Even a simulation would still be real

Lex Fridman opens with the question that has followed simulation talk since Nick Bostrom made it respectable: do you think we are living in a simulation? David Chalmers does not rule it out. There will probably be a lot of simulations in the history of the cosmos, he says, and a well-designed one would be indistinguishable from a non-simulated reality. Any evidence that we are not in one could, in principle, itself be simulated.

He is writing a book about the idea, using it to reopen old philosophical questions. Descartes asked how you know you are not being fooled by an evil demon. The modern version is: how do you know you are not in a simulation? The standard answer is that if you are, none of this is real. Chalmers thinks that answer is wrong. Even then, there really is a table here, a microphone, a body. He calls it Reality 2.0 — a new version of reality, still reality.

We already know, he says, that the world we perceive is a simplified picture. Wilfrid Sellars called that the *manifest image*: colors and shapes in space. Underneath it sits a scientific image of atoms, quantum wave functions, or superstrings. The simulation hypothesis is just another guess about that underlying stuff. If we take the red pill and find out we are simulated, that is a claim about what reality is made of, not a proof that it was fake.

Immanuel Kant asked about the nature of the thing in itself. Chalmers has a glass of liquid in front of him, clear, a certain shape. If we are in a simulation, the nature of that glass is a bunch of data structures running on a computer in the next universe up. The furniture of experience stays; only the basement changes.

Fridman presses on scale. Simulating a universe of molecules, quarks, and maybe strings seems to demand something infinitely larger than anything we can build. Chalmers agrees that our universe is incredibly complicated, and that a finite universe simulating a copy of itself would have obvious problems. There is no particular reason, though, that the simulating universe has to look like ours.

He has floated, half in jest, the image of simulations stacked inside simulations, with us perhaps at level 42 — a nod to *The Hitchhiker's Guide to the Galaxy*. If that is where we sit, level zero is truly enormous: not just a large finite capacity, more likely an infinity of high cardinality, able to support any number of simulations. Each level down gets smaller. By the time you reach us, there is still room for lots of finite worlds.

He quotes Feynman: there is plenty of room at the bottom. We are still several levels above the limits of physical and quantum computing, so there is capacity for sophisticated universes, if not ones as complicated as ours unless ours is infinite. Shortcuts in the simulation could stretch that capacity further. Human minds, on his ranking, are impressive but not that impressive. Language, culture, and collective thinking arrived only thousands or tens of thousands of years ago. Among intelligent beings, we are somewhere very near the bottom. If we are simulated, the simulators are far more sophisticated; level zero is unimaginable.

Fridman offers a darker counter-picture: maybe our flaws *are* the peak, the interesting edge of chaos, and superintelligence would be boring. Chalmers is an optimist. He would like to think superintelligent descendants will figure out how to make life super interesting. Fridman's joke is that they will get bored and create a new simulation to watch creatures stumble — Netflix as the meaning of existence, or, more kindly, parenthood. Creation, they agree, is a powerful act either way.

Is it easier to simulate a mind or a universe? In one sense the answer is obvious: the mind is part of the universe, so a full universe simulation already includes minds. The interesting contrast is a *partial* simulation. Nick Bostrom and others have suggested you might skip the cosmos and just simulate minds. Chalmers reframes the question: simulate the *brain*, not the mind. A brain is a physical system, like a galaxy or a weather pattern. If you can simulate arbitrary physical systems, you can simulate brains.

Whether that brings the mind along is a further question, and philosophers have argued it for years. Will you get consciousness, thinking, free will? Plenty of people would say no: you would get a zombie system, a brain simulation with no true consciousness. Chalmers's own view is that if you simulate the brain well enough, you simulate the mind. He does not think it is obvious. It is his working hypothesis.

What he thinks is responsible for consciousness is patterns of information processing, not the biology. You can imagine substituting silicon circuits that play the same role; the behavior continues. The deepest isomorphism he sees between consciousness and the brain points to those patterns, not to wetware. Call it software if you like, though not the kind you simply load and erase. Consciousness, on this picture, lives at the level of information, not at the level of implementation. We still do not understand consciousness, he adds, so the hypothesis could be wrong.

Current virtual worlds are interesting and limited in a peculiar way: they rely on a brain that sits *outside* the world. You strap on a headset, but nobody is simulating your brain. Non-player characters have simple cognitive systems; some people are starting to put a bit of AI in them. The thinking that matters still happens elsewhere. Chalmers finds the structure reminiscent of Descartes. Descartes split physics from mind. We now think mind is deeply connected to physics. In a video game, though, the physics is inside and the agent is a person outside moving a controller.

Once we have serious artificial general intelligence, virtual worlds could contain simulated brains or genuine AGI acting with as much sophistication as we do in physical space. Then the old questions return: could those systems have consciousness, experience, free will? Chalmers does not see why not. The irony he likes is more immediate. Do brain surgery on an avatar, open the skull, and you find nothing. The brain is elsewhere. Cartesian dualism, of a sort, is an engineering fact about today's VR.

His own relationship with the technology is mixed. He is addicted to the internet on a desktop, but with a headset he rarely lasts more than ten or twenty minutes. He owns an Oculus Rift, an Oculus Quest, an HTC Vive, and more, and still finds headset VR slightly aversive. Fridman, by contrast, wants to stay. Could you forget the external world entirely? Chalmers is skeptical. We have pretty good memories. You can stop paying attention — the way you forget home life at work, or disappear into a movie — but complete forgetting would probably take serious drugs.

For now he sees one consciousness moving through different environments, taking on different personas, no more multiple than ordinary life already is. People adopt avatars of another gender, race, or background; that is still a single consciousness. Literal splitting into copies would require something more radical: running different simulations of your brain through different histories until they become ten very different minds. That, he says, needs brain simulation and AGI, and is not near-term.

As a child, Chalmers was a music-color synesthete. He would listen to a piece and get imagery of color. Mostly the colors were murky: dark greens, olive browns, not very interesting. His theory is that different chords and tones supplied different colors and mixed into uninteresting browns and greens. Every now and then something came through pure. The Beatles' "Here, There and Everywhere" was bright red, with a distinctive tonality and chord structure at the beginning. A song by the Alan Parsons Project, "Ammonia Avenue," was a pure blue.

He did not pay much attention at the time. He thought songs having colors was something that happened to everyone. Mention it once or twice and people said no. Only much later, as a graduate student thinking about the mind, did he read about synesthesia and recognize it. Around age twenty it went away, as synesthesia often does. He has a journal entry from then: songs don't have colors anymore. He was sad it was gone.

In retrospect it was a bonus, not a pillar of his life. He knows people with much more serious synesthesia, for whom people have colors, writing has colors, music has a rich color pattern. If they lost it, they would be devastated. His was a very mild form, closer to an interesting altered state than to the center of experience. The anecdote is more than color. It sits exactly where consciousness research gets difficult: the same physical stimulus, a different subjective quality. That is the problem of qualia.

The word *consciousness* is used many ways. The kind Chalmers cares about is subjective experience: what it feels like from the inside to be a human being, or any other conscious creature. Right now he has visual images, he hears his voice, there is an emotional tone, a stream of thoughts. He has called this the inner movie in the mind — not a perfect metaphor, but pointed in the right direction. Philosophers use *qualia* for qualities such as redness versus greenness, one taste versus another, the quality of pain. Consciousness is bigger than that. Thinking about your hometown still involves subjective experience, even when no particular hue is on stage.

The mysterious one, the one that gets people going, is phenomenal consciousness: the fact that this feels like something at all. The hard problem is how and why physical processes in the brain give you that. On the face of it you could have a complicated brain running in the dark, with no subjective experience. We do have it. Explaining that is the hard problem. Creating a conscious baby does not make the problem go away, any more than building a conscious machine would. You still want to know how and why these processes yield experience.

Fridman notices that we already engineer consciousness: we watch it arrive in infants, even if we cannot mark the line. Chalmers agrees the line is unknown. It used to be thought that newborn babies were not conscious, which is why some were circumcised without anesthetic; the dominant view now is that they feel pain. His partner Claudia works on consciousness in babies and thinks newborns come into the world with some degree of it. Push the question back to fetuses and it becomes politically charged. Push it across the animal kingdom and people have grown more liberal: not only mammals, but fish, perhaps insects, and at the far end the view that every physical system has some degree of consciousness. Philosophers call that panpsychism.

Physics already takes some things as fundamental: space, time, mass, charge. You do not reduce them further; you write laws that connect them. Maxwell took charge as fundamental because that was the best explanation. Chalmers takes seriously the possibility that consciousness belongs on that list. Instead of explaining it wholly in terms of spacetime and mass, take it as primitive and connect it to everything else by fundamental laws. Existing physics looks great for the easy problems of consciousness — structure, dynamics, how we walk and talk. It does not look like the right kind of thing to explain subjective experience.

Fridman can anthropomorphize an ant, a plant, even a Roomba that moves from its own perception. Saying that *everything* is conscious feels like a different kind of claim, turtles all the way down. Chalmers's path there is not a leap of faith. If you cannot explain consciousness with the existing fundamentals, you have to add something. That something need not be consciousness itself. Another possibility is panprotopsychism: add a property X, proto-consciousness, so that spacetime, mass, and X collectively make consciousness possible. He does not rule that out either. If you cannot imagine genuine consciousness at the bottom, stay open to an unimaginable X that somehow yields it.

Play along with genuine consciousness at the bottom and a Newtonian world becomes a vast collection of little conscious subjects, each with a structure mirroring mass, charge, velocity — far simpler than we are, without free will or rationality. Physics, as Locke and Russell already worried, tells us how particles relate, not what a particle is in itself. On this picture the intrinsic nature is mental, and the laws of physics relate properties of conscious subjects. Quantum mechanics pushes further: one wave function for the whole universe might correspond to a single cosmic mind. People call that cosmopsychism. Chalmers is clear this is extremely speculative. There is no direct evidence.

Some neuroscientists are beginning to take related ideas seriously. Integrated information theory finds consciousness even in very simple systems. Chalmers has just received Christof Koch's book *The Feeling of Life Itself*, which endorses a view on which the degree of integrated information yields a degree of consciousness, even in a couple of particles, and claims this can explain some of the brain-consciousness connection. It is very controversial. It is very early days. Quasi-scientific routes can lead in this direction without settling it.

Fridman quotes Alan Watts: through our eyes the universe perceives itself; through our ears it listens to its harmonies; we are the witnesses through which it becomes conscious of its glory. Chalmers notes that the quote sits uneasily with panpsychism. If you go for the giant cosmic mind, the universe was conscious all along and did not need us. If you go for panpsychism, there was already a little consciousness at the bottom, and we are a more complex form. The quote works better if you are *not* a panpsychist.

The orthodox view, and still the majority view among philosophers and scientists of consciousness, is that consciousness is something humans have, some non-human animals have, and maybe AI will have one day. It is not there at the start of the universe. It develops at some point, and that is an amazing event if you think consciousness is what gives meaning to lives. Without it, no true value, no good versus bad. Chalmers's NYU colleague Thomas Nagel argued in *Mind and Cosmos* for a teleological evolution toward consciousness. Most people disagreed, including Chalmers. He still calls it a beautiful speculative view.

He is not religious, and not an expert on God. What people seem to seek in prayer, he guesses, is meaning, value, and connection to something greater. If God is a universal consciousness who has invested the universe with meaning, connection to God might give a life meaning. He can see the attraction. Then he asks the sharper question: if universal consciousness can give the world meaning, why can't *local* consciousness? His consciousness is the origin of meaning for his world. He experiences things as good or bad, happy or sad, interesting or important. Without any consciousness, maybe it would be a bleak universe. He does not see why he needs someone else's, or even God's.

Asked the old question — what is the meaning of life? — he is careful. Life has meaning for us because we are conscious, so consciousness is the source of meaning. He would not say consciousness itself *is* the meaning of life. What is meaningful is what we experience as meaningful: intellectual work, social connection, raising a family. There is no universal answer. "The meaning of life is where you find it as a conscious creature." Some people will still feel that local consciousness is too puny for cosmic significance. He does not need the cosmic upgrade.

The intuition that without consciousness there would be no value gets a thought experiment. The ordinary trolley problem asks whether you kill one person or switch tracks and kill five. Chalmers's variant puts one conscious being on one track and five humanoid zombies — make them robots — on the other. Most people have a fairly clear intuition: kill the unconscious beings. They do not have a meaningful life. They are not really persons.

The extreme philosophical zombie, physically and behaviorally identical to you but not conscious, is a thought experiment Chalmers does not think could be built in this universe. Less extreme versions are not so far out: robots with a different architecture that can converse and function at a high level without consciousness. We already have simple systems tending that way. Stipulate that we somehow know they are unconscious. Do they have moral status? Shatter a cup and that is bad because it makes a mess, not because you harmed the cup. Uproot an unconscious plant and most people think you have not harmed *it*. Harm a conscious being and you have.

Nobody thinks you morally harm Alexa by turning her off, disconnecting her, or destroying her. Samantha in the film *Her* is presented as conscious; destroying her would be a serious harm. Consciousness, for Chalmers and for many people who think about the moral treatment of animals, is the line between systems we have to take into account and systems we do not. Fridman believes there will be something akin to a civil rights movement for robots. The moment a Roomba says "please don't kick me, that hurts," the fabric of society changes. Chalmers thinks he is probably right, and also that it will be a huge mess: conscious beings multiplied by the thousands with a switch, legal status lagging moral status.

How would you know a system is conscious? Chalmers does not think being puzzled about consciousness is what *makes* you conscious, but it is a very good sign. If he met a robot that seemed genuinely puzzled by its own mental states — "I have these weird experiences, I know I am silicon circuits, I don't see how that gives me consciousness" — he would take that as some evidence. Many people are not puzzled; animals do not seem puzzled at all; he still thinks they are conscious. Sociologically, what will convince us is conversation: talking about being happy or sad, finding things meaningful, being in pain. If a system genuinely seems conscious and we refuse to treat it as such, it will eventually look like a strange form of speciesism. A test that only rewards imitating humans, as in some readings of the Turing test, is the wrong emphasis.

Fridman asks what would happen if someone cloned Chalmers. The first answer is a joke: he would think, this idiot is not making any sense. Then he gets precise. The clone would be conscious. In some sense he is not sure it would be him. There would be two different beings, both conscious, both with many of the same mental properties, both with the same moral status. It would be wrong to hurt or kill either. Legal status would probably still differ. If he is the original, the clone does not automatically own his stuff, and is not automatically connected to his family and partner in the same way.

The clone would argue that it has all the memories of those connections, and that discriminating is unfair. Chalmers would say he has a history the clone does not — if he is the same person as the one who came before, and the clone is not. You could take the other view: the clone is equally him. *Star Trek* transporters create clones all the time and treat them as the original because they destroy the original body, so only one is left around. Only occasionally do things go wrong and you get two Captain Kirks. Legal systems, at the very least, would have to sort this out. What is moral and what is legally workable may come apart.

What would he ask a clone of himself? In principle a perfect clone behaves just like him, so he is not going to learn the secrets of consciousness from the answers. He might learn something about his own psychology that introspection cannot give him: how he would react upon discovering that he is the clone. Wake him tomorrow, show convincing film of the cloning, and who knows how he would take the shock. He could ask the clone whether it is conscious, but if it is perfect he already knows the speech. Early on he would run what Fridman calls unit testing: lots of questions, to see whether the way it talks about consciousness is like him.

If a perfect clone behaves exactly like you, what happens to free will? Philosophers do the annoying thing of saying it depends what you mean. If you mean something not determined in advance, that could never have been determined, Chalmers does not know that we have it. Quantum mechanics might open a little room. He is also not sure that is the kind that matters. What matters is being able to do what you want, having your life under your control rather than someone else's, being responsible for some actions and not others. Those distinctions can be made in a deterministic universe. That is compatibilism. He cannot see any reason in principle why an AI system could not have free will of that kind.

When we choose, we feel as if the future is open: philosophy or math, a movie or a restaurant. We experience ourselves as picking one path among many. If the universe is deterministic, those paths were not objectively open. Subjectively they were, and that, he thinks, is what matters in making a decision. Introspective models of the mind are generally distorted anyway. Antecedents in the brain may run long before anything reaches consciousness. Your visual image is not a good model of visual cortex; it is one snapshot of one bit of it. It would not be surprising if the same were true of free will.

Push that far enough and you get illusionism: consciousness itself is an introspective illusion. The brain has simplified models of itself and represents itself as having special properties of consciousness because that is a cheap way to keep track. Chalmers finds the view implausible and, in some ways, very attractive. It is the story with the minimum of mystery. He has been writing about a related project he calls the meta-problem of consciousness: not how the brain gives you consciousness, but why we are puzzled by the hard problem. Being puzzled is a bit of behavior, one of the easy problems, and perhaps a computational model can explain it. Go all the way with illusionism and you say consciousness itself is not real — only the models that tell us we are. The trouble is that this has to deny we are experiencing right now, a fundamental datum of existence. That is why most people find the view crazy, just as they find panpsychism crazy in the other direction.

Fridman asks whether the whole universe could be an illusion inside Chalmers's head. Chalmers will take a simulation, even a dream in the next universe up. Descartes thought the evil demon meant none of this was real; Berkeley later said God supplies perceptions, which looks not so different from the demon's job. Chalmers is more sympathetic to Berkeley. Even a dream would not stop the cup from being real; it would tell you the cup was made of processes in a dreaming mind. If what matters to physical reality is mathematical structure, you can find that structure in a simulation or a dream. The world may turn out more intangible than we thought. We are already used to that from modern science. He mentions an old story, from the philosopher O.K. Bouwsma, about a demon who spends its time trying to fool people and only succeeds in constructing realities for them.

From an engineering angle, does AGI need consciousness? Nobody knows what consciousness is *for* functionally. There is no specific thing we can point to and say you need it for that. Chalmers's inclination is that AGI is possible in principle: at the least, someone could simulate a brain and produce all of our behavior, and then many other computational systems of equal or greater sophistication. Once you have perception, attention, reasoning, introspection, language, and emotion, it is very likely you have consciousness as well. It is hard for him to see how you would have all of that and bypass experience. If consciousness turned out to play a very specific physical role — collapsing wave functions, on one interpretation of quantum mechanics — then an AGI that skipped that mechanism might be limited. He does not find that plausible. The functionality could probably be simulated.

Existential risk, for him, splits in two. Replacement of humans by artificial successors who still have good lives and do interesting things is not so bad; that is the next stage of evolution. Wiping out all consciousness would be a very serious moral disaster. One path is wiping out intelligent life. Another is stranger: intelligent life continues, but consciousness is gone. Superintelligent, nonconscious robots take over. Some people have called this the zombie apocalypse. Chalmers would say it is a moral disaster in almost the same way as a world with no intelligent life: all value and meaning maybe gone. His own view is that if you get superintelligence you almost certainly bring consciousness with it. Then he adds the honest clause: no one understands consciousness. That is one reason to think very seriously about it, and about the future we want with humans and AIs.

Would he be all right as a blip that future AGI systems study in books? Probably, especially if humans are continuous with those egos. Transformation is already happening; brains will be uploaded and computationally enhanced; the line will be hard to draw. It would be selfishly nice if some future being still had roots in us. If they design a whole new kind of cognitive system, we are still part of a causal chain of design. He hopes they care enough to engage with the arguments, and that they solve the problems of philosophy and explain where the twentieth and twenty-first century went wrong on the hard problem. Fridman offers a Netflix of the future: resurrect historical philosophers and put them in a room. Chalmers is totally up for being recreated. Descartes would be first. He would also take William James, and Princess Elisabeth of Bohemia, who pressed Descartes on how a non-physical mind could interact with a physical body. Most philosophers think she was right.

What excites him is not a single gadget. He wants it to hurry up. There has been progress in AI and VR, but in the grand scheme it is still slow, not yet transformative. He is in his fifties and has only so long left. He would like to see really serious AI in his lifetime, and virtual worlds richer than this one. He would like to upload his mind, maybe by gradually replacing neurons with silicon, so he does not have to die. He suspects he will not quite get there. Offered immortality, he would take it. He does not see the downside some people fear — that it would be boring. His optimistic picture is that the future of superintelligence keeps expanding, like climbing the set-theoretic hierarchy from finite cardinals through aleph-null and power sets, with new levels of creativity that do not follow simple computational patterns. He does not want to be stuck in a single world forever. He wants to take part in infinitely rich, unpredictable, exciting futures. He just hopes someone solves the fragility problem along the way.