A common general pattern in games is to not process that which is not seen or known not to have any interaction - e.g there's no point in making NPCs/enemies move or think in the last room of a level (or a far away planet) when the player is at the beginning of it, there's no point in rendering what's outside the frustum, etc... which kind of echoes the deep philosophical questions of "do unobserved things exist?" or "if a tree falls in a forest and no one is there to hear it, did it fall?", or quantum wave collapsing/Schrödinger's cat.
Is there any game engine that does such general lazy evaluation based on the observation effect or are they all using ad-hoc hardcoded laziness rules?
Many side scrolling platformers have ad-hoc rules based on where the virtual camera is looking on the 2d level that are pretty close to what a systematic rule would be.
Consider a hypothetical game like No Man's Sky. A tree falling on a neighbouring planet has zero impact on the current one the player is standing on.
> something where if if not simulated while nobody was around, the simulation would need to catch up on if an observer ever came near.
That's the wave collapsing part: probabilities become reality.
> this eventually amounts to processing work that is equal or even worse compared to just simulating it in full detail in the first place.
Collapsing treefall probabilities for the small area of a planet the player is due to observe when coming near the area is orders of magnitude less than simulating every single treefall on every planet in real time.
This principle could be equally applied to NPCs, e.g a NPC having a probability to be at this or that place, taking into account a lightcone of causality so that they don't warp from one place to the other.
E.g. Elite would bring ships into existence only when you entered a planetary system. If you then shot at police, it'd affect your reputation and impact others reaction to you even after ships have blipped out of existence. But the ships themselves retained no state.
Minecraft, on the other hand, maintains state of objects "too well": Once you visited an area, the landscape goes static once you leave: While mobs disappear and are reinstantiated, nothing will grow or decay while you're away.
If you want to make a realistic lazy simulation and you also want to minimise computation and storage, you could do this:
* For any entity, you assign a function that computes its change over time, including being able to give a result that is effectively "reconverged to base-state of the map at time t". * When a user re-enters a map location, you 1) generate the map as per that time, which might include whatever effects you want to cause change over time, 2) you apply the functions for any user-affected objects that were present, and let them be affected by map changes since last time-step, 3) you purge any user-affected changes that will have re-converged to the base state of the map.
E.g. clear an area of trees? If you come back soon, it'll still be clear; if you come back much later, it will have reverted to the freshly generated state of the biome affected by a time factor so it's not identical.
Leave your farm for a while? A bit later it'll be ready to harvest. Too long, and it'll have fallen into disrepair.
Build a house? Leave long enough, and there will be damage to it, and cobwebs etc. Leave for really long and it might be ruins when you return.
Dug out a tunnel? Wait too long, and things might have fallen and started blocking them.
Put another way: A good enough simulation could potentially short-circuit a lot of the need to retain state and simulate what has happened by applying pure function of time decay. You don't need to simulate the process of a house decaying, just the outcome at time t unless it's being observed at a given time.
Incidentally, when people don't know/remember the details of something we've done, we're great at just making up the details - we know this from split brain experiments - so even if you instantiate simulations of people in this kind of scenario you could likely work your way around inconsistencies just by decay memories as well and letting people fill in the gaps in a way consistent with the effects they see.
I fully agree with you that it won't always be worth it vs. just simulating the full detail in the first place, but I also think you can at least get far closer to realistic even with aggressively pruning what you actually fully simulate. To what extent you can get close enough is another question, and something that's fascinating to think about.
I think you are taking us far too serious, we are not that important...
In other words, maybe you're the only subject. Or maybe I am. Or maybe all that exists are isolated "snapshots" and that we only assume time passes because we process a single momentary "image" of memories that includes the sensation of it passing.
The problem, of course, is that we can not possible ever know. And so we mostly discard this notion because absent someone pulling us through a door and letting us observe it from the outside, we're stuck inside a room and trying to infer the outside shape of the building without having any idea how much of the interior we have access to.
Like some programs escape a virtual machine by escalating priviledges.
I read that the computer game Eve online (never played it) slows down when there are big battles with many players. Sounds a lot like black holes. Time slows there down too - too many calculations? Planck time is the tick rate of the universe? Can pauses happen?
As I said in a different comment, even if the stars in the Milky Way suddenly rearranged themselves to spell out »You are in a simulation.« on the night sky, you could not tell for sure whether this is a message from the simulators in the outer universe or whether this is just some weird behavior arising from the laws of physics in a real universe.
This of course proves nothing, and I'm absolutely not suggesting it's evidence of anything. But it's fascinating when possible design tradeoffs for a simulation might fairly neatly line up with how things actually work in our universe. (At least from a science fiction POV; I have a couple of story ideas I intend to write that makes use of that notion)
I do share your thought that it's harder and harder to dismiss some variant of the simulation argument - especially as the original one is far too constrained (which made sense in as much as you want to cut away as much as possible that might make people casually dismiss it)
Imagine we have a really close look at space and discover that it is made out of small voxels. Simulated universe! No! Who said that real universes are not made out of small voxels? It could be the other way around, the real universe could be made out of voxels but a voxel-based simulation does not fit into memory, so they assigned a real position vector to every object in order to approximate the real voxel universe. In that case it would be wrong to conclude that our universe is real because it is continuous and a flawed analogy with our own computer simulations suggests that simulations must be discretized in some way. [1]
So no matter what you observe in our universe, you have no idea if it is indicative of a real or simulated universe. Maybe if the stars in the Milky Way would suddenly rearrange themselves and spell out »You are in a simulation.« on the night sky, that could be a good hint that we are indeed in a simulation. But even then, maybe that is just how real universes behave, spelling out messages on the night sky from time to time might just be part of the laws of physics. Admittedly, if that actually ever happened, I would also consider the simulation hypothesis the better explanation.
[1] This is just an illustrative example, whether this is realistic or even possible is besides the point.
The best we could hope for would be someone popping their metaphorical head in and telling us there's an outside and then show us, but that just gives rise to the question of whether that "outer universe" is simulated or not, so infinite regress here we come.
Ultimately we just have to accept that we can never conclusively prove whether we're in the "outermost layer", and we can only prove we're not if we somehow "escape" and can look in on our own universe from the outside and for what we know we're in an unexploitable simulation, so the lack of evidence of either certainly does not prove the other and we're left wildly speculating and it's not very useful.
But that doesn't make the mechanisms less interesting to explore, and for me it's mostly a series of fun thought experiments, and your voxel example to me is a good illustration of what an interesting rabbit hole it is even if it doesn't lead to anything practical.
For me it was an interesting one because the simulation argument is really a rehashed variant of Berkeley's philosophical idealism, and the same conflict with materialism exists for the simulation argument as with respect to Berkeley: We can speculate about it, but since we can't prove it, day to day we need to act on the assumption we live in a material world. If someone pops their head in to inform us we're in a simulation, then sure, then we can reasses.
Until then it's an entertaining diversion.
EDIT: I'll add one caveat, though: You're right in as much as talking about what we can say about a simulation by looking at our universe, but the core of the original simulation argument is much more pared back. It's an argument that if we can run simulations and eventually do, then statistically we are more likely to be in a simulation. Or to pare it back further: If there are 3 or more "realities" and 2 of them are simulation, then we are more likely to be in one of the simulated ones than the single "real" world. What that boils down to, then, is how realistic you think it is that someone will simulate this time period, and we can assert some limited things about that if we get to the point of running our own realistic simulations. But until/unless we get there, we can't really assess the odds, and even if we can show that such simulations are common, you're of course still right that it won't prove whether we are in one or not.
To address the actual trilemma, I think the solution is that there are or will be no simulations. Generally the most compact representation of a thing is the thing itself. If you want to simulate something faithfully, then the simulator will need more resources than what you simulate. You are not going to build an electron simulator that needs less than one electron to simulate one electron. You are not going to simulate a computer with a gigabyte of memory on a computer with a gigabyte of memory. That might work under special conditions, when the simulated memory is mostly zeros and you can compress it, but it will not work in general.
You can simulate the weather on earth well enough for the whether forecast to be useful if you throw away enough details, but you are not going to simulate it in full detail with a computer much smaller than earth's atmosphere. Simulating a cubic meter of universe faithfully will generally require at least a cubic meter of computer. I do not see us building galaxy-size computers. The other option would be to go the way of the weather forecast, make the simulation as simple as possible, simulate entire humans but not cells or atoms to make the problem tractable. That seems much more plausible.
In both cases it seems unlikely to me that you could deeply nest simulations. If you have a faithful simulation and simulating a cubic meter of universe requires two cubic meters of computer, then you could at best simulate a universe half the size of the real universe if you turned the entire universe into a computer. The people in that simulated universe could at best build a simulation half the size of their universe and so on. The simulated volume would exponentially approach zero with increasing nesting level.
If you go the other way, if you simulate entire humans and not cells or atoms, then it is not even clear how the simulated people would go about creating their own simulator as simulating arbitrary machines is not part of the simulation. So you are forced to move towards a more faithful simulation mechanism if you want to enable nested simulations which, as said, will probably shrink the simulation volume rather quickly.
There's no reason to believe this, and (I would argue that) any civilization creating ancestor simulations has, in fact, likely reached a post-scarcity economy where it's entirely possible that the "programmers" are doing it for the love of programming, the love of simulation, the love of understanding, etc. Or that they are trying a very large number of simulations with different setups at one time. And the Higgs boson fits perfectly well into any or all of that.
My use of programming terms to describe this is both a nod to 'gosh this game engine has great memory management and algorithmic efficiency' but also me thinking about the relationship between the fundamentals of programming and the fundamental nature of information.
When programming we understand that time is simply a matter of cycles of information processing, and that space in a rendered 3d game is not real, but a representation of relationships between data structures, and it's interesting to think about what computers - machines built in this universe from the stuff of this universe - are revealing about the nature of this universe.
I know that it is easy to mock computer metaphors as the next iteration of the universe is a clockwork, but the people who thought the universe was a clockwork were not wrong, they were taking a small step forward. They had a new metaphor which gave them additional insight that they had not had before - a model built in this universe from the stuff of this universe that shed new light on the nature of the stuff of the universe. This is what a computer is, it is an object built of the stuff of this universe that has given us new insight in the nature of the stuff of this universe.