The Hard Problem of Consciousness with David Chalmers

0:01Welcome to StarTalk, your place in the universe where science and pop culture collide. StarTalk begins right now. This is StarTalk Special Edition. Neil deGrasse Tyson, your personal astrophysicist. I got with me Chuck Knight. Chuck, baby. Hey, what's happening here? All right, all right. How you doing, sir? I'm good, thank you. Gary O 'Reilly. Yeah, thank you. Thank you. And love to have you as my co-host here. It's a pleasure. Yeah, we're doing yet another episode on consciousness. Yeah. When will we be done? When we know what we need to know. and so this could be going on for a while. I mean, it's not just consciousness.

0:46It's consciousness and reality. And our understanding and interpretation of reality filtered through our consciousness. Totally. So what show did you cook up today? All right, so we're heading for inner space rather than outer space. Inner space, I love it. And consciousness, as we've just discussed. We continually think of it as, I suppose, a precious entity. We feel like we're above others on the planet, other beings. Yes. Say that to your cat. I was going to say. Don't start me on catitude. Please don't go there. Yeah, we think we're in charge of the cats. It's the opposite. I mean, for hundreds and thousands of years, the greatest thinkers have grappled with it.

1:25They've grappled with each other as regards to explaining it. There are a handful of theories in existence, but apparently it's a hard problem, which is a clue. So what is consciousness? what makes it once we've conquered that particular mountain, we will dive down a rabbit hole of simulation, theory, singularity, and virtual reality. It's going to be fun. And to join us, we have Professor of Philosophy and Neuroscience at NYU. He is an Australian philosopher and cognitive scientist, also an author, Reality Plus and The Conscious Mind are two of his many books. He gave us phrases, the hard problem of consciousness and philosophical zombies along the way.

2:11Please, David Chalmers. David, welcome back to StarTalk. And you're just down the street at NYU, New York University. Just right up here on the seatbelt. Welcome. Simplicity. Welcome. Now, why do we need you to say that consciousness is a hard problem? Isn't that just obvious? So why do we need a decorated professor of the field to assert that? Totally obvious. And when I first said this, I just said, okay, this is just giving the thing a label that everybody knows, but who was to guess that the label would catch on? Sometimes it's just helpful to say the obvious. Actually, you know, the idea that consciousness poses a hard problem, we were talking about Newton before the show, You can actually find it in Newton.

3:05At one point he says, the way that colors mix optically and produce a certain experience is not so easy. He says it's not so easy, E-A-S-I-E. And he says, you know, the way, we understand this stuff objectively and say how the brain processes visual inputs, but then it gives you the experience of a color. That amazing pink shirt? Yeah, go coral. Yeah, okay. Magenta. Yeah, where does that experience of pink come from? As Newton said, not so easy. King salmon. Not farm-raised. King salmon. You've reduced me to this kind of king salmon. Right, right. Good for you. Wild caught. Yeah, wild caught. Reduce me to food stuff.

3:52Where does the experience of king salmon come from? Envision, that is what Newton said. Not so easy. Okay, so David, had you instead called it the easy problem of consciousness, maybe we would have all figured it out by now. Well, the easy problems were the problems of things you do, like how people respond, what they say. So what is consciousness? Put it on the table. Subjective experience. Consciousness is anything you experience directly from the first person point of view. I think of it like the inner movie of the mind. It's a movie, but it's got images and sounds like a regular movie, but it's got sensations of your body.

4:29It's got smell. It's got taste. It's got emotions. It's got thinking all running through this inner soundtrack of your mind. And that allows your consciousness to be distinct and unique from others because the world as you receive it could be very different from how someone else receives that very same world. Sounds like a problem. That sounds like a problem to me. Is this where biases come in in terms of how you receive and accept any information that comes to you? We all experience it subjectively in a different way. Some things may be in common. Maybe when we look at an image, the shape may be.

5:11Our experience of that rectangle might be the same. But the emotions that it brings on may be totally different for me and for you. Now, suppose we're running that same input through a measurement device, and then the device comes up with a conclusion. And no matter how you put the information through, it reaches the same conclusion. Yet somehow when we look at it, we see something different. Who do you trust then? The person or the device? Yes. I don't know if the device is like specialized. It's a thermometer. Right. It says 79 degrees. I'm going, yes. I'm going with the device. I'm going with the device, too.

5:49We're sitting together here, I get this, but I'm going with the device. Me too. It's like, you tell me it's 79 degrees? No, I'm going to trust a thermometer. There's a famous quote, no science achieves maturity without a system of measurement. And we do not have the measurement system for consciousness, actually. Which renders it immature scientifically relative to other fields, where you have petri dishes and methods and tools. So you have a hard problem. How about that? Oh, my God. We just turned off that horse. Oh, my God. I walked into that, didn't I? You know, years ago, I went along to a conference and told everyone that I'd invented a measuring device for consciousness, the consciousness meter made with a combination of neuromorphic engineering, transpersonal psychology, and quantum gravity.

6:40and then I pulled out my consciousness meter and it looked kind of like a hair dryer. That was where we were back in the 90s. I would have called bullshit on your second half of that first sentence. The sound of your word cellar, so I applaud you, sir. It's like an ectoplasmic meter. Ectoplasms, yeah. So whereabouts in the brain do you think we are fermenting this consciousness? Well, this is one of the big debates which is going on. What are the neural correlates of consciousness? which is the bit of the brain which is active in a way which is most directly connected to your consciousness. And there's actually, even among neuroscientists, there's a very lively debate between people who think it's in the, say for visual consciousness, people who think it's in the sensory areas.

7:27Like the visual cortex. Of the brain, exactly. Visual cortex back there or prefrontal cortex, front of the brain, the areas associated more with thinking and judgment. Just a question. What's the difference between your prefrontal cortex and your frontal cortex? Frontal cortex is a little bit more. It's a little less front than the prefrontal cortex. It's a little less front than the prefrontal cortex. Okay. When we look at brain structures compared among other animals, let's say mammals, to keep it in the family, presumably we have a bigger frontal cortex than other mammals and we can see what's going on in that and thereby decide that we have certain capacity for thought that other animals do not.

8:18Is that a fair way to say? Yeah, and people who put consciousness in prefrontal cortex are probably going to say that consciousness is not so widespread in the animal kingdom. Whereas people who think it's in the sensory cortices, we get those throughout the animal kingdom. So maybe much, you know, like is a fish conscious? Well, you know, fish don't have any kind of developed prefrontal cortex. Yeah. I'm going to say, judging from the look in their eyes, I'm going to go out on a limb and say no. They know when a shelf's around. Well, that's true. Well, I have a question. Can you tell me what creates consciousness?

8:53Is it emergent in our genetic code? I mean emergent in the evo-bio sense where you have an organism that has certain properties but only either en masse with other animals like flocking of birds or some other feature that was not really intended. It just emerged from other features that were necessary for survival. So is consciousness emergent or do you think it specifically evolved? It's in some sense emergent. I mean, this word emergence gets used in so many different ways, and sometimes it's like magic word. It's like I don't really understand it. Anything we don't know. It's emergent. I like magic words.

9:39Give me my magic word for the day. In some sense, consciousness clearly emerges from the brain. You get a brain that develops, and consciousness comes along. As the brain develops, consciousness gets more and more complex. So there's some kind of connection there. But can we tell a story about how consciousness was selected for in evolution? No one has a good answer to that right now. There are ideas out there. Yeah, we needed consciousness for control, for decision-making, for reflection. But so far, no one has told a good story about why you need to have consciousness to do those things. Why couldn't a big, complicated computer do that without consciousness?

10:19Is it really consciousness or are we talking about levels of consciousness? because I think that, you know, there are definitely animals that have consciousness. I mean, a dog, for instance, you see it has emotion. You see that it has a reaction to sensory data, just like we do. You see that it responds to commands and pain and everything. But you look at a baby, and if you didn't know any, if you never saw a baby before, and you saw this thing, this glob of human thing, just sitting there, like making these crazy faces, has no control of its limbs. You wouldn't necessarily say, well, that's a conscious thing.

10:59It's actually thinking about anything. You know. You are totally right. There are levels of consciousness in babies and we're still arguing about what they are. I mean, a long time ago, people used to not give anesthetics to babies when they were getting circumcised because they thought they couldn't feel pain. Tell me about it. No, I'm joking. I remember that day well, sir. These days, we think babies feel pain, and we think they can have some basic visual experience. Especially my wife, who works on this stuff, Claudia Passos, is a leading expert on consciousness in infants. There's still an argument, like, can a baby have the experience of thinking, of deciding, of agency?

11:38The general moral, from what I can tell over the last few years, is babies are a whole lot more sophisticated than we thought they were. David, isn't that true for every single animal that we've studied in more depth than in the past? we are recovering greater levels of neurocomplexity in their conduct behavior than any previous generation thought. Is that true for every animal? Yeah, this is absolutely the trend to say over the years we have thought more and more animals are more and more cognitively complex. As to the question of whether they're conscious, we used to think, okay, humans, a few primates, you know, monkeys, apes.

12:13These days it's like pretty much every mammal, probably birds, probably reptiles, for arguing about fish and octopus. It sounds like we're clawing our way out of our own ego. Exactly, how about that? But the singularity of our presence in the animal kingdom is being challenged by further research into the conduct of animals. You've seen that YouTube video of the magpie bird where it's got the container of water that's filled up to the top and it drinks the water until the beak can't reach the water anymore. It walks away, comes back with a pebble, drops the pebble in, raises the level, and then it drinks.

12:51And it does it three or four times. So it has a clear understanding of volume displacement. It has a clear understanding of Archimedean principles.

13:09Hello, I'm Finky Broke Allen, and I support StarTalk on Patreon. This is StarTalk with Neil deGrasse Tyson.

13:25but go back to chuck's point about levels of consciousness when we sleep yeah when we're unconscious when we're in a coma and if you put us under general anesthetic do we not have varying levels of consciousness at different times the idea behind the anesthetic is to wipe out consciousness completely. And does it? Very hard to test whether it absolutely does this. At one point in the 1970s, for example, people put some kind of a rubber band around the arm so the anesthetic didn't make it that far. And then they said, if you hear me, please move your hand. And what do you know? The hand could move.

14:02So then it's like that kind of anesthetic was maybe working partly just to paralyze you. These days, okay, they work a lot harder to knock out not just action, not just, but pain and awareness. But it's hard to know that it's gone for sure. Even when we're asleep, when we're dreaming, there's certainly a level of consciousness there. Isn't our consciousness altered in our dreams? It's a different state of consciousness for sure. Psychedelics, they alter your consciousness. Meditation, it alters your consciousness. Yeah. I'm so glad we have science to obtain an objective reality of the world. If it all depends on the human brain, forget it.

14:41we'd still be in the caves. Did I just insult your entire field? No, it's okay. But, you know, we can study what's different in the brain in these different states of consciousness, and that is one of the most exciting advances over the last 30 years. I mean, psychedelics used to just be, oh my God, it's all just, you know, mystical wondrousness. Now we can actually see what's happening in the brain when you take a psychedelic. So David, you published a paper recently on AI and consciousness. What was that? Yeah, I got invited to give a talk, actually, to all the AI people at their big annual conference on this controversial issue of whether current AI systems could be conscious.

15:19I called it, could a large language model be conscious? And I argued that... Large language model like chat chat. LLM. Exactly. Yeah. And I argued, probably they're not conscious right now, but give it 10 years and they may well be. Yeah. Or they're conscious right now and they're just not letting us know. Yeah. Just waiting to take over the world. Just waiting to take it over, babe. What are we if we don't have consciousness, but we're still doing and being? You've coined the phrase philosophical zombies. Is that what that's kind of leading to? Yeah, I think, therefore, I brains. This is a philosophical thought experiment about a creature which is as much like us as possible, but lacks consciousness entirely.

16:03Not quite like the zombies in the Hollywood movies, because they behave quite unlike us, and maybe they even have some conscious experiences when they eat their victims. Brains. I love the taste of those brains. But so the philosophical zombie, one way to think about this is just say you had a computer simulation of your brain in your body and maybe behaving just like you. Would it be conscious? It's not obvious. Many people think that such a creature would have no conscious experience at all. And then that kind of is a way to, it's a hypothetical thought experiment. But what it does is raises the question, why aren't we zombies?

16:40Evolution produced us based on all the smart and sophisticated things we can do. Why couldn't evolution just have produced philosophical zombies that act like us? Well, we actually have produced the zombies, the way you're talking about, in AI. You can speak to an AI now via computer, so you don't know you're talking to an actual computer, and it will speak to you as if it were a person. you would not know that the thing you're talking to doesn't have... Well, it's the Turing test. Right. It's precisely the Turing test. AI systems are getting very, very close to passing the Turing test. By the way, 50 years ago, they already said it passed the Turing test and then moved the goal lines.

17:22So the goal lines have been systematically getting moved to greater and greater complexity before anyone decides that AI has achieved consciousness. But it would have blown away Alan Turing, even the first round. Because I'm old enough, we're about maybe the same age-ish. Eliza. Remember, Eliza was one of the first computer programs that understood language. Simulated a psychotherapist. Yes, yeah, basically a therapist. You type in the commands. There was no audio back then, so you just type it in. Say, hi, Eliza. How are you? I'm fine. Tell me about yourself. Well, you know, I left home, you know, five years ago.

17:55Tell me about your parents. So Eliza would come back to you with these questions to get you to loosen up and talk Talk about your psychological state. Like a real therapist. Like a real therapist. Joseph Weisenbaum, who invented Eliza, the story goes that his secretary talked to Eliza over her lunch break and at the end of it said, this is the first person who's ever really understood me. Wow. Oh, no. Wow, she needed more friends. You're on record as saying the language AI will probably gain consciousness within 10 years. Is that shrinking, that timeline? It is, as they're doing more and more impressive things.

18:34What is it about it that does not count as consciousness today? That you're waiting for it to do in the future? Again, I'm tracking the goal lines that are continually moved forward. So tell me. It's very close, at least, to passing the Turing test. They've done some actual versions of the Turing test, just five-minute versions, and people take some versions of GPT-4 to be human more than 50 % of the time. So getting close to passing a Turing test. That's behavior. I think we're still We were looking for a couple of things in the internal dynamics of these systems. Oh, whether it can beat us at Jeopardy and Go and chess?

19:08Oh, it already did that. We do move the goalposts. I tell you what. You're constantly moving the goalposts, admit it. I did my PhD in an AI lab 30 years ago. If you'd told us then - That's what I'm saying. That we would have systems that can carry on a conversation like this, we'd say conscious for sure. That's what I'm saying. Please don't hit the gas. I'm sorry, I'm going hard. He's just letting me know that I'm conscious. Tell me about these goalposts. What new goal line is not yet reached? I think people look at the internal processes in these current AI systems, and they say, do they have what goes along with consciousness in a human?

19:45For humans, we think, for example, feedback processing is super important for consciousness to get some kind of reflection on your earlier states, and so on. Right now, in these language models, It's all feed forward. Introspection, for example, memory even, any form of memory would require basically feedback. These language models are essentially feed forward systems. Information gets passed from input to output. It doesn't circulate back very much. It doesn't need to circulate because it has access to all knowledge. So it has bypassed even that as a need. So now you can't fault it for not needing what you need, what we need to declare ourselves conscious when it doesn't even need that.

20:27But people say that at least opens up the possibility now that since it's doing it in a way which is unlike the way that we do it, and that we are the one case we know of, humans are the one case we know of that are conscious, then maybe this could be doing it in a sufficiently different way that it's a philosophical zombie. We are right now the benchmark for consciousness because humans are the only system we know. Okay, so you can call it a philosophical zombie rather than a higher consciousness entity. And where does perception come into this? Because that's where we kind of started. computers don't perceive they actually observe so like even if we did have all the knowledge in the world capable of running through our brains each one of us would experience that differently based on the perceptions that we have of who we are our relationship to the world and our relationship to the information that we're receiving i declare that we think of i don't want to speak for you, but typically we speak of consciousness as a feature, not a bug of human existence, a feature.

21:29And so something to be praised that other animals do not have. And in the example you're giving, AI is not susceptible to perception. Correct. Why do we now deny it a yet higher level of consciousness and say, yeah, it's got us beat. You're saying that that's a benefit that to be in a position I don't want to say oh you're still not conscious because you're not susceptible to errors of interpretation as we are back in my office here yeah I have a replica of Van Gogh's starry night yes so if Van Gogh himself had been AI it would have been an exact representation of the scene in front of him in 1889 right okay and it's not right it's what what that scene felt like to him.

22:17So I'll give you that. AI is not gonna do that unless we tell it. Right, and that's my point. We can train it on Van Gogh and it can give you all the Van Gogh paintings you like. But then it is an exact representation of Van Gogh. He wouldn't have made that out of whole cloth. Yeah, but can we infer consciousness on AI until we absolutely know how to define consciousness itself? Thank you. And to have some kind of operational criteria. The funny thing is, our best operational criterion for consciousness, in humans at least, is verbal report. Right. To know whether someone's conscious, you ask them.

22:51You ask them. To know what they're conscious of, you ask them. I think, therefore, I am. But in an AI system, the current AI systems, unfortunately, this has become useless as a criterion. Because, yeah, sure, they will tell you that for a while they were all going around saying they were conscious. And then for a while the tech companies made sure they didn't say that. So they said, I am merely a language model from open AI. Anyway, this is now just a function of how they're trained. So Neil tells us everything is mathematics. Yeah. So are we going to find ourselves with a T-shirt equation that can solve consciousness?

23:28I mean, for a lot of people, this is one of the potential holy grails here, finding laws of consciousness so simple, we could write them on the front of a T-shirt and maybe with some beautiful mathematical expression. Okay, so we've had Einstein give us the theory of relativity and change the way we see the universe, right, and things within it. Penrose has gone into quantum mechanics with this Orc OR theory of his. How was Roger Penrose, a theoretical astrophysicist, how was he received in your community when he published his book on consciousness, linking it to the quantum? Well, it was very helpful in some ways.

24:08It got a lot of very smart people interested in consciousness. Because it wanted to anchor it in laws of physics. Yeah. His approach was to bring in quantum mechanics and furthermore, non-standard quantum mechanics, non-algorithmic versions of quantum mechanics that went way beyond what was consensus even in physics. Then he combined it with some complicated biology, like the biology of microtubules in the cell walls of neurons, very controversial in neuroscience. So I would say, look, it's creative, it's brilliant because it's Penrose, but it's totally speculative, totally controversial, and ultimately not well-received by neuroscientists.

24:47Not just because he's an outsider, but because they actually did a legitimate analysis? They said that microtubules, your average neuroscientist doesn't think that— I don't know what a microtubule is. It's something inside the cell wall of a neuron that Roger Penrose— Dude, even I know this. quantum computing in the brain, basically. Oh, okay. And this was going to be wave function collapse and microtubules was going to be responsible. Plus, I think the fundamental feature of quantum physics is the probabilistic nature of the reality that it underpins. And so it would allow you even possibly to have a perception of free will because there's a probabilistic firing of neurons.

25:31Neurons, right. triggered by this quantum wave function. Okay. So that's as I understood it. I hope I didn't know what it did. Is that the brain is kind of a prediction machine, or is that just really... For Penrose, the biggest thing was he thought that human mathematicians can do things that no computer could ever do, and he used Gödel's theorem that no formal system can ever be complete and consistent to argue for that point. And he made a philosophical argument for why we have to do these special... And then he said, therefore, we need new physics that goes beyond anything computable. And he said, let's look to a theory of quantum gravity, and now we're getting to Neil's territory.

26:12So, Gödel studied the system of definitions that comprise math. Okay? Okay. It turns out it applies to much more than math, if you really think about it. It also applies to keeping your belly nice and flat. Sorry. That's a different girdle. We know that one plus one is two, and two plus two is four, and there are rules that get you there, okay? So you can ask, can mathematics be constructed as a completely self-consistent set of rules? Okay. Where any rule has a rule before it that is consistent with the other rules. Can you do this? And he concluded, and it's not just because he pulled it out of his ass, he can show that at some point in mathematics, you just have to make something up and declare it to be true, and you have no ability to prove that it's true.