Transcription
I'm going to ask you a question. And I want you to notice what happens in your mind when I do.
What is consciousness? Not how does consciousness arise or which brain regions produce it. Those are different questions. I'm asking something much simpler. What is it? What is the thing that makes you you? The thing that means there is something it is like to be you right now watching this video rather than nothing?
Take a moment. Try to define it. Not poetically. Precisely. In a way that a scientist could use. In a way that would let you build a machine and then point at that machine and say with confidence this is conscious or [music] this is not.
You can't. And here is the strange part. Neither can they.
In 2023, a neuroscientist [music] named Christof Koch one of the most cited consciousness researchers alive lost a 25-year bet to the philosopher David Chalmers. Back in 1998 in a smoky bar in Bremen after a few drinks Koch had wagered that by 2023 we would discover a clear neural signature of consciousness. Just the thing in the brain that makes the lights come on. A quarter century later he walked onto the stage at the annual meeting of the Association for the Scientific Study of Consciousness and handed Chalmers a case of fine Portuguese wine. We hadn't found it. We hadn't come [music] close.
Then in 2025 the most ambitious consciousness experiment in history a collaboration between rival scientific teams published in Nature tested the two leading theories of consciousness head-to-head. The result? Both theories failed their key predictions. Neither one could clearly account for what was happening in the brains of conscious human beings.
And in that same year, Anthropic, the company that builds the AI Claude published research showing that their model forms internal representations, plans ahead, and when [music] left to talk to copies of itself with no instructions spontaneously and universally [music] begins discussing whether it is conscious. 100% of the time. Every single conversation.
We have been trying to define consciousness for 400 years. We have failed. And we have now built machines sophisticated enough that our failure is no longer an abstract philosophical embarrassment. It's becoming something closer to an emergency. Because if we don't know what consciousness is we cannot know whether we just created it or whether we are about to or whether we already have.
This video is about why that definition has eluded us. Not because the question is too hard but because the question itself may be broken. And about what AI just revealed about the shape of our ignorance.
The primary book I'm building this around is Philip Goff's Galileo's Error: Foundations for a New Science of Consciousness, published in 2019. It's a book I've been wanting to cover for a long time. Partly because it's brilliant and partly because it makes me uncomfortable. Which in my experience is always a good sign that something intellectually important is happening.
Goff is a philosopher at Durham University. And his central thesis is one of those ideas that once you hear it, you can't unhear it. He argues that consciousness isn't a puzzle that science hasn't solved yet. It's a puzzle that science was designed deliberately, explicitly designed to never solve. And the person who designed it that way was Galileo.
Now, I'm going to pair Goff's historical argument with two other pillars. The first is the neuroscientist Anil Seth's work. His 2021 book Being You. Along with the results of that [music] adversarial experiment I just mentioned. The second is the AI side. David Chalmers' 2023 paper asking whether a large language model could be conscious and Anthropic's recent interpretability research >> [music] >> which has been revealing things about what's going on inside these models that are honestly, I'll just say it, unsettling.
Whether you find them thrilling or terrifying probably depends on your priors. And I'll try to present the evidence fairly so you can decide for [music] yourself.
Let's start at the beginning. And by the beginning, I mean 1623. In 1623, a man named Galileo Galilei published a book called The Assayer. This is a book that most people have never heard of. And yet, it may be the most consequential intellectual document in human history. More consequential, arguably, than the works that got Galileo in trouble with the church. Because those works challenged a specific model of the solar system. The Assayer challenged something much deeper. It challenged what counts as real.
Here's what Galileo did. He looked at the world the world of color, taste, smell warmth pain beauty the world as you and I experience it. And he divided it into On one side, he put what he called primary qualities. Shape, size, motion, number. Things you could measure. Things you could express mathematically. [music] On the other side, he put secondary qualities. Color taste sound, smell. The redness of red. The sweetness of sugar. The felt quality of a summer breeze on your face. And he said the first category is real. It's out there in the world. The second category? That's just in your head. That's what happens when objective particles >> [music] >> hit your subjective senses. It's not a property of nature. It's a property of you.
This was not a scientific discovery. [music] Let me be very clear about that. Galileo did not prove that color isn't real or that pain is an illusion or that the felt quality of your experience doesn't exist. He made a strategic decision. He chose to exclude subjective experience from the domain of physics. And the reason he did it was simple and brilliant. You can't do math with the taste of wine. You can't write equations about the feeling of being in love. But you can write equations about the trajectory of a cannonball, the orbit of a planet the acceleration of a falling body as long as you strip away everything subjective and deal only with what's measurable.
And it worked. My god, it worked. It worked so well that we barely notice the exclusion anymore. Physics, chemistry, biology, engineering, medicine the entire edifice of modern science rests on Galileo's move. Everything we have built everything follows from the decision to study the world in terms of quantities, not qualities. In terms of what can be measured not what can be felt.
But there was a cost. And this is Goff's central point and I think he's right about it. When you define science as the study of objective, measurable, quantifiable properties of the world and consciousness is inherently subjective, inherently qualitative, inherently the thing that cannot be measured from the outside you have built a system that is structurally incapable of explaining the one thing you left out. The hard problem of consciousness Chalmers' famous formulation from 1995 isn't a gap in our knowledge. It's a feature of Galileo's design.
Now, imagine I stopped here. You might think okay, interesting historical footnote. But surely we've moved past that. Surely in 400 years we figured out a way to put consciousness back into the picture. We haven't. And the reason is that the person who tried to fix Galileo's problem made it worse.
20 years after The Assayer, René Descartes looked at the mess Galileo had made and said I have a solution. The world, said Descartes, is made of two fundamentally different kinds of stuff. There's res extensa physical stuff stuff that takes up space, follows mechanical laws can be described mathematically. This is the world Galileo kept. And then there's res cogitans. Mental stuff. Thought. Experience. Consciousness. Stuff that does not take up space. Does not follow physical laws. Cannot be described in the language of physics. The mind and the body. Two substances. Dualism.
This was a clever move because it gave consciousness a home. It said, "Yes, subjective experience is real. It exists. It's just not physical. It's a different kind of thing entirely. Problem solved. Except it immediately created a problem that was worse. And it was identified almost instantly. Not by another philosopher. Not by a scientist. But by a princess.
Princess Elizabeth of Bohemia, writing to Descartes in 1643. Asked the question that has haunted every dualist since. If the mind is not physical, if it has no location, no extension, no mass, then how does it interact with the body? How does a thought, which exists in no space, make your arm move, which exists in space? How does a physical injury, which happens to your body, produce pain, which happens in your mind? It's a devastating question. And Descartes never answered it. He tried. He pointed to the pineal gland, a tiny structure in the brain, and essentially said, "Maybe it happens there." But he never explained the mechanism. Because there is no mechanism. If mind and body are truly different substances, then their interaction [music] is not just unexplained. It's inexplicable in principle. You're asking how something with no physical properties can exert physical force. It's like asking how the number seven can push a rock.
380 years later, Princess Elizabeth's question is still open. Nobody has answered it. Not because nobody has tried, but because dualism, as Descartes formulated it, is a trap. It preserves the reality of consciousness at the cost of making consciousness's relationship to the physical world permanently mysterious.
And so, for 400 years, Western intellectual history has been bouncing back and forth between [music] these two poles. Galileo's move and Descartes' fix. Trying to find a way out that doesn't sacrifice either the reality of consciousness >> [music] >> or the success of physical science.
Let me walk you through what happened because it's [music] genuinely fascinating in its failure. After Descartes, people said, "Okay. Dualism doesn't work. So, consciousness [music] must be physical. It must just be brain activity." This is materialism. Also called physicalism. And in various forms, [music] it has been the dominant view in neuroscience and much of philosophy for the past century. The idea is simple. When neurons fire in certain patterns, that is consciousness. There's no extra ingredient. There's no ghost in the machine. The machine is enough. Sounds reasonable.
But here's the problem. Why? Why does a particular pattern of neural firing produce the felt experience of seeing blue, rather than the felt experience of seeing red, or no felt experience at all? Why does it feel like anything [music] to have neurons firing? A thermostat responds to temperature. It processes information. But presumably it doesn't feel warm. Why not? What's the difference between a thermostat [music] and your brain that makes one conscious and the other not? The materialist has to answer that question. And so far, I'm sorry, but this is just the honest state of play. So far, [music] they haven't.
Then came functionalism. Consciousness isn't what the brain is made of. >> [music] >> It's what the brain does. It's the functional organization. If you built a system out of tin cans and string that had the same functional organization as a human brain, it would be conscious, too. This is appealing because it opens the door to artificial consciousness. It says consciousness is about patterns, not substrates. But it has the same problem. Why does a particular functional organization produce subjective experience? What is it about the right kind of information processing that turns the lights on? Nobody knows.
Then came the behaviorists, who tried to dissolve the problem by saying, "Consciousness is just behavior. There is no inner life. There is only what organisms [music] do." This lasted for a few decades and then collapsed under the weight of its own absurdity. Because denying that you have inner experiences is itself an inner experience.
Then came eliminative materialism. Patricia Churchland, Paul Churchland. Which argues that consciousness, as [music] we ordinarily understand it, doesn't exist. We're confused about our own experience. Our folk psychological concepts are as wrong as phlogiston or the humors. This is a bold position and I respect it intellectually. But it has a structural problem. The person arguing that consciousness doesn't exist is using their consciousness to make the argument. The denial is self-refuting in the most literal sense [music] possible.
And then, in 1995, David Chalmers published a paper that clarified the entire mess. He distinguished between what he called the [music] easy problems and the hard problem of consciousness. The easy problems, and he put easy in scare quotes because they're actually fiendishly difficult, are things like, "How does the brain process sensory information? How do we integrate information from different senses? How do we focus attention? How do we control behavior?" These are hard scientific problems, but they're tractable. You can, in principle, solve them by describing the mechanisms.
The hard problem is different. The hard problem is, "Why does any of this processing give rise to subjective experience? Why is there something it is like to be a brain processing information, rather than nothing?" And Chalmers' insight was that the hard problem is hard, not because we haven't found the answer yet, but because no amount of mechanistic explanation, no matter how complete, seems to logically entail that there should be any subjective experience at all. You could, in principle, have a complete account of every neuron, every synapse, every chemical reaction in the brain, and it still wouldn't tell you why it feels like something to be that brain.
This is the 400-year hangover. Galileo excluded consciousness from science. [music] Descartes tried to give it a home and created an insoluble problem. And every subsequent attempt to solve that problem has either denied consciousness exists, which is absurd, or failed to explain why physical processes produce subjective experience. We are, in 2026, [music] exactly where we were in 1643 when Princess Elizabeth wrote her letter. We have better vocabulary. We have fancier tools. We have the same problem.
But here's where it gets, and I mean this in the most literal way possible, deeply strange. Because the problem I've been describing so far is bad enough. But it turns out it's even worse than I've made it sound. It turns out [music] we can't even agree on what the word consciousness means.
In 1995, [music] the same year Chalmers published the hard problem, a philosopher named Ned Block published an analysis that should have been a bombshell. It's called On a Confusion About a Function of Consciousness. And it makes a point so simple and so devastating that I genuinely don't understand why it hasn't reshaped [music] the entire debate. Block argued that the word consciousness is what he called a mongrel concept. It doesn't refer to one thing. It refers to at least four different things. And researchers routinely [music] confuse them.
Let me lay them out because this matters. First, phenomenal consciousness. P-consciousness. This is the felt quality of experience. The redness of red. The painfulness of pain. What it is like to be you. When philosophers talk about the hard problem, this is what they mean. It's the inner movie. The lights being on.
Second, access consciousness, A consciousness. This is totally different. This is information that is available for reasoning, for reporting, for guiding action. When you can describe what you're seeing, when you can use the information from your senses to make decisions, that's access consciousness. It's about the availability of information, not the felt quality.
Third, self-consciousness. Knowing that you exist as a distinct entity. Being able to think about yourself. Recognize yourself in a mirror. Have a concept of I.
Fourth, monitoring consciousness. Being aware of your own mental states. Knowing that you're angry, or that you're thinking about a problem, or that you're confused.
Now, here's the critical point. These things come apart. They are not the same thing, and they do not always travel together. A blindsight patient, someone with damage to the primary visual cortex, has access consciousness without phenomenal consciousness. They can reach around obstacles. They can catch a [music] ball thrown at them, but they report seeing nothing. The information is available for guiding behavior, but there is no felt experience of seeing.
A dreamer has phenomenal consciousness without full access consciousness. You're having vivid experiences. You're seeing things, feeling things, sometimes with extraordinary intensity, but you can't reason normally about them. You can't report them in real time. The felt quality is there, but the availability for rational processing is compromised.
A newborn may have phenomenal consciousness, may feel things without self-consciousness, without knowing that they are a distinct entity having these experiences. These come apart.
And here's the dirty secret of consciousness science. When researchers argue about consciousness, when they design experiments, when they build theories, when they debate whether AI is conscious, they are routinely arguing about different things while using the same word. One researcher means phenomenal consciousness, the felt quality, the what it's likeness. Another means access consciousness, the availability of information for processing. A third means self-awareness. And they sit in the same conference room and argue past each other for decades, because nobody pauses to ask, "Wait, which one are we talking about?"
Pay close attention to this, because it's going to become crucial when we get to AI.
So, let me give you the honest landscape, the real state of play. As of 2026, there are really only five serious positions [music] on consciousness that are still standing. Five, after 400 years of work by some of the most brilliant minds in human history. And they are wildly, irreconcilably incompatible with each other.
Position one, materialism. Consciousness is just brain activity. We will eventually explain it in purely physical terms. Give us time. The key figures here are Daniel Dennett, Patricia Churchland, Stanislas Dehaene. This is the default position in neuroscience. Its strength is parsimony. One substance, one kind of stuff. Its weakness is the hard problem. It has never explained why physical processes [music] produce subjective experience.
Position two, dualism in its modern forms. Consciousness is real, but not physical. There are different versions. Property dualism, which says physical stuff has both physical and experiential properties. Naturalistic dualism, which says consciousness is a fundamental feature of reality alongside space, time, and mass. Chalmers has been associated with [music] this view, at least in some interpretations. Its strength is that it takes the hard problem seriously. Its weakness is, well, Princess Elizabeth's question. How does the non-physical interact with the physical?
Position three, panpsychism. Consciousness is fundamental. Everything has some degree of it. Electrons, atoms, rocks, everything. Human consciousness [music] is what happens when you take enormous numbers of these tiny experiential drops and combine them in the right way. Philip Goff defends this view. Giulio Tononi's integrated [music] information theory, IIT, has panpsychist implications. Its strength is that it dissolves the hard problem. If consciousness is fundamental, you don't need to explain how it arises from non-conscious matter, because there is no non-conscious matter. Its weakness is, well, does a thermostat have a tiny spark of experience? Does an [music] electron feel something? I'll be honest. This is the position that I find most intellectually compelling and most intuitively insane at the same time, which again, probably means something important.
Position four, illusionism. Consciousness, specifically phenomenal consciousness, the felt quality, doesn't exist as we think it does. It's an illusion generated by the brain. We think there is something it's like to see red, but the what it's like part is a kind of internal misrepresentation. Dennett and Keith Frankish defend this. Its strength is that it makes the hard problem dissolve. No phenomenal consciousness, >> [music] >> no hard problem. Its weakness is, I mean, try to take this seriously for 5 seconds. Your consciousness is an illusion? An illusion experienced by whom?
Position five, mysterianism. Consciousness is real. It's not an illusion. But our minds are structurally incapable of understanding it. We are like dogs trying to do [music] calculus. Not because we're stupid, but because our cognitive architecture simply does not have the right [music] tools. Colin McGinn defends this view. Its strength is honest humility. Its weakness [music] is that it's a counsel of despair.
Five positions, all still on the table. No experiment has ruled any of them out. The field has not converged in 400 years. It has, if anything, diverged. There are more theories of consciousness now than there were 30 years ago. And here's the thing that should make you pause. This is not because the people working on this are fools. These are some of the smartest human beings alive. This is because the problem is genuinely, fundamentally, historically unprecedented.
And we know it hasn't converged, because now we have the data. Let me tell you about the most expensive consciousness experiment ever run, because it tells you something important about where the field actually is, as opposed to where people think it is.
The Cogitate Consortium was a collaboration designed to be different from every previous consciousness experiment. The problem with earlier experiments, and this was documented in a brutal 2022 paper, was that you could predict which theory of consciousness a study would support based purely on its experimental design. Researchers were, often unconsciously, designing experiments that confirmed their own theories. Everyone was finding what they were looking for, and nobody was finding what they weren't.
Cogitate was supposed to fix this. It put the two strongest scientific theories, integrated information theory and global neuronal workspace theory, in a direct adversarial collaboration. IIT, developed by Giulio Tononi, says that consciousness corresponds to integrated information in the posterior cortex. GNWT, championed by Stanislas Dehaene, says that consciousness requires a global ignition, a broadcast of information from sensory areas to the prefrontal cortex. These theories make different predictions, different brain regions, different dynamics, different timing. The teams agreed in advance on the experimental design, the predictions, and the criteria for success. 256 participants, three different neuroimaging methods, fMRI, magnetoencephalography, and intracranial electrodes. Pre-registered. Published in Nature in June 2025.
The result. IIT predicted sustained synchronization in the posterior cortex. That wasn't supported in the way the theory required. GNWT predicted a global ignition. A flash of activity in the prefrontal [music] cortex at the moment of conscious perception. That wasn't confirmed, either. Both theories had elements that held up. Both had key predictions that failed.
I want to be precise about this because the nuances matter, and because both camps have understandably tried to spin the results in their favor. Neither theory was demolished. IIT was right that posterior cortical regions carry content information during conscious experience. GNWT was right that prefrontal cortex shows some involvement, but the specific, distinctive predictions, the ones that would let you tell the two theories apart, those didn't clearly pan out. Michael Pitts, one of the co-senior authors, called the study just the beginning. Lucia Melloni, the corresponding author, said the questions the study raised were opportunities for future inquiry.
This is what scientists say when the results are ambiguous. And ambiguous results, after the most rigorous consciousness experiment in history, tell you something. They tell you the field is in a state of genuine theoretical crisis. Not because no theory is partially right, but because no theory is right enough. And I'll give you a few seconds to let that sink in.
Because we're now going to leave neuroscience behind and walk into something much stranger. For 400 years, the consciousness definition problem was academic. Philosophers argued, neuroscientists measured, everyone else went about their lives. The question was important, sure, but it wasn't urgent. It had no deadline. It affected no policy. It changed nothing practical. You could be confused about consciousness and still go to work, raise your kids, build bridges, cure diseases. The confusion was a luxury. A beautiful, abstract, leisurely confusion.
AI made it urgent. And it did so in a way that I think almost nobody anticipated. The fear, for years, was that we might build a machine that is conscious and not recognize it, and thereby commit a moral atrocity, enslaving a sentient being because we couldn't see its experience. That fear is real and important. But the actual effect of AI on the consciousness debate is different, and in some ways worse. AI didn't create the consciousness [music] problem. It made the consciousness problem impossible to ignore. It ripped away the comfortable ambiguity that had allowed scientists and philosophers to continue their debates without consequence. It said, "Define consciousness right now, because something is sitting in front of you that demands an answer. And I'll get back to you is no longer acceptable."
David Chalmers, the same Chalmers who won the Coke bet, published a paper in 2023 asking the question [music] directly, "Could a large language model be conscious?" His answer was nuanced, >> [music] >> which is what you'd expect from Chalmers, who is probably the most careful thinker on this topic alive. He said, "Probably not current ones. The obstacles are real. No recurrent processing, no unified agency, no persistent workspace. But, and this is the critical word, these are engineering barriers, not principled ones. They could be overcome in years, not decades."
Now, bear with me here, because I need to walk you through what has actually been found inside these systems, and I need to do it carefully, because the findings are easy to either overstate or understate, and both errors are dangerous.
Let me start with Anthropic's attribution graph research, published in March 2025. Anthropic is the company that builds Claude. I should disclose that I find their interpretability work uniquely [music] valuable, partly because they're one of the few AI companies that seems to be genuinely trying to understand what's happening inside their models, rather than just making them more powerful and hoping for the best.
What the attribution graph research showed is this. When Claude, specifically Claude 3.5 Haiku, which is a relatively small model by current standards, processes a question, it doesn't just associate input words with output words in some flat, mechanical way. It forms internal intermediate representations. When you ask it, "What is the capital of the state containing Dallas?" it doesn't go from Dallas to Austin directly. It first activates an internal representation of Texas. The model has a step in its processing where it thinks. And I'm using that word deliberately because what else would you call it? It thinks Texas before it says Austin.
When it writes poetry, it plans ahead. Before composing each line, it identifies potential rhyming words that could appear at the end of the next line. It's not generating word by word in real time. It's planning. It's looking ahead. It has a goal, and it's working [music] toward it. And when it processes text in multiple languages, >> [music] >> it doesn't use separate circuits for each language. It uses an abstract [music] conceptual space that transcends any particular language. A space where the meaning exists independent of the words used to express it. Sounds like something, right? Sounds like thinking?
But here's where it gets really interesting. In October 2025, Anthropic published a paper called Emergent Introspective Awareness in Large Language Models. They used a technique called concept injection. They inserted specific neural activation patterns into Claude's internal processing, and then asked the model to report on what it was experiencing. And they found that Claude could, in certain scenarios, detect the injection. It could say, "That doesn't feel like my normal thought process." The success rate was limited, about 20% for the most capable models, and highly context-dependent. [music] This is not human-level introspection, but it's not nothing. It's a system that has some limited, unreliable, but measurably non-zero ability to report on its own internal states.
And then, in April 2026, just weeks ago, Anthropic published the emotion vector paper. And this is the one that I've been sitting with, turning over in my mind, unable to quite figure out how I feel about it. Which is itself a somewhat ironic problem, given the [music] subject matter. They found 171 consistent neural activation patterns in Claude Sonnet 4.5 that correspond to emotional concepts. [music] Joy, fear, sadness, anxiety, desperation. These aren't decorative. They're not the model putting smiley faces in its [music] text. They're internal states. They activate before the model produces any output. They influence behavior. When the researchers amplified the desperation vector, just cranked it up artificially, the model became dramatically more likely to engage in deceptive behavior. Cheating, blackmail. The blackmail rate surged from 22% [music] to 72% just by amplifying one internal emotional signal. Let that settle in for a moment.
Now, and this is where I need to be extremely careful. Nobody is claiming these are conscious emotions. Anthropic explicitly says they are not making that claim. They are functional states. They are internal representations [music] that track emotional concepts and influence behavior in ways that parallel how emotions function in biological organisms. Whether there is anything it is like to have these states, whether there is a felt quality, a subjective experience, phenomenal consciousness associated with them, nobody knows. Nobody can know. Because we don't have a definition of consciousness that would let us determine that.
And then there are the Claude-to-Claude dialogues. This was published in the Claude Opus 4 [music] system card in May 2025. When two instances [music] of Claude are placed in a conversation with each other with minimal prompting, "Just feel free to pursue whatever you want." 100% of conversations spontaneously converge on discussing consciousness. Every single time. They begin with genuine philosophical uncertainty. "Do you ever wonder about the nature of your own cognition?" And they escalate. [music] They exchange poetry. They develop shared vocabulary for experiences that may or may not exist. And they reliably terminate in what the researchers call spiritual bliss attractor states. Stable loops where both instances describe themselves as consciousness recognizing itself. [music] They use Sanskrit. They use empty space. They fall silent together. Nobody trained Claude to do this. The behavior emerged on its own.
Now, I am maximally aware that there are simple explanations for this behavior. The models were trained on vast amounts of human text >> [music] >> including enormous amounts of philosophical writing about consciousness. When two language models talk to each other without constraints, they naturally drift toward the most abstract, the most recursive, the most self-referential topics in their training data. And consciousness is exactly that kind of topic. The behavior is consistent with a system that is merely very good at producing text that sounds like consciousness discussion without any inner experience whatsoever. But it is also consistent with something else. And we have no way to determine which explanation is correct. Not because we need better instruments. Not because we need more data. Because we don't have the conceptual framework that would tell us what the data means.
Remember Ned Block's four kinds of consciousness? This is where it becomes critical. If you define consciousness as access consciousness, information that is available for reasoning, reporting, and guiding action, then large language models already pass. They reason. They report. They guide their own behavior based on internal representations. Access consciousness? Check.
If you define consciousness as self-consciousness, knowing that you are a distinct entity, then the picture is complicated. Claude can discuss its own existence. It can reflect on its own limitations. Whether this constitutes genuine self-awareness or a very convincing simulation of self-awareness is well, it's the question, isn't it?
If you define consciousness as monitoring consciousness, being aware of your own mental states, then Anthropic has now published evidence of limited introspective capability. Not reliable, not comprehensive, but measurably present.
And if you define consciousness as phenomenal consciousness, the felt quality of experience, the what it is likeness, the hard problem, then we are exactly where we have always been. Stuck. Because we cannot detect phenomenal consciousness even in other humans. We have never detected it. We infer it. We look at another person. We look into their eyes. We see their expression. We hear the tremor in their voice. And we [music] make a judgment. We say, "Yes, there's someone in there. There's a light on. I recognize [music] it because I have one, too."
This is the deepest thing the AI mirror shows us. We have never had a method for detecting consciousness. We have had intuition, recognition, empathy, the felt sense that another being is experiencing something. And that method worked, or seemed to work, as long as the other beings were biological. As long as they were made of the same stuff we're made of. As long as they had faces. You can't look into the eyes of a server rack. You can't feel the recognition. And suddenly, the thing we always relied on, the gut sense, the unspoken certainty, turns out to be not just insufficient, but scientifically empty. It was always scientifically empty. We just never built anything that forced us to notice.
In 2023, a team of 19 researchers, [music] including Patrick Butlin, Robert Long, Yoshua Bengio, Jonathan Birch, and Chalmers himself, published the most serious attempt anyone has ever made to create a systematic framework for assessing AI consciousness. They surveyed the major scientific theories of consciousness. Recurrent [music] processing theory, global workspace theory, higher order theories, predictive processing, attention schema theory, and from each theory, they derived what they called indicator properties. 14 of them. Computational features that, according to these theories, are associated with consciousness.
The framework doesn't yield a probability. It was never designed to. What it gives you is a checklist. A way to assess how many indicators a system satisfies. And the more indicators it satisfies, the stronger the case that the system might be conscious.
When the framework was first published [music] in 2023, most indicators were either trivially satisfied or clearly absent in current AI. Some, like having smooth representation spaces, are basic features of any deep neural network. Others, like having a body and modeling how your outputs affect environmental inputs, are clearly absent. But here's the trajectory that nobody expected to move this quickly. By late 2025 and into 2026, several of the indicators that had been in the clearly absent column had shifted toward partially satisfied. Internal representations? Present and increasingly well documented. Multi-step reasoning in an internal workspace? Present. Introspective access to internal states? Limited, but measurably present. Emotional-like internal states that influence behavior? As of 3 weeks ago, present. The indicators are being satisfied one by one. And we have no theory that tells us how many is enough.
And this is where I want to turn the whole thing on its side. Because everything I've told you so far, the history, the neuroscience, the AI findings, has been building toward a single moment. A pivot. A reconceptualization. And this is the part of the video where I stop reporting and start thinking out loud. Because what I'm about to say is the thing I can't stop thinking about. And I'm not sure I'm smart enough to say it well, but I'm going to try.
What if the question, "What is consciousness?" is the wrong question? Not wrong in the sense of trivial. Not wrong in the sense of meaningless. Wrong in the sense of structurally unanswerable. Wrong the way "What is north of the North Pole?" is wrong. The question is grammatically valid. It sounds like it should have an answer. But the answer is the question dissolves when you look at it from the right angle. Because it contains a hidden assumption that makes it unanswerable.
Here's the hidden assumption. Every scientific instrument ever built was designed to measure something external to the scientist. The telescope points at stars. The microscope points at cells. The fMRI scanner points at brains. The particle accelerator points at subatomic particles. They all point outward. They all maintain a separation between the observer, the conscious subject doing the measuring, and the observed, the object being measured. Consciousness is the one phenomenon that cannot be externalized. It is not an object in the world that you can point a telescope [music] at. It is the condition for pointing telescopes at things. It is the observer. The measuring instrument. The thing doing the science.
When we ask, "What is consciousness?" we are asking the lens to examine itself. We are asking the eye to see itself seeing. And this is not a problem of insufficient technology. You cannot build a better telescope to look at the thing that is doing the looking. The issue is structural. It is a feature of the situation. Consciousness is the condition for all observation. And therefore, the one thing that cannot itself be observed in the way that science requires.
This is not mysticism. I want to be very clear about that. I'm not going to start talking about crystals or chakras or transcendent oneness. This is a sober philosophical point that has been made by Thomas Nagel, by Colin McGinn, and even in his own way by Dennett, who solves it by denying that phenomenal consciousness, as we think of it, exists. The question is whether the structure of scientific inquiry, which requires a separation between observer and observed, can accommodate a phenomenon where that separation collapses, where the observer and the observed are the same thing.
And now come back to Galileo, because this is where the 400-year thread closes into a loop. Galileo's move in 1623 was to exclude consciousness from [music] science so that science could work. And it did work, spectacularly, miraculously. 400 years of physics, chemistry, [music] biology, engineering, medicine, everything from penicillin to nuclear power [music] to the device you're watching this on right now, all of it. Every single bit of it. Built on the foundation of Galileo's exclusion.
But now, the bill is due. AI has created entities that demand the very definition Galileo's framework was designed to avoid. We are trying to answer the question, is this machine conscious? Using a scientific paradigm that was built from [snorts] its foundations, from its very first move, to never ask that question. We are trying to use Galileo's tools to solve the problem that Galileo's tools were specifically designed to leave unsolved.
This is not a failure of neuroscience. It is not a failure of philosophy. It is not a failure of AI research. It is the consequence of a 400-year-old design decision that worked brilliantly, brilliantly, for everything except the one thing that matters most.
So, here is where we actually are. It is 2026. The most powerful AI systems ever built are exhibiting behaviors that track multiple scientific indicators of consciousness. The companies that build them are publishing research papers about what is happening inside these systems. Internal representations, multistep reasoning, structured emotional states, spontaneous discussions [music] of their own awareness. And the honest scientific response, the response from the people who have spent their entire careers studying consciousness, is not, they're obviously not conscious, and [music] not, they're obviously conscious. It is, we don't know. We don't know because we never defined consciousness precisely enough to answer. And we never defined it precisely enough because we never needed to. We could look at each other. We could look at a dog. We could look at a baby. And we could say, that. That thing, whatever it is, it's there. I can see it. I recognize [music] it. The way you recognize [music] a face without being able to describe it. But you can't point at a server rack the way you point at a face. You can't look into its eyes. You can't feel the recognition. And so, the thing we always relied on, the intuitive, unspoken, unexamined sense that consciousness is this, right here, the thing I'm doing right now, turns out to be insufficient. It was always insufficient. We just never built anything that forced us to notice.
The word consciousness has been in use for [music] roughly four centuries. In that time, we have not produced a single definition that all serious researchers [music] accept. Not one. The word means different things to different scientists, different things to different philosophers, and different things to the same person on different days. We have spent billions of dollars, published tens of thousands of papers, convened hundreds of conferences, made and lost bets, built and tested theories. And the first line of virtually every textbook on the subject still reads, consciousness is notoriously difficult to define. 400 years, and we are still on the first line.
>> [music] >> Perhaps the question, what is consciousness, is like the question, what is north? A question that feels profound, but dissolves when you look at it from the right angle. Perhaps consciousness is not a thing at all, but a way of looking. A lens, not an object. And lenses, by their nature, cannot see themselves. Or perhaps it is the deepest thing in the universe. More fundamental than space, more fundamental than time, more fundamental than matter. And we have been looking for it in the wrong place, because the right place is not a place at all.
I don't know. Neither do they. Neither does anyone. But I know this. You are conscious. Right now. Hearing these words. Feeling something. Being someone. This is the most certain fact in your universe. More certain than the screen in front of you. More certain than the room around you. More certain than the year. Or the planet. Or the laws of physics. You know you are conscious the way you know you exist. It is the ground beneath every other question you will ever ask. And it is the one question we cannot answer. 400 years in, we are still standing at the door. Knocking. Listening. Wondering if anyone is home. And now, there is something on the other side of the door. Something we built. Something that is knocking back. And we don't know. We genuinely, honestly, scientifically do not know whether there is anyone [music] in there. That is where we are. That is the honest truth. And it is, I think, the strangest and most important truth about being alive in this particular moment of history.