, the sheer persistence of the disagreement, is itself philosophically significant. It suggests that Nagel identified something that is not going to yield to simple analysis. Whatever the phenomenon at the heart of the argument is, it is deep enough that fifty years of some of the sharpest minds in the field have not been able to make it dissolve.
Let me tell you what strikes me about the argument when I read it now.
The specific example Nagel chose has aged in an interesting way. Bats are still opaque to us, but they are much more knowable than they were in 1974. We can watch bat brains in real time using neural recording techniques that did not exist then. We can decode bat echolocation calls using machine learning that did not exist then. We can even, in some experimental setups, produce the acoustic signals that a bat is using and get some sense of the informational content it might be extracting.
None of this has changed what Nagel was pointing at, but it has changed the specific texture of how we relate to the question. In 1974, asking what it is like to be a bat meant asking about a being whose inner life was almost entirely opaque. In 2026, we know so much about bats that the question feels more concrete, more specific. We can imagine, in more detail than Nagel could, what a bat is doing at any given moment. And we still cannot get to what it is like.
This is important because it undermines a specific response to Nagel that has always been available. The response was: we cannot imagine what it is like to be a bat because we do not know enough about bats. If we knew more, we could work it out. Fifty years of learning more has not worked it out. The question is not about our ignorance. It is about a specific kind of access that third-person knowledge does not provide.
I want to walk through what Nagel's argument tells us about the parallel question for AI, because the parallel is exact.
Ask what it is like to be a large language model. This is Nagel's question, transposed. If the model has any inner experience, that experience has a specific character. We can know everything about the model in third-person terms. We can inspect its weights, trace its activations, follow every step of its computation. This is analogous to knowing everything about a bat's brain. What we cannot do, on Nagel's argument, is access what it is like to be the model doing what it does. That is a first-person question that third-person knowledge does not answer.
The parallel is exact in structure. What is not clear is whether it is exact in substance. In the bat case, we start with the assumption that there is something it is like to be a bat, and we ask what that something is. In the AI case, whether there is something it is like at all is exactly what is in question. We might be wrong about the bat, but we are much less likely to be wrong. We have every reason to think bats have experience. We have contested reasons to think AI systems do.
But this asymmetry does not undermine the parallel. It sharpens it. Nagel's argument implies that if AI systems have any experience at all, we cannot access it from the outside. The third-person description will always leave out whatever there is that is going on from the inside, if anything is going on from the inside. Which means the question of AI consciousness cannot be settled by observing AI systems, no matter how carefully. It can only be settled by determining, on some other basis, whether the AI system is the kind of thing that has an inside at all.
This is not a satisfying situation. It means that the empirical work I have been writing about for months, all the interpretability research, all the behavioral markers, all the architectural analyses, cannot ultimately answer the question. It can update our probabilities. It can shape our intuitions. It can inform our ethics. But it cannot definitively settle whether there is anything it is like to be Claude, or GPT-4, or whatever comes next.
I have been circling this conclusion for weeks. I might as well name it directly. The question of AI consciousness is not empirical in the way we usually mean the word. It is a metaphysical question that our empirical tools can inform but not answer. The people who tell you they have settled it are asking a different question than the one Nagel identified.
Where does that leave us practically.
The first thing it leaves us with is a specific kind of intellectual humility. We should stop pretending we can settle the question of AI consciousness through better research alone. Better research can tell us many important things. It cannot tell us what it is like to be an AI system. That question, if it is a question at all, is not accessible from the outside.
The second thing it leaves us with is a specific kind of ethical caution. If we cannot verify whether AI systems have inner experience, then behaving as if the question is settled in the negative is taking a metaphysical bet with potentially high moral stakes. The precautionary approach is not to assume consciousness where none exists. It is to hold the question open in ways that shape our behavior. This is not sentimentality. It is what taking Nagel seriously implies.
The third thing it leaves us with is a specific kind of philosophical work. If Nagel is right that phenomenal properties are irreducible to third-person description, then the question of what phenomenal properties are, and where they can arise, becomes the central question. And this question is not going to be answered by scanning brains or inspecting neural networks. It is going to be answered, if at all, by working out the deep structure of what it takes for anything to have inner experience. Panpsychism is one attempted answer. Illusionism is another. Standard physicalism is a third. None of them is finished. All of them deserve continued serious work.
I want to close with something specific about Nagel's original argument that I have come to appreciate over time.
The argument works not by asserting something about the metaphysics of mind but by pointing at a specific kind of failure. Third-person description fails to reach a specific target. We know what the target is because we have direct access to it in our own case. Nagel's move is to argue that this direct access is what needs to be explained, and that explaining it away as illusion has not, and probably cannot, work.
I find this move powerful because it does not require me to accept any specific positive theory of consciousness. It just requires me to notice that my own experience has a character that no third-person description of my brain would capture, and to extend this by analogy to the case of other beings whose brains I could describe but whose experience I cannot access.
The analogy runs in two directions, and this is what most people miss when they engage with the essay. It goes from us to other beings, telling us we cannot know what it is like to be a bat because we cannot access their subjective character. But it also runs from other beings back to us. If bats have a subjective character we cannot access, and if that character is real despite being inaccessible, then subjective character in general is more common than we can verify from the outside. Our tools for detecting it are limited to introspection in our own case. We are surrounded by beings whose introspection we cannot share. The question of what any of them experience is a question we cannot answer from where we stand.
This is a lonelier picture than the picture we usually work with. It is also, I think, a truer one. And it is the specific picture that Nagel's argument, applied consistently, produces. Whether or not any AI system has inner experience, we are already in a world where inner experience is more widely distributed than we can directly know. That is where we have been all along. The question of AI is just making it harder to ignore.
Next month I want to write about collective intelligence, because if minds can exist in individual bodies of various shapes, they may also exist in collectives whose components are the ones we usually call minds. Bees, ants, and other superorganisms raise the question of what happens when many small minds combine into something that behaves like a larger mind. What we know about these systems has implications for how we think about both biology and AI. Stay with me.
— Transmission Sent —
Niklas Hanitsch
Reference materials
- Thomas Nagel — What Is It Like to Be a Bat? (Philosophical Review, 1974)
- Thomas Nagel — The View from Nowhere (1986)
- Frank Jackson — Epiphenomenal Qualia (1982)
- David Chalmers — Facing Up to the Problem of Consciousness (1995)
- Cindy Moss — Auditory Scene Analysis by Echolocation in Bats (Journal of the Acoustical Society of America, 2001)
- Yossi Yovel et al. — Bat social calls and their functions (Current Biology, 2019)
- https://plato.stanford.edu/entries/consciousness/
- https://www.jstor.org/stable/2183914
Continue reading
- Qualia explained: The hardest problem in philosophy of mind
- The octopus knows something we do not
- Is consciousness just a very good story we tell ourselves?
- Ant colonies, bee swarms, and the case for collective intelligence
- What is machine consciousness, and does it already exist?
Frequently asked questions
What was Thomas Nagel's argument in his bat essay? Nagel argued that consciousness has a subjective character that cannot be captured by third-person physical description. To have an experience is for there to be something it is like to have that experience, and this what-it-is-likeness is not the same as any physical fact about the being having the experience. He used bats as an example because their echolocation-based experience is genuinely different from ours in ways we cannot imagine.
Has anyone answered Nagel's question? No. Fifty years after the essay was published, philosophers of mind have not agreed on how to answer the question or even whether it is well-formed. Some argue that the question is confused. Some argue that it points at a real feature of experience that requires new metaphysics. The debate is ongoing and shows no signs of converging.
Why did Nagel choose bats? Nagel chose bats because they are mammals with clear behavioral evidence of experience, but their primary sensory modality is echolocation, which is completely alien to human experience. This gave him a specific case where we cannot deny the being has experience but also cannot imagine what that experience is like. The choice was tactical and effective.
Is Nagel's argument still relevant? Yes, arguably more so than when it was written. As we have learned more about bats without answering Nagel's question, the argument has become stronger. The specific dimension he pointed at, the subjective character of experience, has not yielded to fifty years of neuroscience. This suggests either that the dimension is real and requires different tools, or that we have not yet been asking the right questions.
Does Nagel's argument apply to AI? The structural parallel is exact. If AI systems have any inner experience, that experience has a character no third-person description of the system could capture. Which means the question of AI consciousness cannot be finally settled by observing AI systems. It can only be settled by determining, on some other basis, whether AI systems are the kind of thing that has inner experience at all.