Bostrom's Simulation Argument is a well-known philosophical argument, a trilemma that can be taken to suggest that it is likely that we are living in a simulation. The argument suggests that, perhaps, our descendants became vastly more technologically advanced than we are, today. Out of curiosity, they began to simulate past histories and we happen to inhabit one of those simulations. One property of a computer simulation is that whoever controls the simulation has absolute control over time, so objections to Bostrom's argument based on the subjective passage of time are unconvincing.
This argument is very interesting and might even be convincing to some people. But I do not find the argument in itself convincing and I think that it fails to pierce through to the root issues.
When we talk about simulating the entire world (physics, the five senses, psychological effects, and so on), we run into two problems. The first problem is called the substrate problem -- if the observable world is a simulation running on a computer "somewhere else", what is simulating that other place? The term "base reality" is sometimes used to refer to some kind of "really real" world, but the moment we set sail on the roiling ocean of simulation theory, it's too late to wish we were back on the terra firma of "base reality." If our world is being simulated in a way that is so convincing that it cannot be distinguished from "base reality," then how could any conscious mind distinguish between being in a simulation and being in "base reality"? If we admit the possibility of simulation, then the idea of "base reality" must be jettisoned entirely.
The second problem is the problem of metaphysical contingency of the simulation. In plain language, physics feels convincing to us (in the sense of being "really real", not just a dream or illusion of some kind) because the laws of physics are, in some sense, inevitable, inexorable. The laws of physics seem to be logically necessary (non-contingent).
This point is closely related to the previous point. What we mean when we use a term like "base reality" is that which could not be any other way. In metaphysics, a world in which things could not be otherwise would be a logically necessary (non-contingent) world. It is generally held that necessary worlds are abstract (not concrete) and that the concrete world is inherently contingent (non-necessary). Part of what we mean when we say that the concrete world is real (not an illusion) is that it could be some other way than it is, that things could have unfolded some other way. It seems that there is some ineradicable element of indeterminacy to the physical world, and this indeterminacy is part and parcel of why the real world feels real. If we were trapped in some kind of immersive, scripted, movie-like illusion, we would be aware of its farcical nature, even if we were unable to escape it or do anything about it. Of course, the mere contemplation of such a dimension seems hellish to us and I will argue that this is one of the reasons that the simulation argument is so difficult to handle correctly.
So, what we're looking for is one or more principles that would explain (a) why the world must be a simulation and (b) why the simulation is convincing in all the ways that we mean that the observable physical world is convincing (does not seem to be an illusion, dream or farce).
The first principle that we are looking for is the substrate-independence of computation. What this means is that it does not matter what machinery you use to perform a computation, the computation itself can be performed by any suitable mechanism. Today, most computers are built from silicon transistors in integrated circuits but computers have been built from discrete transistors, electronic tubes, mechanical gearing systems, and so on. The determined hobbyist could construct any Boolean logic function from water, pipes and gravity -- a purely passive system, discounting the potential energy of the water in the reservoir.
The second principle that we are looking for is the universal utility of computation for making choices. Any mind -- even an artificial, non-cognitive mind -- will find that computation is useful for making choices. Our mammalian brains use vast amounts of physical computation to inform the choices that we make on a conscious level, even though we are almost completely unaware of all the real computation that our brains are performing. But even from a purely abstract perspective, we find that computation is the foundation for choice-making. Computation can allow an agent, even in a purely abstract game, such as Chess or Go, to recognize losing choices and identify choices that will bolster its own aims and ends.
The third principle that we are looking for is the universal summum bonum (henceforth SB.) In plain language, this means that we are looking for the goal or aim that any agent (that is, being capable of making choices) would have, or would tend to have. Philosophers have been arguing for thousands of years about what is the SB (Latin for "highest good") and we cannot solve that argument, here. However, we can provide a reasonable proxy for the SB that I think serves as a placeholder for whatever the actual SB might one day turn out to be when human philosophy has finally become sufficiently enlightened to agree on such matters.
What is this proxy SB, you ask? First, we must note that a SB that is "beyond reasonable argument" will have to be objective, since humans will always argue about any subjective point, whether out of obstinacy, devotion to the truth, or for some other reason. Second, this proxy SB must be sufficiently all-encompassing that no one could reasonably name something bigger or greater. I think we already have such a SB although the general awareness of this SB is still not very high.
Let us suppose, for a moment, that the famous P vs. NP problem will one day turn out, to everyone's shock and surprise, to resolve to P=NP. I'm not going to review the P vs. NP problem here, there are many good introductions to it. What I will do is quote an informal argument given by Scott Aaronson for why would should believe that P≠NP:
If P=NP, then the world would be a profoundly different place than we usually assume it to be. There would be no special value in “creative leaps,” no fundamental gap between solving a problem and recognizing the solution once it’s found. Everyone who could appreciate a symphony would be Mozart; everyone who could follow a step-by-step argument would be Gauss; everyone who could recognize a good investment strategy would be Warren Buffett. It’s possible to put the point in Darwinian terms: if this is the sort of universe we inhabited, why wouldn’t we already have evolved to take advantage of it?On its face, Aaronson's point is very persuasive. Despite all the technical reasons to believe that P≠NP, this non-technical argument probably has the strongest appeal to anyone's intuition.
But...
Let us suppose that we really do live in a simulation. Then, it is possible that the conditions of the observable world have been tuned by the Simulators to make it seem, for whatever reason, that P≠NP. For the purpose of physics proper, such speculations are obviously useless. But for the purposes of metaphysics, such speculations are unavoidable... we're not doing real metaphysics if we simply shy away from contemplating inconvenient possibilities.
I intentionally capitalized "Simulators" in order to play on the common notion of simulation as some kind of computer-driven hall-of-mirrors. But I think this notion of simulation is deeply flawed and leads to all kinds of mistakes of reasoning about the simulation argument. At a very high level, the primary uses of computer simulation in the year 2020 are: science/technology, commerce, governance, warfare and entertainment. When we talk about world simulation, and we try to imagine generalizations of any of these contemporary uses of simulation to a universal scale, the consequences are invariably terrifying. A commerce-and-entertainment-driven simulation could lead to a dystopian world like that in Westworld or Ready, Player One. A warfare-driven simulation could lead to a dystopian nightmare like that depicted in the Terminator series or Source Code. A purely technological, machine-driven simulation could lead to a dystopian nightmare like that depicted in The Matrix series. And so on.
But all of these contemporary, human applications of computer simulation have a very contingent (non-necessary) flavor about them. For example, we need entertainment but it is not at all obvious that we need computerized entertainment. So, it is hard to imagine how there is anything truly universal about an entertainment-driven simulation. Similar arguments can be made for each of the major human interests that drive modern applications of computer simulation. Simply compare human uses of computation to natural ones (bearing in mind that all living systems, even viruses, are fundamentally computational systems.) In short, our contemporary applications of computer simulation are all distinctly anthropocentric, contingent, non-necessary.
What does this have to do with the P vs. NP problem? Well, there is one constraint that binds all living things in the observable world: scarcity. We live under the constraint of scarcity of space (mechanical mutual exclusion), scarcity of time (imminent bodily death), scarcity of energy, scarcity of mental capacity, memory, attention, and so on. All of these constraints prevent us from being the Mozart, Gauss or Buffett that Aaronson points out we would be if we lived in a world where P=NP. That we do not know how to escape these constraints does not, in itself, prove that these constraints are actually inescapable.
We earlier supposed, for the sake of argument, that P=NP. The point of this supposition is to present the following conundrum. Suppose that an unmistakably divine oracle or angelic messenger appeared to humanity and gave us the following message: "To all humanity: be it known that you will one day solve the problem of P vs. NP and you will succeed in proving that P=NP because this is the case. However, the arduous task of finding this proof has been left to you to discover. We look forward to you solving it and joining us in the Higher Realm." I think that basically every reasonable mathematician, scientist, computer scientist, engineer, and so on, would realize that the single most valuable goal of all human effort would be to discover the proof that P=NP. But the heavenly envoy already informed us that this task would be arduous -- the real question would then be, how arduous?
The most obvious, "dumb" method to solve the P vs. NP problem would be to apply a technique called proof-search. The basic idea is to just use blind search - write out every syntactically valid formal proof, in order, and check whether its conclusion is "P=NP". Since we have already been given the answer by the divine oracle, we know that this proof-search problem is decidable -- if we search long enough, eventually, our proof-search will halt and it will give us the proof that P=NP. Unfortunately, even knowing that P=NP, the question of how long we will have to search to find the proof is still undecidable. Given that mathematicians, as of yet, have no better ideas on how to prove that P=NP, it would be inexcusable not to devote all free computing resources to this monumental task. Perhaps with all free computing resources devoted to the task, globally, it will only take a few months or a year to find the solution.
But the problem could turn out to be much, much harder to solve. In this case, brute-force proof-search will not yield the answer, even after months or years of running. The next step is to design more intelligent proof-search. In this case, we would want to build better tools for mathematicians, tools that help convert their brains into "proof-search cores." Basically, we imagine the brains of all mathematicians on the planet as some kind of massively parallel, wetware computer that is driving search heuristics into our already-existing blind search proof-searcher. We want to accelerate this wetware computer as much as possible by building powerful assistive machines that handle all the intercommunication and processing tasks as seamlessly as possible. Between the massively parallel wetware computer comprised of the brains of human mathematicians, and the mechanical proof-searcher that they are guiding, we would have heuristic proof-search that is at least as intelligent as the most intelligent mathematician in the world and, hopefully, much more intelligent than that, collectively.
But the problem could turn out to be so hard that even this wetware-based heuristic proof-search machine fails to solve it. In this case, we would have to re-task the wetware proof-search machine. Rather than trying to directly solve the problem of P=NP, this proof-search machine would instead turn to solving a less ambitious problem that, hopefully, will help us solve the original problem as fast as possible. Notice that everything we have described about how to prove P=NP has been seat-of-the-pants. There is no a priori reason to believe that a wetware computer comprised of human brains interconnected by all the latest technology guiding a brute-force co-processor for proof-search will necessarily outperform just plain old proof-search. That is, it might turn out that such a computer is only as fast at finding the proof that P=NP as a plain old, unguided brute-force search would have been.
So, our new, less ambitious goal would turn to solving the general problem of proof-search acceleration for decidable problems. Since the best model we have of applied computation is the human brain, we would "cheat" off the brain itself, as an existing instance of a fast problem-solving machine. We would devote our wetware computer (and its associated co-processing machinery) to solving the brain and generalizing its capabilities for problem-solving in order to derive rigorous criteria for faster proof-search. The results of such investigations might be that the human brain can be significantly improved upon by making alterations to our DNA or making other tweaks to its operation. The result is that we would likely start redesigning ourselves to make improvements that would make us structurally more suited to solving any kind of proof-search problem, so that we will eventually find the proof that P=NP, as quickly as possible.
But once we start to engage in this self-referential process of tweaking our own makeup to enhance our ability to find the proof that P=NP, we will need to simulate the long-run effects of such tweaks. Because the human organism is made up of trillions of cells and because we are extremely complex dynamical systems, the amount of simulation required to calculate the long-run consequences of major changes to our brain and physiology is enormous. In fact, in order to do it properly, we would need to simulate a small, representative ecosystem, including air, water, microbes, radiation, and so on and so forth. We would need to construct simulations capable of exhaustively simulating the earth environment, at least to the scale of a small geographical region.
And here, at last, we have come full circle to the idea of a simulation which has some purpose that is not merely a generalization of narrow human interests, such as warfare, commerce, entertainment, and so on. Obviously, no angel has announced to us that P=NP. But if it is true that P=NP, it would be rational for the human species to devote an enormous amount of its available resources to discovering the proof of this. It's like we're ants living on top of a mountain of sugar, encased in bullet-proof steel. No matter how impossible it might seem to penetrate that steel armor, we are rational to devote almost all of our available resources to discovering any possible technique that would allow us to get through to that effectively infinite energy resource.
In closing, let us return to Bostrom's original Simulation Argument. It is possible that we have always been living in a simulation and it is possible that the motivation of the Simulators is not curiosity about their origins but, rather, to find a proof that P=NP! Imagine that finding the proof that P=NP is so hard that our original cadre of wetware mathematician brains transformed themselves into something completely un-human-like: universe-spawning replicators that seed all available computational resources (including the natural environment of Earth) with viral computer code that hijacks whatever it comes in contact with and converts it into P=NP proof-searchers. It's like Gray Goo that doesn't kill you, it just makes you search for a proof that P=NP.
Bostrom's original Simulation Argument has no particular normative aspect to it. But it should be obvious that my tweak on his argument does. Specifically, if it is the case that we are in a simulation that has been spawned in the process of some kind of proof-search for the proof that P=NP, it would behoove us not to blindly thrash our available resources around according to whim or fancy but, rather, to try to perceive the template according to which the proof-search process is already unfolding and align ourselves with that template.
The reason for aligning is superrationality. While the idea of some kind of superior entity (or entities) capable of spawning a simulation on such a terrifying scale can lead to feelings of Lovecraftian horror, there is every reason to believe that a mind or minds that are searching for a proof that P=NP are basically aligned with human ethical values. Specifically, there is every reason to believe that such a mind or minds are predisposed to the principle of reciprocity -- if we approach the possible existence of such hypothetical Simulators with a charitable orientation (instead of a hostile orientation), then we can expect that they would hold themselves to have an equal obligation of charity towards us. If they were not also searching for the proof that P=NP, we might hold the reservation that we are simply being exploited as an animal resource or in some other way. But their mutual goal in finding a proof that P=NP means that they are subject to the constraint of scarcity which we also want to escape (if possible).
There is some interesting game theory that opens up if we posit the real existence of some kind of abstract agency or agencies. If we are in a simulation, then it is possible that the local rules of our observable universe are not truly universal and, in that case, it could be that the Simulators are interacting with us in ways that do not conform to our ordinary notions of the limits of agency. To cut to the chase, I specifically mean to invoke something like the pagan notion of "the gods", but without all the superstitious baggage they attributed to them. In other words, it could be the case that what our ancestors referred to as "the gods" were real, yet abstract, agents. These abstract agents could be nothing more than purely deterministic scripts coded into the simulation by the Simulators, something like the "bootstrap loader" used in digital computers to initialize and invoke the operating system software. But it is also possible that these agents are cognitive beings that are not limited by the laws of physics, as we understand them today. If this is the case, we might be able to leverage this fact to accelerate our alignment with the goal of the simulation, supposing that it is some kind of proof-search for P=NP, or a similar goal.
There is a lot of buzz surrounding quantum computation but there are also some jeremiad voices among leading quantum researchers. While we know that quantum computers are capable of solving large classes of interesting and useful problems faster than digital computers can, we do not know how cost-effective, scalable and practically useful quantum computers will be. It is not obvious that quantum computers are going to moot the constraints that I already mentioned in respect to the P vs. NP problem. Even if quantum computers turn out to be less immediately useful for applied computational problems than hoped, it is still very possible that they will play an important role in enabling the kind of extremely large-scale, detailed simulations mentioned above, simulations that will be required in order to make significant alterations to the human make-up. If the Simulators exist and if they are searching for a proof that P=NP, it is virtually certain that they are operating quantum computers.
The Universe is observationally indistinguishable from a quantum computer...
- Seth Lloyd (https://arxiv.org/pdf/1312.4455.pdf)
