Reality+: Virtual Worlds and the Problems of Philosophy

by David J. Chalmers

Nonsims will choose not to create sims. In this scenario, human-level populations develop the ability to create many population simulations, but they don’t. Perhaps they think it’s too risky. It may be that the only populations to survive all the destructive technologies are extremely risk-averse and are worried, for example, about sims escaping from the simulation and taking over their world. Perhaps they think it’s unethical...

Given the ability to create simulations and strong incentives to do so, it would be surprising if few populations create them.

More nonsims than sims will be created.

we should expect humanlike sims to greatly outnumber humanlike nonsims. (Here humanlike beings are those with experiences broadly like ours.) Still, the argument has an underlying assumption: that it will be cheaper and easier for nonsims to create humanlike sims than to create humanlike nonsims. If it turns out to be easier to create nonsims for the relevant purposes—perhaps using nanotechnology, exploiting infinite space, or creating baby universes—then we’d expect nonsims to proliferate instead.

We have not definitively ruled out sim blockers, but any of them would be somewhat surprising. We certainly can’t know that there will be uncomputable laws, insufficient computer power, near-universal extinction, a near-universal choice to avoid simulations, or more efficient nonsims. And if we can’t know that there are sim blockers, then for all we know, most intelligent beings are sims.

More precisely, it is a feature that a sim is more likely to have than a nonsim. For example, a sim is perhaps more likely than a nonsim to experience glitches in physical reality arising from approximations, shortcuts, and programming errors—a Matrix-style experience of the same cat crossing one’s path twice, say. If so, these glitches are sim signs.

Perhaps our most significant sim sign is that we seem to live quite early in the universe. We haven’t discovered intelligent life elsewhere in the universe,

and we haven’t yet created simulated universes with intelligent beings of their own.

where sims are concerned, it’s likely that early-universe simulations will be especially common, partly because later creatures may be interested in simulating their history, and partly because early-universe simulations will be far less demanding than later-universe simulations.

A nonsim sign is a feature that tends to attach to nonsims. More precisely, it is a feature that a nonsim is more likely to have than a sim. In the world where all sims get a “You are a sim” signal, the absence of such a signal is a nonsim sign. If we know we have a nonsim sign, then even if we know that 99 percent of beings are sims, we should be less than 99 percent confident that we are sims.

Some important objections to premise 3 point to nonsim signs. These are the objections that say, We’re special. The potential nonsim signs here include consciousness (simulations won’t be conscious), our minds more generally (simulated minds won’t work like ours), the complexity of the world (simulated worlds will be simpler than ours), and more.

Sims can’t be conscious! The most obvious potential nonsim sign is consciousness itself. Suppose one thinks, as some philosophers do, that only biological systems can be conscious and that therefore simulations can’t be conscious. Given this view, the fact that we’re conscious will indicate that we aren’t simulations. We could be ...

This view is controversial. I’ll argue in chapter 15 that it’s false and that simulated beings will be as conscio...

Nick Bostrom rules out the view that simulations aren’t conscious by making an assumption of substrate-independence (or equivalently, substrate-neutrality)—that is, consciousness depends only on the organization of a system and does not depend on the subs...

Simulators will avoid creating conscious sims.

ethical simulators may aim to use the nonconscious versions wherever possible.

Sims won’t have minds like ours.

you’re certain that an aspect of your mind—consciousness, say—cannot be replicated in a simulation, then you’ll regard that aspect as an absolute nonsim sign. If it’s merely less likely that this aspect—emotions, say—will be found in simulations, then it will be a probabilistic nonsim sign.

Sims won’t experience large universes.

shortcut simulation—a simulation that takes shortcuts so that our world is not as large as it seems.

To simulate where we’ve been, the people we interact with, and the media we read and watch, it will need to simulate a fair amount of our planet, and it will need reasonably detailed simulations of the Sun, Moon, and other planets (of which we now have detailed images). We’ll need a decent simulation of at least the visible stars and galaxies, and of background radiation and other observable phenomena. The simulators will need to be ready to expand the simulation—for example, if we travel to the stars or acquire new ways of getting information from them. Expandable simulations are already familiar in video games such as No Man’s Sky, in which new planets are algorithmically generated as players travel to them. Sophisticated simulators may be masters of these techniques, knowing just where a local simulation can cut corners. Something similar goes for using shortcuts to simulate the microscopic depths of the world. In some contexts, a simple Newtonian physics of macroscopic objects may be all we need, but there are many contexts for which more is needed. Many observable properties of ordinary objects depend on chemistry, which itself depends on quantum mechanics, so it will be difficult to simulate macroscopic objects really well without going very deep. Making our simulation consistent with the observations of scientists about atomic physics and the like will require even more work. Perhaps not every last detail will have to be simulated. Some low-level details may never s...

It is arguable that this degree of complexity will be relatively rare in simulations.

Nevertheless, it remains plausible that simulators will create a reasonably large number of complex-world simulations. If so, we can expect most beings with complex-world experiences to be sims.

The potential nonsim signs we’ve considered, such as consciousness and a large world, may decrease the probability that we’re in a simulation. At the same time, we need to weigh these against potential sim signs, such as the fact that we seem to be early in the universe, which may increase the probability that we’re in a simulation. Do the sim signs outweigh the nonsim signs, or vice versa?

Nick Bostrom takes another approach to the sim sign issue by focusing only on ancestor simulations: exact simulations of the entire mental history of humankind. Any ancestor simulation of my world will include an exact simulation of me. If there are many ancestor simulations of my world, it is guaranteed that there will be many sims with experiences just like mine.

If so, there is no need to worry that my experience contains nonsim signs. Every feature of my experience will be replicated in many sims.

I don’t think the “ancestor simulation” version of the argument works in this form, since I don’t see good reason to believe that the...

Constructing such simulations would require knowing something close to the exact state of human brains at every point in human history, and there’s not much reason to think that’s possible. Perhaps it will be possible to do this inside simulated universes (via backup records for simulated brains, for...

Bostrom later notes that for the argument to work, the simulated creatures needn’t have exactly our experiences. It is good enough if they have “human-type” experiences, the ...

I think this is right, though it means that there is no longer any need to mention ancestor simulations in the argument. Any human-type simulations are good enough. In fact, I think “human-type” experiences as Bostrom defines them may still be unnecessarily narrow. In principle, the argument will still work with a broader class of minds,...

I would therefore focus the argument more broadly on the likelihood that there...

Humanlike beings are beings with roughly the same major sim signs and nonsim signs as humans: for example, they’re conscious, they experience a large universe, and their society is...

it’s possible to formulate the simulation argument in a simple general form

If there are no sim blockers, most humanlike beings are sims. If most humanlike beings are sims, we are probably sims. ______________________ So: If there are no sim blockers, we are probably sims.

“Most humanlike beings are sims” means that most humanlike beings in the cosmos (including past and future beings, beings who created us or whom we created, and so on) are sims.

Importantly, a sim blocker is now defined as something that prevents the creation of enough humanlike sims to ensure that most humanlike beings will be sims.

Because premise 1 builds in If there are no sim blockers as a condition, sim blockers are no longer an objection to it. Premise 1 now requires only the plausible assumption that if nothing prevents the creation of many humanlike sims (enough of them that most humanlike beings are sims), then there will be many humanlike sims.

Because premise 2 builds in most humanlike beings, nonsim signs are no longer an objection to it. Premise 2 now requires only the assumption that if there are many beings with the same sort of experience as...

This is sometimes called an indifference principle, because it recommends indifference between each of th...

From this assumption it follows that if 90 percent of beings with experiences like mine are sims, then I should be 90 p...

Even if we accept premises 1 and 2, we’ve really just relocated the sim blocker and sim sign issues. In weakening the premises, we’ve also weakened the conclusion, which now explicitly builds in the possibility of sim blockers. Furthermore, the notion of a sim blocker has now been broadened to include anything that blocks creation of enough humanlike sims. As a result, the notion of a sim blocker now covers things that we previously counted as nonsim signs. For example, Sims won’t be conscious is now a potential sim blocker: if conscious sims are impossible, then humanlike sims are impossible. Sims ...

From the conclusion, it follows that we can be highly confident in Either there are si...

Bostrom puts the conclusion of his argument in roughly this form: This paper argues that at least one of the following propositions is true: (1) the human species is very likely to go extinct before reaching a “posthuman” stage; (2) any posthuman civilization is extremely unlikely to run a significant number of simulations of their evolutionary history (or variations thereof); (3) we are almost certainly living in a computer simulation.

Bostrom’s options (1) and (2) are both sim blockers, closely related to Nonsims will all die before creating sims and Nonsims will choose not to create sims. These are reasonable sim blockers to consider, but they’re far from the only ones.

Of the sim blockers for humanlike sims we’ve considered, I’d add at least Intelligent sims are impossible, Conscious sims are impossible, Sims take too much computer power, Simulators will avoid creating conscious sims, and More nonsims than sims will be created. Instead of Bostrom’s three-way co...

A simpler approach is to divide sim blockers into two groups. First, it could be that humanlike sims are impossible or highly impractical to make (to simplify, I’ll understand “possible” to mean “practically possible,” so that impracticality counts as impossibility). This group includes sim blockers like Sims won’t b...

Second, it could be that humanlike sims are possible and practical, but few humanlike populations will create them (in sufficient numbers for most humanlike beings to be sims). This group includes sim blockers like Nonsims will die before creating sims, Nonsims will choose not to create sims, Simulators...

If this is right, we can cash out the conclusion more explicitly in a three-way form. My conclusion is that we should be highly confident that either (1) we are sims, or (2) humanlike sims are impossible, or (3) humanlike s...

We can’t know the physics of the next universe up;

Are we in a simulation? What are the implications of the simulation argument for skepticism and the Knowledge Question?