Unless you, dear reader, are a web-scraping software bot quietly pulling this text into a data-hungry LLM, you’re probably a human. And though you’ve likely never seen me in person, you have good reason to believe I’m human, too. But we don’t know one another exist, for sure. In fact, we have no definitive proof that anything is conscious beyond ourselves—a clunkier analogue to René Descartes’ famous dictum: Cogito, ergo sum — “I think, therefore I am.”
Forget one another—what if none of it’s real? Not our coworkers, the trees, the constellations—not even ourselves? That’s a question Gizmodo recently posed to a slew of experts across the fields of computer science, philosophy, social psychology, and neuroscience.
For this Giz Asks, we asked these experts a straightforward question: How do we know we’re not living in a computer simulation? Their answers—varied and nuanced—offer thought-provoking insights into a question much older than computers themselves, yet one that has remained compelling, if not obsessive, for thinkers through the centuries. With so much technological innovation happening right now, it’s only natural to consider the age-old question with greater scrutiny. We’re creating technology so uncanny, it seems like it will climb out of the valley and dust itself off. Strange, surreal times—the perfect time, I would suggest, to ask how we know what is real at all. Some responses are lightly edited for clarity.
For all we know we are, in fact, in a vast computer simulation. Here’s a version of the simulation argument that I am fond of. Suppose that the commonly held view in astrobiology that there is life on other worlds is correct. Further suppose that at least some of these life forms evolve to become technologically sophisticated, developing artificial intelligence to such a degree that they run sophisticated computer simulations. Now, if there are only two civilizations out there that do this, the chances are 2 to 1 that you are in a computer simulation now!
Of course, there are many issues with this version of the argument, and those like it, well worth debating. Can consciousness be something that is simulated by a computer? Is it possible to build highly sophisticated AIs, such as quantum computers capable of simulating Earth? Do civilizations survive long enough to develop such technologies, or do they hit some sort of wall or ‘great filter’ in their development. Indeed, perhaps life does not even exist on other planets. (Of course perhaps we are being simulated by future Earthlings, as in the Matrix.) The jury is out, and the issue is a fascinating issue indeed!
The hypothesis that we live in a simulation seems provable: it could be the discovery of a flaw in the simulation, such as a distant region of the universe that cannot be zoomed in on, where a telescope would not be able to obtain a clear image. Of course, an even more advanced simulation could roll back time, erase the flaw, and then restart the simulation. This suggests that simulations may vary in type and quality.
Nick Bostrom’s Simulation Argument is a brilliant piece of philosophy, suggesting that the initial probability we assign to the idea that we are currently living in a simulation should be revised upward. However, the argument does not convince me, mainly because simulations can be of very different natures. In particular, the Simulation Argument is based on a reference class of simulations created by post-humans, where the simulated beings are nearly indistinguishable from real humans but are unaware of their simulated status. However, it is also conceivable that simulations just as sophisticated could be created, where the simulated beings are aware of their nature. Depending on which of these reference classes is chosen, the conclusion of the argument changes significantly. If the majority of simulations involve beings aware of their simulated nature, it eliminates the disturbing conclusion that we ourselves are simulations, since we are not aware of being simulated. The choice of reference class is therefore crucial. If the argument is based on simulated humans who are unaware of their simulated nature, it leads to the unsettling conclusion that we are likely living in a simulation. Conversely, if the argument focuses on humans who are aware they are being simulated, the conclusion is far more reassuring: we are not such simulations. There is no a priori justification, however, for favoring the choice of humans unaware of their simulated nature over those who are aware of it. Furthermore, one could even argue that simulated humans who are unaware of their nature might be prohibited in a post-human civilization for ethical reasons.
One might wonder why the question whether we are in a computer simulation should even be taken seriously. The answer is the simulation argument (which I published back in 2001). The simulation argument shows that at least one of the following propositions is true: (1) the human species is very likely to go extinct before reaching technological maturity; (2) any technologically mature 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. To the extent that may think that (1) and (2) are unlikely, therefore, we need to assign substantial probability mass to (3). In any case, the simulation argument reveals an interesting coherence constraint about what one can coherently believe about the future and about our place in the world.
The closer we get to reaching the point where human civilization becomes capable of creating ancestors simulations—with conscious minds in it that has experiences of the type that we have—and provided we still retain some inclination to use at least some non-negligible fraction of our resources for that end, the less likely (1) and (2) would appear; and hence the more likely that we ourselves are living in a computer simulation. I think the progress in computer game graphics and in artificial intelligence that we’ve seen over the past couple of decades has made it easier for people to see the simulation hypothesis a plausible view about the nature of existence.
Using evidence-based reasoning, we can’t conclusively determine whether we are living in a simulation or not. A sufficiently sophisticated simulation can be totally indistinguishable from reality, and any bugs could always be deleted and then the simulation restarted from before their emergence. There’s also no fundamental barrier to simulating our experiences, especially since one only has to simulate an individual’s perceptions of reality, not the reality itself. Think about how a computer game doesn’t render the whole universe all at once, just the section your located at.
A common argument used to claim that we are very likely in a simulation is that there should be far more simulated realities than the singular base reality, thus it’s pretty unlikely we’d happen to be in the base reality. However, this argument is predicated upon the assumption that simulating realities is possible in the first place. Although it’s true that we don’t know of any fundamental objections to doing so, we also don’t know that it is actually possible either! I would tend to give these two possibilities 50:50 odds in the absence of any evidence either way. Another subtly is that if simulated realities can themselves birth simulated realities, then there will be an enormous hierarchy of realities and each one will necessarily have less computational capacity than the ones above it. Eventually one might reason that there will be a lowest level, a sewer of reality, where occupants are simply incapable of building a computer sophisticated enough to ever produce their own convincing simulations. In fact, there should far, far more of these sewer realities than all the others put together and thus we should conclude that’s where we live. This forms somewhat of a contradiction though. We have concluded that we likely live in a reality where simulations are impossible, yet our entire argument is predicated upon the assumption that simulations are possible. These kinds of contradictions deserve more attention and perhaps will eventually help us make sense of this persistent question.
Nobody knows whether or not we live in a computer simulation. I put the probability around 20%. I think we might be living in a simulation because I can’t rule out the following premises (one or both may be false, of course, but that is not obvious to me). First, simulation technologies will continue to improve, and they will eventually be capable of simulating people with experiences like ours. Second, there would be an irresistible temptation to use such simulations to study things about the past. (For example, if you wanted to study the connection between intelligence and violence, you could run many simulations of the history of humanity in which you vary the average IQ and observe trends in the prevalence of warfare.) If these premises are true, then we should conclude that eventually people will run millions of simulations concerning humanity’s history. But if so, how do we know we’re not already living in one of those simulated histories? After all, there would exist millions of simulated realities and only one unsimulated reality. So the odds would be good that we are in a simulated reality.
This way of thinking in philosophy started with Nick Bostrom’s 2003 paper “Are You Living in a Computer Simulation?” Eventually, physicists got on board with the idea, and in 2014, some suggested that we could scientifically test the simulation hypothesis by taking a close look at the movements of cosmic rays. In 2017, more potential tests were proposed. In my 2019 paper and New York Times piece, I argued that the move to scientific experiments in studying the simulation hypothesis is dangerous because such experiments risk causing the shutdown of our simulation. Essentially, the experiments are attempts to reveal something that would happen only in a simulation—for example, a “glitch” in the movement of a cosmic ray. But whoever might have created our simulation probably cares that it is accurate. So, if we manage to reveal some glitches, our simulators might simply end the simulation and develop a less glitchy one (as is the current practice in computer programming). Philosophical theorizing about the simulation hypothesis doesn’t have this risk, because philosophers draw conclusions based on logical inference and the nature of probability, and not by looking for glitches.
The idea that we “live in a computer simulation” is too vague to be scientific. It only becomes scientific once you specify what you mean by computer simulation.
If you say, for example, that it means that our observations can be calculated, then it just means that the laws of nature are mathematical, which they are. But this is usually not what people mean when they talk about a “computer simulation”. They tend to mean that there is an algorithm that can be executed on a computer that works similar to the computers we use ourselves and that will reproduce reality as we observe it.
There is no known algorithm which does that. If we had such an algorithm it would be a theory of everything. People who claim that it can be done tend to underestimate the problem, usually because they have no idea how physics works in the first place. Maybe they should have a look at how difficult it is to even make a weather forecast despite the fact that we do have the equations. And let’s not get started about quantum physics or quantum gravity. Generally, if someone claims we live in a computer simulation I think it’s up to them to please tell us what the algorithm is and not just claim that it exists.
Another thing that people sometimes mean when they refer to the simulation hypothesis is that not only is the universe a simulation but it’s a simulation created by someone or some thing, so a god basically. I do not know how one could find evidence for any sort of creator behind an algorithm that explains our observation, even if we had such an algorithm, and I don’t think this is a scientific question.
There is no definitive way to prove that we aren’t in a simulation. The simulation could be so perfect that we may not be able to detect it. However, that doesn’t mean that we can’t find some evidence that we may be living in a simulation, and various groups have been working on this. The way to understand the simulation hypothesis is to break it down into two constituent parts: 1) that the universe consists of information, and 2) that the information gets rendered for us, like in a video game.
That the universe consists of information isn’t hotly debated as it might have been when John Wheeler came out with his famous statement of “it from bit”—i.e. that when you look for this thing called a particle, physicists can’t seem to find it—the only thing they find is a set of properties which are like answers to a series of yes/no questions—i.e. bits of information. For example, Tom Campbell’s group at ColyPoly is trying to run some experiments to show that that a conscious observer is needed to collapse the quantum probability wave through a series of double-slit eraser experiments; Melvin Vopson in the UK is working on the second law of infodynamics which shows that information entropy declines over time in certain biological systems. Stephen Wolfram is working on models which derive physical laws from types of cellular automata and hypergraphs—i.e. computation. Some groups have found that there is geometric shape in the form of lattices which resemble pixels.
However, perhaps the more important question is what I call the NPC (Non-player characters) vs. RPG (role-playing game) flavors of the simulation hypothesis. In the first version, everyone is just AI, or NPCs; in the second we exist outside the game and are players, with avatars or characters inside the simulation, just like with today’s MMORPG video games (massively multiplayer online role playing games). We are rapidly approaching the simulation point, a kind of technological singularity where we can create realistic simulations which are indistinguishable from reality, with AI characters that are indistinguishable from biological characters. If any civilization reaches that point, to use an argument put forward by a philosopher, Nick Bostrom of Oxford, then there are likely to be many simulated worlds (billions) with many simulated beings (trillions), then we are more likely to be simulated beings inside a simulated world than a physical being in a physical world. It was this argument that led Elon Musk to state in 2016 that the chances that we are in base reality (i..e not in a simulation) is one in billions. If we are able to reach the simulation point in the next few decades then it has probably already been reached by a more advanced civilization, and we are inside their simulation—unable to tell the difference.
