Simulation scenarios and philosophy

I will mostly discuss the epistemological side of Reality+, especially the idea that we should take seriously the possibility that we are, right now, living inside some sort of computer simulation.1 I'll argue against this view; I don't take the possibility seriously. I won't worry about how many of our beliefs are true rather than false if we are inside a simulation. For Chalmers (2022), “virtual reality is genuine reality” and the familiar objects around us still exist in this scenario. I assume that simulation hypotheses are very surprising no matter how that issue is handled, especially as our lives are observed and controlled by hidden powerful agents. Chalmers gives scenarios like this a probability of about 25%. My discussion has three parts. I first make some arguments more or less within Chalmers's background assumptions. I then discuss the “substrate independence” of the mental. This bears on the question of what kinds of simulations we might inhabit – what sort of hardware would be needed. At the end, I return to the epistemological side and look at questions about time and memory. Reality+ argues that we should take simulation scenarios about our present situation seriously because our own picture of things tells us to. That picture tells us that some of these scenarios are technologically plausible, and will probably be realized in the future if not already. “Our picture” includes mainstream science and reasonable guesses about what intelligent agents are likely to want to do. All this we use to show that we should doubt some of our usual assumptions about what is going on around us. That “our own picture tells us…” element is important. If, at some stage, the discussion lapses back into “Who knows? Something weird might be happening,” then that's a different matter, and not one that puts much pressure on us. At least partly embodied brain-in-vat or Matrix sorts of beings, with a biology, controlled by agents of some sort, Pure software beings in a computing artifact, also with a hidden controller of some kind (or a history of hidden control), or Boltzmann brains – biological systems, or perhaps other physical systems not tied to computing artifacts, not controlled but spontaneously arising and probably short-lived.2 Let's consider first the scenarios with hidden agents involved. We are in a simulation that has been built for us to inhabit (or perhaps one simulation for each of us), by someone we don't know about, for their purposes. There is a lot of computing technology behind it – a fine-grained model of a world we interact with, a model feeding us sensations and responding to our decisions, whether we turn our head, blink, drive 100 miles east, or build a microscope. I wonder whether Chalmers underestimates the difficulty of this task – as candidates for VR-resistant or simulation-resistant experiences, I would offer a hot shower, or a skateboard crash. But I'll not worry much about that side of the feasibility question. The threat to our familiar assumptions is to come from the idea that a large majority of minds like ours, a majority over some long span of time, will inhabit a simulation of some sort. The minds inhabiting these simulations won't be able to tell that this going on. Given that (probably) most minds will be in this situation, why not ours? Why would people (or other agents, ChatGPT-11 or whoever) bother to do this? In large numbers, what would be the point? It might be a research project. Why then make it “total,” in the way it seems to be, rather than local in space and time? You might say: we don't know, and as long as there's a chance that they might get up to this, the argument can be posed. We can introduce a multiplier, a probability-reducer, in the argument. The pressure might be reduced but it is still real. The question of motive has more effect than this, though. Again, the argument is supposed to be that within our own picture, we can see why it would make sense for things to look completely different, to most sentient agents, from how they really are. Whenever the argument includes a “Who knows why they'd do it?” or “Who knows how they'd manage this?” step, it takes a lot away. Perhaps it's not science but recreation? It would be a rather sadistic form of recreation, if our virtual world is more or less as it appears in terms of who experiences what. (A “problem of evil” arises for some versions of this view.3) There's a lot of unexplained deception, anyway. Might it be something like the scenario in The Matrix? Perhaps the motive is economic, and we are a power supply? The Matrix was a good movie, but it was just a movie; the scenario made no energetic sense at all. They imagined humans being used as a source of electricity. But we've got to be fed. So you're putting chemical energy in to get electricity out. That is the opposite of what it would make sense to try to do, as the renewable energy industry has learned; the hard process is to get from electricity, via solar and wind, to a good storage medium. Our picture says that what they were doing in the movie makes no energetic-economic sense and we should not expect anything like it to actually happen. Any biologically based simulation is expensive, for reasons like these. It has to be a “pure” sim, where people are just software, in order to make large numbers feasible. And large numbers of people inhabiting simulations are needed in order to exert probabilistic pressure on our usual picture. Is a “pure sim” scenario plausible? If we are “pure sims,” in a computer or being shuttled around between computers, then the mental, including the experiential, has to have a lot of “substrate neutrality” or “substrate independence” (SI). Are there good reasons to believe this? I used the phrase “a lot” of SI, which makes it sound like a matter of degree. This is indeed how we should think about SI, and I don't think there's much of it. Other people think the mental just is substrate-independent, or at least that there's a great deal of it. This is a major disagreement, and not something that can be resolved in a short commentary. I can argue, though, that some of the reasons people give for believing in SI (or a lot of it) are not good reasons. I can also sketch an alternative, though that will be more speculative. The brain simulation is a digital system running on a computer. … We can suppose that every neuron is simulated perfectly, as is every glial cell and other cells throughout the brain. The interactions between neurons are simulated perfectly, too. All the electrochemical activity is simulated, and so is other activity, such as blood flow. If there's a physical process in the brain that makes a difference in how the brain functions, it will be simulated. (p. 287-88) Philosophers have gotten used to talking like this, but the “perfect” duplicate idea is problematic. We can start by thinking about details of timing. Some facts about timing in a brain are certainly “functionally relevant” – the time things take as a whole, also the simultaneity of parallel processes. Other details of timing look like they don't matter much, but they're still there; the brain processes are different if they're changed. Time tends to be ignored in classical functionalist discussions – you've got the machine table, or imagined network, and if it gets things done, then it gets things done. But biologically, timing is a big deal. How much fidelity of timing must be present in a “perfect” duplicate? That is an ill-posed question. Suppose something takes a millisecond longer, or the synchrony of two processes is not quite the same… What there is, in these cases, is functional similarity to various degrees. Timing can make the point, as can other details – a few more ions here or there. Chalmers says, “If there's a physical process in the brain that makes a difference in how the brain functions, it will be simulated” – but is another millisecond a “difference”? Yes, but a tiny one. It's a difference to experience, too; if the process that is the basis of experience goes on for another millisecond, then experience is different. It might be denied that millisecond differences are part of experience. Perhaps one physical system might duplicate another in all ways that matter to experience, even if there are microfunctional differences as well. This thought can reflect a cause-effect model for the relation between brain processes and experiences, a model that is covertly dualistic. If sensations are brain processes, then when the relevant brain processes go on for a bit longer, so does experience. The denial that small timing differences are part of experience might also be based on the thought that no one could compare two experienced events from the “inside,” and judge that one took a millisecond longer than the other. But they might still be different in this way.4 So far I've been opposing a way that these issues are often framed. I'll now offer some more positive ideas that push away from SI. What is it within a brain that is the basis for felt experience? It might be just the neuron-to-neuron effects that involve action potentials, but it might not be. A family of views holds that biological properties other than these are important, and these tend to be hard to realize or approximate in a nonbiological system. 5 Within this family, the view I defend features a role for large-scale dynamic patterns in the brain. Brain activity includes both relatively discrete, point-to-point influences between cells, and less local electrical oscillations that seem to involve movements of ions across cell membranes below the threshold that initiates a “spike.” Sub-threshold oscillations affect spikes, and spikes affect the oscillations. The literature on the apparent experience-relevance of these processes is now large.6 Perhaps the details of this view are wrong. Other views in the family I mentioned have a similar upshot – they reject the idea that all that matters to experience in our brains is the networked interactions that lend themselves to realization in a computer. They might be wrong, too, but I do want to resist the idea at the present time we have positive reason to believe in (much) SI. Both the points I made about functional similarity and the ones about brain processes matter to the believability of “neural replacement scenarios” often used in support of SI. During a slow replacement of neurons with physically different devices, the system will change micro-functionally. It will also change micro-behaviorally, in all sorts of details. What is happening, inside and outside, is different. As a result, there's no reason to believe that experience is unaffected in a slow replacement. First a little, then a lot. This could include all sorts of transformation and fading. In response to these arguments about large-scale dynamic patterns and other biological features of nervous systems, a reply might be, “We can simulate those features, too.” But there's a difference, especially in cases like this, between simulation in the sense of representation and simulation in the sense of realization. It can't just be a representation; it would not be enough to handle large-scale oscillations by having a collection of memory addresses that form a coordinate system, and that you write values to very quickly, going from one represented value of a variable to another within the usual hardware. There needs to be a version of the process itself taking place in the nonbiological, non-cellular hardware. If a lot of SI is unlikely, large numbers of simulated minds are less likely. Setting aside, again, the difficulty of getting any sort of simulation to provide minds with a realistic pattern of experience (remember the hot shower), simulated minds themselves are only cheap if we're talking about solar-powered computers with conscious software. Otherwise they're expensive, and then they'll probably not be very numerous, if they come to exist at all. Many discussions of skepticism about the external world include a convenient lacuna. If we're going to consider the possibility that our physical embedding in our environment is completely different from what we usually assume, due to large-scale deception or something similar, then we should suspend belief in just about everything about the past, including one's recent remembered experiences, as well. We shouldn't entertain an external-world “spatial” suspension of belief without taking seriously the temporal side. This is one reason I like Alex Proyas's movie Dark City, a film in which skepticism about the past is true. The ordinary people in the movie have their memories artificially reconfigured every 24 hours (or half day?), as part of something like an alien experiment. The kind of consistency I have in mind would require that one doubt everything in one's mental history from perhaps half a second or so ago. Some have argued from this to an anti-skeptical conclusion. Rinard (2018) argues that external-world skepticism leads to skepticism about “complex reasoning,” and this leads to a kind of reductio. My attitude is different. It's not that this gives us a path out of skepticism, but some kinds of “working one's way back” projects collapse. If you don't at each moment take yourself to be reasonably sure what you were thinking a few seconds ago, you can't have confidence in conclusions, even small ones, reached at previous steps in a train of thought. Reasoning involves a tacit assumption of mental continuity. As the theoretical physicist Sean Carroll has pointed out, however, the thesis that I am almost certainly a Boltzmann brain is “cognitively unstable.” If it's true, then I cannot stably endorse it. If I endorse it, I then must endorse that my perception of the external world is almost certainly an illusion. But then I must reject all of my scientific reasoning that's based on my perception of the external world. In particular, I should reject the scientific reasoning that led to the physical theories on which the existence of Boltzmann brains was based in the first place. Those theories are the only reason to take the Boltzmann brain hypothesis seriously. Without those theories as support, we're back to the original situation: The hypothesis that I am a Boltzmann brain is extremely improbable. (p. 457) I don't agree that thoughts like these get us “back to the original situation,” confident that we are not Boltzmann brains. The Boltzmann brain idea brings those issues about time and reasoning to the fore. If Boltzmann brains ever exist, most will be very short lived. To wonder if one is a Boltzmann brain is not just to suspect that one can't rely on a body of scientific reasoning and history of experimentation. It's to suspect that one can't rely on even short trains of thought, on any topic. This bears on the discussion of simulations in earlier chapters of Reality+. To seriously suspect you are in a simulation is to suspect that many or all of your memory traces might be cooked up, and your background knowledge is no good at all. You could not trust whatever might be the apparent basis for your assumptions about what the simulators might be up to and why. Chalmers, echoing Carroll, says that when we entertain the Boltzmann brain possibility, this requires that we stop trusting the ideas in physics that made Boltzmann brains seem likely. The same point applies to simulation hypotheses – to suspect that we are in such a situation requires that we stop trusting a lot of what was used to get us to that thought. The Boltzmann brain case is extreme, as a very short time window is built into the scenario, but in the other simulation scenarios, much of what's in our memories should not be trusted. If the Boltzmann's brain scenario is “cognitively unstable,” this would apply to other simulation hypotheses as well. Does this mean both scenarios are self-undermining? Not in the sense seen in the Carroll and Chalmers treatment of Boltzmann brains, where an anti-skeptical conclusion is supposed to follow, and also not in a sense that would stop us going through an exercise like the one in the early parts of Reality+. We can still ask, as Chalmers does and as I did earlier in this commentary, whether our picture of things gives us reason to think there will come to be large numbers of minds with experiences like ours existing within simulations. I answered no to that question. We can do the same thing with Boltzmann brains. Do we have reason to think that the universe will come to be populated by huge numbers of short-lived brains having experiences like ours? I'd have to hear more than the very general points usually offered in defense of this scenario, but in this case as with the other one, a yes answer would be interestingly disconcerting. We can do this, taking a lot for granted as we go. We might also try to do something else, and try to work out what we can conclude from what is available to us, without making the usual “external world” assumptions and also without assuming the integrity of memory. Then we can't rely on our memory of how the world has seemed to work, or even what seemed to be in our minds a few seconds ago, and we won't get far.