Reality+: Virtual Worlds and the Problems of Philosophy

by David J. Chalmers

Assuming that creatures in a simulation are deceived, then Sim Descartes is deceived. If a perfect being is present, then the existence of a perfect...

“What is real? How do you define real? If you’re talking about what you can feel, what you can smell, what you can taste and see, then ‘real’ is simply electrical signals interpreted by your brain.”

The idea is that reality is all in the mind. If something looks like reality and feels like reality (and sounds and smells and tastes like reality), then it is reality. If something appears real, and there’s no appearance to the contrary, it is real. In philosophy, Morpheus’s appearance is reality thesis is a central form of idealism: the thesis that reality is made of minds. Idealism in philosophy isn’t so much about ideals as about ideas. Idealism often says that reality is made of ideas: sensations, thoughts, feelings and other components of the mind.

In Indian philosophy, idealism has been a common view in both Buddhist and Hindu traditions. In his Twenty Verses, an in-depth defense of idealism, the 4th-century-CE Yogācāra philosopher Vasubandhu starts by attributing idealism to the Buddha: Everything in the three realms is nothing but mind. Or as Vasubandhu puts it: Reality is consciousness only. On Vasubandhu’s view, when I see a tree, all there is in reality is the idea of a tree, or the appearance of a tree, or consciousness of a tree. There is no tree outside the mind.

George Berkeley. In his 1710 book A Treatise Concerning the Principles of Human Knowledge, Berkeley put forward his famous slogan, esse is percipi. To be is to be perceived. A spoon exists if it is perceived. This means roughly that if a spoon appears to you, then the spoon is real. To put it another way, appearance is reality.

Berkeley and Vasubandhu argue that reality is made of minds. At the bottom level, there is perception, thought, and feeling. These element serve as building blocks for the world as a whole.

If appearance is reality in this way global skepticism is ruled out.

if our appearances determine reality, this rules out the perfect simulation hypothesis, where there’s never any clue that you’re in a simulation. In such a scenario, there’s a simulation in reality but not the slightest hint of a simulation in appearance. If appearance is reality, that can’t happen.

There are a lot of objections to idealism. One objection is, “Whose minds is reality made of?” If it’s my mind alone, then we have solipsism: I’m the only one who truly exists—or, at least, it’s my mind that makes up the universe. That way megalomania looms. If reality is made up of all our minds together, though, then there will be a gap between my mind and reality as a whole. Maybe where I see a unicorn, everyone else sees an elephant, meaning there’s an elephant in reality. That threatens to bring us back to skepticism. How can I know that everyone else’s perceptions match mine?

Another serious objection: What about unobserved bits of reality? For example, what happens to my desk when I leave the room? And what about parts of the universe where there are no observers, and times long ago, ...

There once was a man who said “God Must think it exceedingly odd If he finds that this tree Continues to be When there’s no one about in the quad.” A reply, from Berkeley’s viewpoint, is expressed in another limerick: Dear sir, your astonishment’s odd. I am always about in the quad. And that’s why the tree Will continue to be, Since observed by, Yours faithfully, God.

As long as God is always watching the whole world, unobserved reality is not a problem. God’s experiences sustain the ongoing reality.

Our own experiences derive from God’s experiences. The constancy of God’s experiences explains why we always see the tree when we return to the quad.

This is okay so far as it goes. But now God has inherited the role that was played by the external world. Instead of having a physical tree out there, sustaining my experience of the tree and everyone else’s, we have God’s idea of the tree, sustaining my experience of the tree and everyone else’s. That...

And if we can’t know, how can we be sure that the tree exists wh...

Another question: Why do we need God here? Couldn’t an evil demon or a si...

The underlying problem for idealism is that in order to explain the regularities in our appearances (the fact that we see an identical tree day after day, say), we need to postulate some reality that lies beyond these appearances and sustains them. Berkeley appeals to God’s mind as this further reality. But now we have created a gap between our own perception and reality, so the skeptical problem rearises.

How can we know about the reality (whether God or an external world) behind the appearances?

Here’s where we can run a version of the Simulation Riposte. We create a rich simulated world, with a simulation of George Berkeley inside it. Sim Berkeley tells us, “Appearance is reality.” Since there is no appearance of a simulation, there is, in actuality, no simulation. Sim Berkeley concludes, “I am not in a simulation. My experiences are all produced by ideas in God’s mind.” From our perspective, Sim Berkeley looks a bit ridiculous. He says he’s not in a simulation, but he is wrong: He is in a simulation. He says that appearance is reality, but there is a vast realm of reality beyond what appears to him. He says his experiences are produced by God, but in fact his experiences are produced by us, with the help of a computer. His world is sustained by the computer, not by ideas in God’s mind. Now, Berkeley might have some comebacks here. He could say that everything in our world is sustained by God’s mind, including the computer. But once we see that a computer can do the job, why do we need God? Berkeley could also say that even if reality goes beyond Sim Berkeley’s appearances, Sim Berkeley’s reality—the world of tables...

If we can’t get evidence for or against the perfect simulation hypothesis, this means at most that it isn’t a scientific hypothesis—one that we can test using the methods of science. But as a philosophical hypothesis about the nature of our world, it makes perfect sense.

verificationism, which says that a hypothesis is meaningful only if it can be verified as true or false by sensory evidence. Verificationism is now widely rejected because it seems that there are many meaningful hypotheses that can’t be verified by sensory evidence.

Bertrand Russell’s appeal to simplicity. The renowned British philosopher argued that the commonsense hypothesis that objects in the external world are real is the simplest explanation of our observations. By comparison, the dream hypothesis is extremely complicated. Presumably he would say the same about the simulation hypothesis. In general, we should accept the simplest explanation of our observations and reject overly complicated explanations. So we should accept the real-world hypothesis and reject the simulation hypothesis.

Ockham’s (or Occam’s) razor, after the 14th-century English philosopher William of Ockham.

Do not multiply entities without necessity! This says that, other things being equal, we should favor the most parsimonious theory—the theory that postulates the fewest things. You should accept a complex theory only when there’s no simpler one consistent with the data.

For example, the ancient mathematician Ptolemy proposed a theory in which the Sun goes around Earth, while the Renaissance astronomer Johannes Kepler proposed a theory in which Earth goes around the Sun. Ptolemy’s theory postulated many epicycles to give the right results, whereas Kepler’s theory worked without postul...

If we focus on hypotheses about the external world, the real-world hypothesis certainly seems simpler than the simulation hypothesis. After all, the simulation hypothesis postulates both a nonsimulated world and a simulated world, whereas the real-world hypothesis has on...

But simplicity is just one factor among many. Often, simple theories turn out to be false, and more complicated theories turn out to be true. Simplicity can be overridden by other factors. One way it can be overridden is when we know there’s complexity in the environment. For example, suppose we find the letter A scratched on a rock on Mars. There are two hypotheses: It was formed by random movements of other rocks, or it was put there by an intelligent being. The first seems simpler, since we otherwise have little reason to postulate intelligent beings on Mars, so we may favor it. On the other hand, if we find a letter A scratched on a rock on Earth, we should favor the intelligent-being hypothesis, even though (because it involves human behavior) it is more complex. We know there...

If we otherwise have no reason to believe there are simulations, then the simplicity of the real-world hypothesis gives us good reason to favor it. On the other hand, if we believe there are many perfect simulations of whole universes in our world, as Bostrom’s simulation argument tends to suggest, then this simplicity reasoning would be overridden. We may not have seen any perfect simulations, but we have good reason to believe that they’re possible and may well be developed at ...

Russell’s colleague G. E. Moore offered another famous reply to the external-world skeptics. Moore said “Here is one hand. Here is another. Therefore the external world exists.” Moore called this a proof of the external world, arguing that the premises are obviously true and are far more plausible than any piece of philosophy. Given that there are hands, there must be an external world.

From the inside, SimUniverse will be indistinguishable from the universe it’s a simulation of. Suppose that we’re simulating a possible universe containing ten billion people; in that case, SimUniverse will contain a simulated person—a pure sim—for each of them. After a while, every teenager may be running SimUniverse on a mobile device. Even if the use of the technology is restricted, we can imagine researchers running many simulated universes for scientific, historical, financial, and military purposes. Within a century or two, this could easily lead to millions or billions of different copies of Sim Universe running here and there. As a result, there will be vastly more sims than nonsims. Across history, sims may outnumber nonsims by at least a million to one. The same goes for intelligent beings anywhere in the cosmos. If any aliens have human-level intelligence, they should eventually develop computers and program them. If these alien civilizations survive long enough, they’ll likely create simulated universes. Let’s run some numbers, using small numbers to keep things simple. Any nonsim population that survives long enough will eventually be able to create (let’s say) at least a thousand sim populations, each with about as many sims as the original nonsim population. It’s arguable that at least one in ten human-level nonsim populations will eventually do this. If one in ten nonsim populations creates at least a thousand sim populations each (as depicted in figure 14)...

We can then ask: What are the odds that we are among the relatively few nonsimulated beings? Since sims outnumber nonsims by at least a hundred to one, the natural answer is “less than 1 percent.” It is much more likely that we’re sims than that we’re nonsims. Conclusion: We are probably in a simulation.

The definitive version of the simulation argument was put forward by Nick Bostrom in his 2003 article “Are You Living in a Computer Simulation?” Bostrom gives a mathematical argument for a complicated conclusion with a choice between three options, focusing especially on a version of the simulation hypothesis that involves simulating one’s ancestors.

The reasoning is straightforward. Simulation technology is likely to be so ubiquitous that most beings in the universe (or most beings with experiences like ours) are sims. If so, then we are probably sims. This reasoning is not bad, but it’s not irresistible. Where might it go wrong?

One sort of objection says: It will never happen! You might deny that there will ever be many sim populations with human-level intelligence. This might be because simulation is impossible, or at least too difficult. It might be because no one will choose to make simulations. It might be because all human-level populations will die before they can build the simulations. If so, then there won’t be many (or any) simulations, and we’re much less likely to be in a simulation. Another sort of objection says: We’re special. You might say that even if there are lots of sim populations, we have special features that make us unlikely to be sims. For example, we are conscious, and we have very specific sorts of minds, and some might deny that sims could be conscious or have minds like ours. Alternatively, you might hold that we live in a distinctive sort of world, and that most simulated worlds won’t be anything like ours. If so, then even if there are many simulations, we are much less likely to be in one.

Let’s say that intelligent beings (or just “beings,” for short) are creatures with at least human-level intelligence.

with the ability to program computers being especially relevant.

We can then lay out the argument as follows. This version of the argument is far from perfect (I’ll give my preferred version toward the end of the chapter), but it’s a good starting point that helps to bring out the underlying issues. At least one in ten nonsim populations will each create a thousand sim populations. If at least one in ten nonsim populations will each create a thousand sim populations, then at least 99 percent of intelligent beings are sims. If at least 99 percent of intelligent beings are sims, we are probably sims. ______________________ So: We are probably sims.

The objections to premise 1 are

It will never happen objections: Simulation is impossible, simulation is too difficult, nonsim populations will all die before creating sims, nonsims will choose not to make simulations.

Intelligent sims are impossible. One objection is that processes that produce intelligent behavior are uncomputable: that is, they can never be successfully simulated on a computer. This could be because the nonphysical mind affects behavior in uncomputable ways. It could also be because there are physical processes in the brain that can’t be simulated. For example, the mathematical physicist Roger Penrose has speculated that a quantum gravity theory (that is, a theory unifying quantum mechanics and general relativity) may involve processes with a nonalgorithmic element that is crucial to human behavior. If so, it could turn out that intelligent sims are impossible—and therefore that sims are impossible, as we’ve said that sims are intelligent by definition. The existence of these processes would be surprising as there is currently little evidence of uncomputable processes in nature.

Even if there are processes in nature that no classical computer can simulate, it’s arguable that they could be harnessed to build a new sort of more powerful computer. We already know that quantum mechanics can be harnessed to build quantum computers.

Penrose is right that quantum gravity involves processes that are not classically computable, we should be able to harness these processes to build more powerful quantum gravity computers that no classical computer can simulate. Then these quantum gravity computers could simulate our brain processes, and we’d end up wi...

Sims take too much compu...

This is far from obvious. Brains are large but finite. They contain around 100 billion neurons with around 1,000 connections (or synapses) each. On current estimates, the brain performs the equivalent of around 10 quadrillion (1016) floating point operations per second (or flops): a computing speed also known as 10 petaflops. That’s a lot, but it’s about on a par with the best existing supercomputers.

If this is right, then once we know enough about the brain, one second of supercomputer time should be able to simulate one second of brain processing. If technology advances at anything like its usual speed, we can expect computers to speed up by a factor of ten every decade, or a factor of 10 billion (1010) over a century. That suggests that within a century (at a speed of 1026 flops), a second of computer time will be able to simulate 10 billion brains for a second each. Within another century (at a speed of 1036 flops), a second of computer time will be able to simulate 10 billion brains for a lifetime of 100 years (or 3 billion seconds) each. A full-scale simulation will require the environment to be simulated, too, but it’s hard to see why this should add more than another two or three orders of magnitude to the workload. All this would require a computing speed of around 1039 flops. Even if progress slows, this should in principle within the range of future computers.

computronium, a hypothetical state in which matter is used as efficiently as possible for computing.

Nonsims will die out before creating sims.

the hypothesis that intelligent civilizations inevitably destroy themselves should be taken seriously.

This hypothesis might explain some of our observations. For example, it would explain why we’ve never seen signs of extraterrestrial intelligence: Intelligent populations destroy themselves around the same time that they gain the ability to send signals. It might also explain why we seem to be living early in the history of humanity. This issue is at the core of the so-called Doomsday argument, developed by the astrophysicist Brandon Carter and the philosopher John Leslie. Their key claim is that probabilistically any one of us should expect to be somewhere in the middle of all the humans who have ever lived or will live. Given the rapid increase of human population, this means we should expect human life to end within centuries (in which case we’d be comfortably in the middle of all humans) rather than to go on for millions of years (in which case we’d be extremely early).

That said, the hypothesis that almost all human-level populations die before they’re able to create many simulations would be surprising if true. One could reasonably hope that at least one in ten human-level populations will be collectively rational enough not to destroy itself; or if that’s too optimisti...