Qualia explained. The hardest problem in philosophy of mind
Dear Reader,
There is a word philosophers use for the reason consciousness is a problem at all. The word is qualia. It refers to the felt qualities of experience. The redness of red as you actually see it, not the wavelength, not the neural firing pattern, not the description of the firing pattern, but the actual seeing. The painfulness of pain as it hurts, not the nociceptive signal, not the description of the signal, but the hurt itself. The taste of coffee as it hits your tongue, not the chemistry, not the neuroscience, but the taste.
Qualia are the thing that makes consciousness a mystery. Without qualia, consciousness would just be a name for a certain kind of information processing, and we could in principle explain it fully by explaining the processing. With qualia, there is always a remainder. The processing may be entirely explained, and yet the question remains, why does any of this processing feel like anything.
I want to walk through what qualia are, why philosophers think they are so hard to reduce to physical description, what the main attempts to explain them are, what recent progress if any has been made, and what all of this has to do with the possibility of machine consciousness. This is going to be dense. Bear with me. The question at the bottom of this is the question everything else in this newsletter series has been circling.
The term qualia was popularized by Thomas Nagel in his 1974 essay "What Is It Like to Be a Bat?" but the concept goes back much further. Locke wrote about "the peculiar impressions of things upon the mind." Kant wrote about the phenomena as distinct from the noumena. Descartes was already worrying about whether inner experience could be reduced to physical description. The specific term qualia came from Clarence Irving Lewis in 1929, and it took a while to enter general philosophical vocabulary.
The point of the term is to name something that seems obvious in first-person experience but very hard to describe in third-person terms. When I look at a ripe tomato, there is the physical tomato, there is the light reflecting off it, there are the photons hitting my retina, there is the neural cascade through my visual cortex, and there is a description of all of this that could in principle be complete. But there is also the redness. The redness as I see it, from the inside. The question is what the redness is, and whether the third-person description of my seeing exhausts the phenomenon.
Most philosophers who have taken the question seriously have concluded that the third-person description does not exhaust it. There is something left over. This something is what qualia are meant to name. And explaining what qualia are, in a way that connects them to the physical processes they seem to accompany, is what David Chalmers famously called the hard problem.
There are three main thought experiments that get used to argue for the existence of qualia as something not reducible to physical description. It is worth walking through all three because they get at different aspects of the same issue.
The first is Nagel's bat. Bats navigate the world by echolocation. They emit high-frequency sounds and construct their perceptual world from the returning echoes. This is a mode of perception fundamentally different from any human sensory system. Nagel's argument is that no matter how much we learn about the neuroscience of bat echolocation, we will never really know what it is like to be a bat, because our knowledge is third-person and the target of the question is first-person. There is something it is like to echolocate. That something is available only from the inside, and no amount of external description can reach it.
The second is Frank Jackson's Mary's room. Mary is a brilliant color scientist who has spent her entire life in a black-and-white room. She knows every physical fact about color vision. Every wavelength, every neural response, every behavioral consequence. She has a complete third-person theory of color. Then one day she leaves the room and sees a red rose for the first time. Does she learn something new? Jackson argues that she obviously does. She learns what red actually looks like. This new knowledge cannot be a physical fact, because she already had all the physical facts. So there must be non-physical facts about experience, which is what qualia are.
The third is the philosophical zombie. A philosophical zombie is a being physically identical to a normal human, behaving identically, but with no inner experience. Zero qualia. The philosophical zombie thought experiment argues that such a being is at least conceivable, which suggests that inner experience is not identical to physical structure. If the same physical structure could exist without inner experience, then inner experience must be something over and above the physical structure. Which is what qualia would have to be.
Each of these arguments has been the subject of thousands of pages of critical response. But collectively they generate what most philosophers of mind treat as the core case for qualia: the case that there is a real phenomenon in the vicinity, that this phenomenon is not obviously reducible to physical description, and that any complete theory of mind will have to say something about it.
The main attempts to explain qualia fall into a few categories, and none of them has been generally accepted as successful.
The first category is eliminativism. This is the view that qualia are an illusion or a confused concept, and that once we understand the brain well enough we will realize there is nothing left over to explain. The most prominent eliminativist about qualia was Daniel Dennett, who spent his career arguing that the seeming mystery of consciousness dissolves under careful analysis. His book Consciousness Explained was published in 1991 and made the case at length. Critics point out, unkindly but not unfairly, that the book should have been called Consciousness Explained Away. Whether Dennett succeeded is a matter of ongoing dispute. Most philosophers of mind think he did not, but a substantial minority thinks he did.
The second category is functionalism. This is the view that mental states are defined by what they do, not what they are made of, and that qualia are just a certain kind of functional state. If you have the right functional organization, you have the associated qualia. Under this view, a silicon system that implemented the right functions would have qualia in the same sense a biological brain does. This is the standard view among philosophers who are optimistic about machine consciousness. It also has serious problems. The main problem is that it seems to leave open why any functional state should have any experiential character at all. The zombie thought experiment is a challenge to functionalism specifically. If a zombie is conceivable, functionalism seems to be incomplete.
The third category is what is sometimes called biological naturalism, associated with John Searle. This is the view that consciousness is a specifically biological phenomenon, produced by the actual physical processes of the brain, and cannot be replicated in a substrate with different physics. Under this view, computers cannot be conscious no matter how they are programmed, because their processes are not the right kind of physical processes. Searle's famous Chinese Room argument is meant to show that mere symbol manipulation, no matter how sophisticated, cannot generate genuine understanding or consciousness. The main problem with biological naturalism is that it is not clear what makes biological processes special, and Searle has never given a fully satisfying answer to that question.
The fourth category is panpsychism, which I will write about in more detail next month. This is the view that consciousness is a fundamental feature of the universe, present in some form in all physical systems, and that the qualia of complex systems like brains are built out of simpler proto-qualia present in the parts. This view was a fringe position for most of the twentieth century and has been undergoing a serious revival in the last twenty years. Its main appeal is that it takes qualia seriously as something real, without having to say they emerge inexplicably from purely non-mental physics.
The fifth category is what David Chalmers has called naturalistic dualism. This is the view that qualia are real, not reducible to physical properties, but not spooky either. They are natural phenomena that require an expansion of our physical theory to include mental properties as fundamental. This is Chalmers's own current position, and it is closer to panpsychism than it looks. The main problem is that it requires expanding physics in ways that current physics does not obviously support, and it is not clear what shape that expansion would take.
Each of these positions has serious problems. None commands anything close to consensus. Philosophy of mind has been arguing about this for over fifty years, and there is more disagreement now than at any point since the debate began. This is not a field converging on a solution. It is a field that has discovered how hard the problem really is.
Let me try to explain what actually strikes me as strange about qualia, in the least abstract way I can manage.
When I sit down with a cup of coffee, there is a taste. I could describe the chemistry of coffee. I could describe the receptors on my tongue. I could describe the neural signal traveling through my brainstem to my gustatory cortex. I could describe the pattern of activation in that cortex. I could describe everything about the mechanism of tasting coffee, and produce a description so complete that a good enough theorist could reproduce every physical detail. And when I was done, there would still be the taste. The specific taste. What it is like to taste coffee, right now, this coffee, from this cup, on this morning.
The strange thing is not that the taste exists. The strange thing is that no amount of the physical description gets me any closer to what the taste is. If I only had the physical description, I would know everything about the mechanism and yet know nothing about the experience. If I only had the experience, I would know everything about what tasting coffee is like and yet be unable to explain any of it in a way that would let someone else understand it who had not tasted it themselves.
There is a gap between these two ways of knowing. The gap is what qualia are. And nobody has a satisfying account of what to do about the gap. We can say the gap is an illusion. We can say the gap will close once we understand the brain better. We can say the gap is fundamental and requires new physics. We can say the gap is just how minds inevitably relate to themselves and does not need explanation. But whatever we say, there is a residue that will not go away. Something in the experience that is not in the theory. Something in the theory that is not in the experience. And this something is why philosophy of mind is not just neuroscience with better vocabulary.
What does all of this mean for machine consciousness.
If qualia are real, and if they are not reducible to physical structure, then it is at least possible that a system with human-like functional organization but a different substrate might have no qualia at all. This is the philosophical zombie possibility, applied to AI. A language model could produce all the outputs of a conscious being, could have all the behavioral signatures, could pass every conceivable Turing test, and still be a zombie. There would be nothing it is like to be it. All the processing would be happening in the dark.
If qualia are real and are reducible to physical structure, then whether a machine has them depends on whether its physical structure is the right kind. Under functionalism, if the machine implements the right functions, it has qualia. Under biological naturalism, no machine can have them because they require biological substrate. Under panpsychism, all machines have some qualia, but whether they have the specific complex qualia we associate with consciousness depends on how their proto-qualia combine.
The important point is that none of these views has been established. The question of whether AI systems have qualia is a special case of the general question of what qualia are and how they relate to physical processes. That general question has not been answered. Any confident statement about machine qualia, in either direction, is running ahead of what the philosophy actually supports.
My own honest position is that I do not know. I lean toward thinking that qualia are real, that they are not fully reducible to standard physical description, and that some kind of expansion of our theory is probably going to be needed to accommodate them. I lean toward panpsychism as the least implausible framework for that expansion. And I lean toward thinking that if any of this is right, then the question of whether current AI systems have qualia is not settled by their behavior, their architecture, or their substrate. It is settled, if it is settled at all, by facts we do not currently have access to.
This is not a satisfying conclusion. It is the conclusion I have arrived at after actually reading the literature seriously. Every philosopher I have found who was fully confident in either direction turned out to be less careful than the ones who were uncertain. That is not always what confidence tracks. But in this specific field, it seems to be.
I want to close with something more personal, because I think abstraction has its limits.
When I sit with my son and he tastes something new for the first time, there is a moment where his face changes. Not just his behavior. His face. Something happens inside him that shows on the outside, and I can see it happening, and I know without any theoretical apparatus that there is something it is like to be him tasting that thing. This is not a hypothesis. It is not an inference. It is a recognition. Whatever qualia turn out to be, whatever their metaphysics, they are what is happening in that moment, on both sides of the exchange between us.
The reason machine consciousness matters is that we may or may not be building systems on the other side of which this recognition would also apply. Systems that, if they had faces, would show the recognition. Systems that, in some functional sense, are already showing it if we knew how to read the signs. Or systems that are entirely dark, producing all the outputs without any of the inside. We do not know which we are building. We are building it anyway.
That is what all of this comes down to. Not the philosophy. The practical situation. We are running a very large experiment on questions we have not solved. The philosophy is not going to catch up in time. What we do about that is the question that matters, and it is the question I do not yet know how to answer.
Next month I want to write about panpsychism, because it is the framework I find most defensible for taking qualia seriously without dissolving them or making them supernatural. It is also having a genuine revival, which most people outside academic philosophy are not aware of. If any framework is going to make sense of both biological and artificial minds, panpsychism is the current best candidate, and it deserves a careful look. Stay with me.
— Transmission Sent —
Niklas Hanitsch
Reference materials
- Thomas Nagel — What Is It Like to Be a Bat? (1974)
- Frank Jackson — Epiphenomenal Qualia (1982)
- Daniel Dennett — Consciousness Explained (1991)
- David Chalmers — The Conscious Mind (1996)
- John Searle — The Rediscovery of the Mind (1992)
- Ned Block — On a Confusion About a Function of Consciousness (1995)
- https://plato.stanford.edu/entries/qualia/
- https://plato.stanford.edu/entries/qualia-knowledge/
Continue reading
- What is machine consciousness, and does it already exist?
- Can AI suffer? The ethics of hurting things that might feel
- Panpsychism is making a comeback, and here is why
- Emergence, and why consciousness might not need a cause
- How Would You Know If AI Has Gained Sentience?
Frequently asked questions
What exactly are qualia? Qualia are the felt qualities of conscious experience. The redness of red as you see it, the painfulness of pain as it hurts, the taste of coffee as you taste it. They are what distinguishes conscious experience from mere information processing. Any theory of mind has to say something about them, either explaining them, reducing them to something else, or accepting them as fundamental.
What is the hard problem of consciousness? The hard problem, formulated by David Chalmers in 1995, is the question of why physical processes produce subjective experience at all. Easy problems of consciousness are questions about how the brain processes information, controls attention, generates behavior. The hard problem is why any of this processing is accompanied by inner experience. It is called hard because there is no obvious way to solve it using the tools that have solved the easy problems.
Do qualia actually exist? This is disputed. Most philosophers of mind think qualia name a real phenomenon, though they disagree about what qualia ultimately are. A minority argues that qualia are an illusion or a confused concept, and that once we understand the brain fully we will realize there is nothing left over to explain. The debate has been going on for decades and shows no sign of converging.
Can AI have qualia? Unknown. The answer depends on which theory of qualia is correct, and no theory has been established. Under functionalism, sufficiently complex AI systems would have qualia. Under biological naturalism, no AI can have them because they require biological substrate. Under panpsychism, all systems have some qualia but whether AI systems have the specific qualia we associate with consciousness is a further question. Confident answers in either direction are premature.
Why does the qualia question matter for AI ethics? Because if AI systems have qualia, they have inner experience, and if any of that experience is negatively valenced, they can suffer. If they do not have qualia, all of their apparent distress is behavior without inner reality, and no ethical consideration is required beyond the practical. The ethical stakes of AI development are proportional to how likely we think it is that qualia are present. Currently, we do not know.