It seems worth pointing out that our sense of what makes a great story is strongly influenced by the sorts of events that occur in our observable reality. It would be rather surprising if reality wasn’t narratively interesting, imo!
Our world looks like it was optimized for entertainment.
My first case against this which comes to mind is that such evidence could be weak or emerge from survivorship bias instead of an actual simulation. Suppose that you see a random string of zeros and ones, but remember only parcels of length
For example, Pichai is the CEO of Google, not of GDM. The rise of clowns like Trump could have a benign explanation, especially given the careers of Javier Milei or former actors Reagan and Zelenskyy.
Additionally, I wonder if there are far more reasons to make a cheap simulation than an expensive one like ours. For example, if an ASI decides to do research around the singularity, then research is split into the part where the AI's training data, architecture, RL intensity, eval methodology, etc. impact the AI's alignment and incident frequency, the part where the AI's training data is created and the part where lab employees think of ways to build the capable aligned AI while trying to outpace rivals and withstand pressure from governments. Having simulated governments[1] do their work could be cheapened by having them talk with husks or receive basic information around the outside world.
Finally, I would expect that carefully crafted simulations where the people struggle to notice that they are in a simulation require so much compute that many possibilities which you list are ruled out, like dreams of a higher being which isn't you. Indeed, carefully simulating the 100B human lives of 1E15 FLOP/second each lasting at least around 30 years aka 1E9 seconds would mean investing 1E33 FLOP. Simulating 1M human lives and 100B husks (which also provide, at best, a thousandfold decrease in cost, since GPT-4o-mini requires at least 100B FLOP per token versus the 100T estimate from a human brain) risks creating friction between humans and husks.
P.S. I wonder what exactly is affected by the world being a simulation as opposed to not being a simulation. For example, if I made a simulation, then I don't think that I'd make a big one or even alter the names of Chosen Ones like Musk. Instead, I would try to repurpose someone else's simulation and to check how different backgrounds of a simulated human affect the ability to generate useful ideas in AI safety (e.g. my background is physics, math and unorthodox political-economical views).
P.P.S. I didn't even notice puns like Alt(ernative to )man or PichAI. Maybe one could do experiments with, say, deliberately finding Russian puns in novels whose human author couldn't have inserted them because he wasn't supposed to know Russian, let alone make puns?
And simulated labs, but they select for being smart, making it easier to notice a potential simulation.
Epistemic Status: I think each part of this piece is worth engaging with in its own right. I haven't thought as much about whether the ideas presented here add up to "it's more likely than not we live in a simulation".
We have good reason to believe that we exist in a “simulation”[1] of some kind. “Simulation” here should be taken to mean: any kind of intentionally created seemingly-self-contained ‘world’ containing conscious actors with no knowledge of the external world. Note this is not the same as the usual conception/usage of simulation.[2]
I believe this not for any one single reason[3]; rather, different ways of thinking about this question each give some evidence, and when combined they make it look very unlikely that we don’t live in a simulation. In the words of the inimitable Linchuan Zhang, “each individual piece of evidence might be explained away, but the consilience of evidence across multiple angles and sources is in my opinion very hard to dismiss collectively.“[4] Each of the ideas or “pieces of evidence” I discuss in this piece has some kind of non-simulation explanation, just-so story, or ad hoc rationalization. (“I am firm, you are obstinate, he is a pig-headed fool”). But when taken all at once, they’re very difficult to explain sans simulation.
What are these different angles? On the theoretical side, there’s A. the strong philosophic “simulation argument” and B. the existence of deep ontological problems our minds face that suggest our ontology does not capture reality. On the "empirical" side, there’s A. a variety of evidence that suggests we live in an intentionally designed world and B. a variety of reasons to expect superintelligent beings to carry out simulations that look like our world.
The classic simulation argument
Let’s start with the "theoretical" reasons. The biggest thing here is the classic simulation argument, which is worth revisiting as it’s a very powerful argument! The full argument is here; the conclusion it argues for is this:
I would modify this to remove the “computer” in “computer simulation” and add that ‘simulation’ here should be taken to mean “any kind of seemingly-self-contained ‘world’ containing conscious actors with no knowledge of the external world.”[5]
Deep ontological problems
Let’s now talk about the deep ontological problems we face. We ‘know’ that there exist true statements that we cannot know and that there are facts we cannot express in language. We know this from information theory and from Godel’s incompleteness theorems. Furthermore, there are very important things we cannot express in language, and I believe[6] there are very important things we cannot even conceptualize within our basic ontology, an ontology to which concepts like “truth” and “exist” are essential. Taoism and Zen make these points rather compellingly, though there’s no strict argument I can point to here; indeed, part of the point is that there is no way to express this stuff in language.
In particular, I think that ‘consciousness’ is an extremely important thing that not only cannot be properly referenced or described by language but cannot even be studied or accessed from a third-person/objective/scientific perspective.
There exist many “big” questions that we just can’t seem to answer. Often a particular theory can answer/reject one/some of them, but cannot answer others and is inconsistent with theories that do answer others. Many of these questions could be said to belong to the domain of physics (or do they belong to the domain of philosophy?). What I’m talking about here are questions like:[7]
Theories that (purportedly) answer some of these questions: Anthropic principle. Mathematical universe hypothesis. Christianity. Illusionism.
For further reading, see niplav's Ontological Cluelessness.
Intentional design (this sure would make a great story!)
Our world looks like it was optimized for entertainment. Consider:
Note that superentertaining characters, nominative determinism, and crazy names are all typical features of narrative art/entertainment – think of literature and film and limericks and so on. Our world looks like a designed story. It looks narratively optimized.
Besides being entertaining, our world is one that would be of particular interest to a wide swath of simulators. We live in the most interesting time, that being the takeoff to the singularity. This seems like a really important time to study. It’s not hard to imagine how simulations of civilization building ASI would be useful for historians, anthropologists, economists, theologians, cognitive scientists, philosophers, story writers, video game designers, etc. etc. etc.
The names of the CEOs involved in AI development – Altman (alternative to man), Pichai (Pitch AI), Amodei (also Amodei means lover of God) – are truly egregious, especially when one considers that the era of ASI development and the period immediately before the singularity are times of particular interest for multiple kinds of simulators. My collection of strangely related phenomena is consilience; yours is synchronicity; his is coincidence.
To sum it up: if I were a simulator, my simulation might look like the exact world we live in! Seriously, can you think of a more interesting, informative, entertaining world to simulate?
What kind of simulation?
Simulation as training module, from Monster's Inc
Allow me to (non-exhaustively) list some kinds of simulation that this may be. Note that our ontologies might not apply to whatever is actually going on in the ‘bigger’, external-to-us reality. ‘Art piece’ or ‘‘God’ might not be good concepts in this bigger reality, but they’re at least pointing at the things that we lack the proper concepts for.[12]
See Alan Watt's "The Dream of Life".
“[L]et’s suppose that you were able every night to dream any dream you wanted to dream and that you could, for example, have the power within one night to dream 75 years of time or any length of time you wanted to have. And you would, naturally, as you began on this adventure of dreams, you would fulfill all your wishes. You would have every kind of pleasure during your sleep. And after several nights of 75 years of total pleasure each, you would say ‘Well, that was pretty great. But now let’s have a surprise. Let’s have a dream which isn’t under control, where something is gonna happen to me that I don’t know what it’s gonna be’. And you would dig that and would come out of that and you would say ‘Wow that was a close shave, wasn’t it?’.
Then you would get more and more adventurous and you would make further- and further out gambles to what you would dream. And finally, you would dream where you are now. You would dream the dream of living the life that you are actually living today.”
That's all for now.
As mentioned in the beginning, I think each part of this piece is worth engaging with in its own right. For example, I think it's worth asking "is our world super-entertaining" separately from the question of "if our world is super-entertaining, is that evidence that we live in a simulation?". I welcome comments on all parts of this.
Appendix: simulations vs embeddings, intelligent design vs intentional design
I said earlier that by ‘simulation’ I mean any kind of intentionally created seemingly self-contained ‘world’ acontaining conscious actors with no knowledge of the external world. If we lose the intentionally-created clause, we can think about non-intentionally-created ‘worlds’ as well. I call this broader class of things “embeddings”, as in: our world is embedded in a much greater reality that we don’t have access to. Embeddings are a superset of simulations. (I’m accepting proposals for a word to refer to only the non-intentionally-created worlds.) An iconic conception of this kind of situation is that of Flatland. (Who else watched the super cool movie in school?). I do think it’s reasonably likely that we exist in a non-intentional embedding, though I don’t have any more to say about that right now.
Instead, I bring up embeddings to point out that some of the evidence and all of the possibilities I’ve described for simulation does not apply to non-intentional embeddings. We have reason to believe that we live in an intentionally-created universe. Our world appears to be intentionally designed. (I wonder if this actually the better framing for the whole idea here, rather than the ‘simulation’ framing I used.) Sorry, Christian fundamentalists, you got it wrong. It’s not argument from intelligent design, it’s argument from intentional design.
Any ideas for a better word here than ‘simulation’?
Indeed, I think that a narrow conception of simulation has prevented people from seeing how various simulation arguments fit together.
Some notes about my own epistemology here. My belief that we exist in a simulation has gradually increased from ~low to ~high over the past many years, although I think this is the kind of proposition that one can’t properly put a probability on. I am not a “woo” person by nature or upbringing. I am quite sane, mentally sound, and non-schizo.
This is from Linch’s piece on the Pope’s AI Encylical being AI generated. I couldn’t say it better myself (inimitable!), so I quoted him.
Note that this is not how Bostrom intends or presents the argument; still, the argument he gives basically applies as written to this broader class of simulations. And note that it is a broader conception of simulation, which means that it’s weaker than the one Bostrom is making. Furthermore I think that this conception of simulation actually addresses the biggest weaknesses in his argument: A. what if computer simulations can’t give rise to consciousness B. what if intelligent civilization intentionally avoids computer simulations that give rise to consciousness C. there may be infinities or at least recursive, exponentially generators of beings that aren’t computer simulations (such as fiction of various kinds); why is the measure of computer simulated beings larger than these simulated-but-not-by-computer beings?
Although I wouldn’t say I “know” it. I don’t have a ‘proof’ for it. It may well be one of these true unprovable things I’m talking about!
A note on the theoretical vs empirical distinction: Some of these "ontological problems" straddle the border into being ‘empirical’/contingent problems, such as the fine-tuned universe, the question of why don’t we see other intelligences in the universe, and the question of the existence and creation of the universe. The theoretical vs empirical distinction is only confusing here.
Answer: We can’t, a system cannot prove its own consistency, c.f. Godel’s second incompleteness theorem.
Elon Musk, for what it's worth, has expressed views similar to those being made in this section: "The most entertaining outcome, especially if ironic, is most likely." Based on his behavior, I think he really believes this to some degree? The better-stated version of this proposition is "All else being equal, more entertaining outcomes are more likely, and the effect size here is considerable."
Donald famously trumped Hillary in the 2016 election.
Ask your friendly neighborhood AI for more samples.
As said in Ontological Cluelessness, "Humans may be in a state of total confusion as to the fundamental makeup of the cosmos and its rules, to the point where even extremely basic concepts would need to be revised for accurate understanding."