So, in Part One of this series on Simulation Theory, we more or less got our heads around the Simulation Hypothesis and the Simulation Argument.
If you’ve not read Part One I’d suggest checking it out before going any further, unless you’re already comfortably clued-up on Nick Bostrom’s Simulation Argument.
Let’s quickly recap…
In a nutshell we agreed (for the sake of understanding the argument, at least) that it’s possible for a technologically advanced civilisation to create a simulated universe, including conscious beings within it.
For the purposes of the argument, this is agreed on the basis that we have no reason to believe otherwise. It is logically sound to say that one day (could be in 100 years, could be in one million years) our human civilisation could get to that point, because technology will continue to advance. And we can say that the same is true for any other technologically advanced alien civilisations ‘out there’.
This being the case, we can then consider the hypothesis (originally put forward by University of Oxford philosopher Nick Bostrom) that we are actually living in a simulation. By putting forward this hypothesis, we’re not saying that we necessarily believe it to be true.
In order to explore that hypothesis, we look to Nick Bostrom’s Simulation Argument which, surprisingly, shows that one of the following statements almost certainly must be true, if we accept the assumption that it is possible to create such a simulation:
The human civilisation (and any other technologically advanced civilisation/s) does not reach the stage of developing the technology necessary to create a simulation of the universe (even though it is possible). This could be because they all go extinct before they get there.
Of the advanced civilisations that create technology capable of simulating a universe, all or virtually all choose not to use it to create ancestral simulations. Perhaps they decide that it is immoral to create conscious beings capable of suffering, or decide it’s too risky.
We are very likely to be living in a simulation.
To put it another way, we can consider all three statements together and say that these three possibilities lead to the conclusion that, unless we are now living in a simulation, our descendants and other technologically advanced civilisations are unlikely to ever create ancestral simulations.
That is Nick Bostrom’s trilemma, also known as the Simulation Argument.
Here it is in his own words:
Now that we’re clear on all that, in Part Two of this series on Simulation Theory we’re going to explore which of these three statements may be more or less likely than the others, to see if that helps us to work out how likely it is that we’re living in a simulation…
How Likely is It That We Are Living in a Simulation?
In order to try to determine how likely it is that we’re living in a simulation, we have to work out whether we can assign more, or less, probability to any of the three possibilities in the argument.
If you’re after a quick answer, I will just tell you now that it’s not possible to definitively assign any meaningful amounts to the probabilities at this point in time, because everything is an unknown.
So, if you’re satisfied with the answer ‘we just don’t know’, you could skip to the end of this rather long post for an interesting final thought!
If, like me, you can’t bring yourself to just leave your philosophical itches unscratched, read on!…
Possibility One in The Simulation Argument
The first possibility from the Simulation Argument is that the human civilisation (and any other technologically advanced civilisation/s) does not reach the stage of developing the technology necessary to create a simulation of the universe (even though it is possible). This could be because all relevant civilisations go extinct before they get there.
To be completely accurate, this is how Nick Bostrom words this possibility:
‘The fraction of human-level civilisations that reach a posthuman stage (that is, one capable of running high-fidelity ancestor simulations) is very close to zero.’
The main proposition within this possibility is that there is some kind of ‘great filter’ whereby the technology necessary to develop a simulation of the universe will, in itself, destroy the civilisation by default once it is created.
It’s also possible that the technology is so difficult to create that it takes a ridiculously long time for a civilisation to get to that point. Should that be the case, the odds that the civilisation becomes extinct through other means increase.
Take our current human civilisation for example (for this point, assume that we’re not a simulation!). We could be wiped out by a huge solar flare.
In the absence of any specific information to the contrary, there is a greater chance of this happening at some point (any point) during the next 100 000 years than there is of it happening tomorrow.
So, if an advanced civilisation can’t get to the point of creating the technology within a couple of million years, for example, the odds that the civilisation gets wiped out by some other means during that time increase.
However, we can look to our rapid advancement in technology over a short period in recent history to suggest that this scenario looks unlikely.
We seem to be close to AGI (Artificial General Intelligence). My own view is that we’re probably extremely close, but for now we can think of ‘close’ as meaning within the next hundred or so years.
Once we get to AGI, it’s then probably not a huge leap to get to ASI (Artificial Superintelligence).
Superintelligence would far surpass the greatest of human minds in all aspects of intelligence.
It could also, therefore, find ways to make itself even more intelligent. In doing so repeatedly, it could quickly get to a level of intelligence that we wouldn’t even be able to understand, and could achieve things that we had thought impossible.
The fact that we can say that this scenario is quite likely to play out within the next hundred years, means that we should consider the possibility that it takes advanced civilisations a ridiculously long time to develop simulation-level technology to be reduced.
In other words, if we start out from a standpoint of attributing equal probability to each of the three possibilities, the fact that we appear to be heading (quite quickly, relatively speaking) in the direction of simulation-level technology should lead us to reduce the likelihood assigned to this possibility – because we have less time available (before reaching the necessary level of technology) to go extinct.
However, this does nothing to address the possibility that the technology itself could be the cause of our demise.
There are many gifted scientists, philosophers and computer scientists (and normal people like me!) who are currently very worried about the possibility of AGI and subsequent ASI presenting an existential threat to humanity.
So it could be that the closer we move towards those levels of technology, the closer we move towards extinction.
With that said, there’s nothing in particular to suggest that an Artificial Superintelligence couldn’t go on to create an ancestral simulation by itself, after it got rid of all the humans.
In summary, we’ve gone round in circles and come back to a place where we are really no further forward in determining whether Possibility One is any more or less likely than Possibilities Two and Three.
All we can really say at this point is that, should we manage to achieve the technology needed to create a simulation of the universe, whether through ASI, quantum computing or other means, we will then be able to discount this possibility.
If that happened, it would greatly increase the odds that we are living in a simulation.
Until then, we’re no further forward… so let’s move on to Possibility Two.
Possibility Two in The Simulation Argument
The second possibility from the Simulation Argument is that, of the advanced civilisations that create technology capable of simulating a universe, all or virtually all choose not to use it to create ancestral simulations.
Perhaps they decide that it’s immoral to create conscious beings capable of suffering, or that it’s too risky.
This is the specific wording from Nick Bostrom’s original argument:
‘The fraction of posthuman civilisations that are interested in running simulations of their evolutionary history, or variations thereof, is very close to zero.’
So this possibility is proposing that our human descendants and/or other technologically advanced civilisations do get so far as being capable of running simulations of the universe, but that they all (or virtually all) choose not to do so.
To me, based on observations of our own human nature, this seems very unlikely. If we can do something, we generally do do it, given time.
Not only that, this possibility proposes that not only do our human descendants decide not to run the simulation, but that all other technologically advanced civilisations (assuming there are some, which is extremely likely given our current understanding of the infinite nature of the universe) also decide not to run theirs.
I would argue that this fact alone – that it would be necessary for there to be a convergence of the decision across all technologically advanced civilisations – should suggest that we consider assigning a lower probability to this possibility.
However, we also have to bear in mind that this kind of technological achievement would likely be made by a posthuman civilisation with a level of intelligence far beyond ours.
That could mean that we can’t claim to have any meaningful insight into whether or not they’d choose to run a simulation, because their thought process would be so far beyond anything we can understand. We would be in the position that dogs (or maybe even ants!) currently find themselves in when they try to work out what the humans are up to!
A subsection of this part of the argument is the suggestion that, even if our descendants/other advanced civilisations do decide to run a simulation, there’s no reason to suggest that they’d be likely to run an ancestral one.
Neil deGrasse Tyson, for example, has previously said that he thinks it’s unlikely because the percentage of films we make today that are set in the past is very small compared to the amount we make that are based in modern times or in the future.
Personally I don’t think that this is a strong argument.
We do still make a lot of films that are based in the past, and even if we didn’t we shouldn’t assume that advanced posthuman civilisations would take the same decisions that we do in our current time.
Plus it implies that they’d be running the simulations simply for entertainment – we don’t know if that would be their motivation.
So once again we find ourselves no further forward!
I am inclined to suggest that this possibility seems less likely due to the convergence aspect, but that’s impossible to quantify and is just my own opinion.
Let’s hope that examining Possibility Three can lead us in some kind of meaningful direction, because it’s not looking good so far!…
Possibility Three in The Simulation Argument
The third possibility from the Simulation Argument is that we are very likely to be living in a simulation.
Or, in Nick Bostrom’s words:
‘The fraction of all people with our kind of experiences that are living in a simulation is very close to one.’
If we were to discount the other two possibilities, we would be left with no choice but to draw the surprising conclusion that we are actually almost certainly living in a simulation.
But, just how likely is this?
We can’t really argue it out as we did in the previous two possibilities because, in a sense, this is just ‘what’s left’ after you discount the other arguments.
So we could immediately just say that our conclusion to the question ‘are we in a simulation?’ has to be ‘we don’t know’ and leave it at that.
To be fair, that is the most correct conclusion to draw.
But, where’s the fun in that?
So let’s see if we could find a way to come up with a better answer.
Elon Musk has publicly stated that he thinks it is billions of times more likely that we are living in a simulation than that we’re not.
But what Elon Musk appears to have done is arbitrarily strike out Possibilities One and Two, along with the possibility that it might actually not be possible to create this kind of simulation.
I tend to be on board with him on the assumption that it must be possible, but no-one can say that it’s a given!
And he may indeed consider Possibilities One and Two to be exceptionally unlikely, but he doesn’t know that to be a fact, so the billions to one thing is just his opinion and nothing more.
What we actually need to do is look at Possibility Three more closely, while still acknowledging the clear existence of Possibilities One and Two, and the possibility that it’s just not possible!
To do that, we can now group all three of those things together – because they all effectively say ‘not in a simulation’ – and weigh them up against Possibility Three which says ‘in a simulation’.
We now have two hypotheses to work with:
Hypothesis One: The Simulation Hypothesis in which a technologically advanced civilisation/s creates a simulation/s.
Hypothesis Two: There is no simulation (either because it is not possible, no civilisation gets there, or no civilisation chooses to run the simulation). We live in base reality.
We can now attempt to weigh those two simple possibilities against each other – Simulation vs No Simulation.
In doing this though, we should keep it at the forefront of our minds that we’re just doing the best we can with the information we have. And that information is very incomplete!
Is the Simulation even possible?
First of all, let’s briefly address the way in which some people attempt to shut down this whole debate by saying that creating a simulation of the universe along with conscious beings in it is just not possible.
One thing I notice is that those who say this often present this as a fact, as opposed to their opinion, which is interesting.
Here’s one of my favourite quotes, which I mentioned in part one, but I’ll go ahead and stick it in here as well!…
‘Any sufficiently advanced technology is indistinguishable from magic.’Arthur C Clarke
Imagine speaking with an eminent scientist from 500 years ago and telling them that within a few hundred years it would be possible to send messages around the Earth (which is not the centre of the universe, btw) almost instantly, hold a little box in front of your face and see someone on the other side of the planet (or even in a rocket in space!) in almost real time, or that you could type a little prompt into a computer (“a what?!”) and ask something called AI to create a picture of a cat working on a laptop (“huh?”) while sitting on a futuristic toilet (“toilet?”) in space, and you’d have it within half a minute or so.
What would they say to that?
Well, perhaps if they were a touch on the arrogant side, they might get a bit cross with you for even suggesting it and say it was categorically impossible pseudoscience.
(Actually, maybe they didn’t have the word ‘pseudoscience’ back then, but you get my point).
I always struggle to understand why a lot of people think we’ll never be more ‘right’ than we are right now.
Anyway, rant over.
The main reason given for the view that it’s just not possible to create a simulation of a universe with conscious beings in it, is that there could never be enough computing power and resources to do it.
However, even if that is true for a constant simulation of the entire universe, it might not be necessary to simulate every part at the same time.
When you play Minecraft, the whole world appears to be there whenever you go, right? Same with virtual reality games – your perception is that the entire environment is there all the time.
But we know that it’s not really all there all of the time, so it’s possible that a simulation could be running on some kind of similar mechanism, without every part of it being rendered (or at least fully rendered at an atomic level) at any one time – thus decreasing the compute needed.
Some have suggested that the observer effect in quantum physics could support this idea. I’ll link to a scientist explaining that in the next part of this series.
Several sceptics have also suggested, along the same lines, that it would be impossible to render a simulation in enough detail to ‘trick’ us into believing that this was base reality.
A fair point, but it’s worth remembering that we all manage to ‘trick’ ourselves into believing that our dreams are real every night during the time we’re in them, and we only need a computer the size of our brains to do that.
Now that’s out of the way, let’s get back to the main subject…
Can we determine the probability that we are in a simulation?
So, we now find ourselves at a point of realising that it’s impossible to assign a meaningful probability to the possibility of living in a simulation.
But our human nature won’t allow us to leave it at that (or at least mine won’t), so we’re going to try anyway.
We’re going to find the best answer we can.
Remember that we’re now just dealing with two hypotheses – Simulation vs No Simulation.
One approach at this point is to say that, since we do not have any meaningful data on either side, we assign an equal likelihood to each hypothesis.
This is known as the principle of indifference in Bayesian statistics.
If we do that then, seemingly, the odds of us living in a simulation vs not living in a simulation are 50/50.
However, that is not quite accurate.
That’s because, within the Simulation Hypothesis (Hypothesis 1), we still have the small chance that we’re living in base reality. It could be a tiny chance, but it’s still there.
That means that in Hypothesis 2 (No Simulation) we have the 50% chance that we attributed under the principle of indifference, PLUS a tiny bit ‘carried over’ from Hypothesis 1, because Hypothesis 1 holds within it a small chance of us still being in base reality.
If that doesn’t quite make sense initially, try looking at it the other way round – the fact that the Simulation Hypothesis holds within it a small chance that we are living in a simulation, you have to take away a slight amount of its ‘likeliness’ and put that in the pot for Hypothesis 2 (No Simulation) instead.
The result is that the probability we assign to Hypothesis 2 (No Simulation) has to be larger than the probability we assign to Hypothesis 1 (Simulation), even if only by a fraction of a percentage.
We don’t know how much more of a probability to assign to Hypothesis 2, we just know that it has to be a bit more likely than Hypothesis 1, if we start out from a place of saying that we don’t know anything about either one and that therefore, on the face of it, they are equally likely.
So at this point we could say that there is likely to be a very slightly higher chance that we are living in base reality rather than in a simulation. We could say that the chance of living in base reality is 50.0001%, but the point would be that it still has to be a touch higher.
And that’s probably the best we can do for an answer, if it has to involve specific numbers.
But what would happen in a scenario where the simulations created by the original advanced civilisations started creating their own simulations? And then some of those secondary simulations created their own simulations, and so on down a long chain?
Would that make it any more or less likely that we’re living in a simulation?
Lets’ find out…
What if there are multiple nested simulations?
First, let’s clarify a common misconception.
It’s often suggested that, should our descendants and/or an advanced civilisation create a single simulation, that simulation would be able to create its own simulation given time. Then that secondary simulation could create its own, and so on.
I have heard it suggested that, because that could happen over and over again resulting in many many simulations, we are more likely to be in a simulation because there are more of them.
That logic simply doesn’t make sense in a scenario where the base reality and each simulation only produce one simulation.
This is because we couldn’t possibly be in any of the simulations other than the one at the end of the chain. The end of the chain could be in position one or position 10 000, but that’s the only one we could be in. The only other possibility for us is the original ‘base reality’ at a point in time where it has not yet created a simulation.
Why? Because all of the simulations that come before the end of the chain have created their own simulation. We know that we have not (yet) created a simulation, so we can’t be any of those.
This seems really obvious, but is often overlooked.
However, if the advanced civilisation created multiple simulations, then there would end up being lots and lots of ‘first’ simulations and, in this ‘nested’ scenario, they could go on to create their own multiple simulations and so on and so on.
That being the case, there would then be lots of ‘end of the chain’ simulations and we could be any of those, or a base reality at a point in time when it has not yet created a simulation.
Does that make it more likely that we’re living in a simulation?
We can see here that, should this scenario play out, the majority of simulations occur at the ‘basement level’ and do not (yet) have their own simulations. It works just like a pyramid scheme.
It makes sense then, to assume that if we are a simulation, this is the spot we’d be most likely to be in – at a place where we don’t yet have our own simulation.
So at this point we can see that the fact that we don’t have our own simulation doesn’t make it any less likely that we are a simulation – statistically, under this model, that’s where we should expect to be.
It also doesn’t make it any more likely though, because all of the previous unknowns from Hypothesis 2 are still relevant.
So once again, we’re no further forward.
It has been argued that thinking in these terms (multiple nested simulations, or even the linear nested model) makes it less likely that we’re living in a simulation.
That’s because, in a structure in which simulations create simulations, the amount of compute available at each level has to decrease because it is ultimately all coming from the original base reality.
Based on our current understanding of ‘how stuff works’, that does make sense.
This being the case, there would come a point where an ‘end of the chain’ simulation could not create its own simulation, because there would be insufficient compute available. And bear in mind that we’ve shown that this is where the vast majority of the simulations in this structure would hang out.
So it could be said that this shows a contradiction in the entire theory, because there would be a level of simulated ‘realities’ at the bottom that could not create their own simulated realities.
This would mean that they were not true simulated realities, because they could not do what could be done in base reality, and so the whole theory falls apart.
While this is a realIy good point, I don’t think it’s necessarily a strong enough argument to push the chances of a simulation down.
First of all, it’s not inconceivable that there could be a mechanism for each simulation to achieve the same amount of computing power as the original simulation. It could even, somehow, be infinite – just like the universe. Just because we can’t see how it could be done, doesn’t mean that it couldn’t. But let’s put that to one side because it’s pure speculation.
A better point is that it’s feasible that the base reality could create a perfect simulation, the next level down could be very slightly less perfect, the next one more so and so on until you reach the bottom of the chain, and at that point perhaps no more simulations are possible.
Each simulation is therefore not a perfect simulation of reality (except perhaps the first), but it’s good enough until it gets to the point where it has to stop.
Perhaps we’ll create a simulation that tops out at the level of VR headsets!
That doesn’t really tell us anything about whether or not we currently exist within the final level, which may indeed not be as ‘good quality’ as the first simulation, but we wouldn’t know any differently.
Perhaps at the higher levels up all the beings can communicate telepathically – but they had to drop that feature at our level!
It could be that the simulations at the end of the chain operate fine for the most part, they’re just a bit glitchy now and then! This could even account for things like the ‘Mandela Effect’, which I’ll go into under the ‘Fun Stuff’ section in part three!
None of this really moves us any further forward, except to say that the amount of simulations could be finite and of decreasing quality.
Oh, and that perhaps we should add in another possibility to our ever-growing list of current existential threats to humanity. Because if we are a simulation then, under this logic, if any of the simulations above us in our chain go extinct, or decide to turn their simulation/s off, presumably it’s lights out for us.
Finally, then, we find ourselves at a point where we can’t attribute any meaningful probabilities to whether or not we’re living in a simulation. Which is what I said right at the start.
But, if we feel that we really have to, we can use a statistical analysis to suggest that it’s slightly more likely that we are living in base reality.
And it looks like that’s about the best we can do.
What would happen if we created a simulation? A surprising turn of events!
As a final thought on the matter of probabilities, let’s consider whether the odds in this model might shift if we managed to create our own simulation.
What if, tomorrow, a secret society of scientists revealed that they’d been working on a classified project for years and had finally succeeded in creating a simulation of the universe with conscious beings?
Well, let’s take another look at the previous diagram…
It showed that a huge majority of simulations in a model like this exist in a kind of ‘basement level’, and they haven’t created their own simulations.
So, if the only information you had about a civilisation was that it was a simulation, that’s where you would bet on it being positioned in the model. At the bottom, without its own simulation.
But, now that this secret society has created a simulation, we know that we can’t possibly exist there at the basement level.
We can strike out the vast majority of the possible simulations.
We have to exist higher up the chain.
What’s more, we can’t be one of the simulations higher up the chain that hasn’t created its own simulation, so we can strike out any of those too and just leave the ones that do have their own simulation/s.
Let’s take a look at that on a variation of the previous diagram. We could now be any of the red circles, or the base reality at the top (at a point in time when the first simulation was created and there was no subsequent chain yet), and the rest are discounted.
For the sake of being pedantic, I’ve only circled simulations that have created only one other simulation (which is then at the end of the chain), because that’s the point in time we’d be at.
Remember, there could be hundreds, thousands, or millions of these.
So, at this point we have struck out the majority of simulations in the model, but there could still be many many simulations left, and any one of them could be us.
Or, we could be base reality and are the first advanced civilisation to create a simulation.
Let’s imagine there are one million red dots – they’re the simulations that could be us. And one of those red dots is base reality.
We would then find ourselves in a situation where the odds in this model are HUGELY in favour of us being a simulation AND we have closed down Hypothesis 2, because we now know that it is possible to create a simulation, you don’t necessarily go extinct before you can do it, and you do go ahead and use it.
In other words, we would find ourselves in a situation where we could be almost certain that we ourselves were living in a simulation, and had just created another.
It would be an interesting day.
My view on this is of no consequence to anyone, so feel free to skip it!
But since we’ve come this far, I’m going to give it.
As I’ve now said many times, the answer to the Simulation Argument is that there is no answer, in fact it’s not even really a question.
It’s just a really interesting thought process to follow, because the logical progression does show that one of the three possibilities must be true.
Or… does it?
If I were to be backed into a corner and forced to say what percentage likelihoods I’d give to each possibility, all I could say is that I would assign a greater chance to Possibility 1 than to Possibility 2.
So I think, based on the direction in which humanity appears to be heading, extinction before achieving a simulation (Possibility 1) is not unlikely.
And I think that basic human nature, together with the convergence aspect we discussed, push down the likelihood of Possibility 2.
But I still don’t know where that would leave me for the ‘bit that’s left over’ (Possibility 3, the simulation).
So the best I could do is go with the Bayesian statistical analysis and say that we’re slightly more likely to be living in base reality.
But I’d still think ‘I don’t know’ would be a better answer.
My main objection to the whole premise boils down to something we haven’t discussed, but can be summarised extremely quickly – the unknown unknowns.
We know that the argument rests on an assumption that technology can reach the point of creating a simulation – that’s an explicit assumption.
But in some ways, we don’t know what else it assumes.
We should take that into account, particularly when we’re thinking about an argument that’s predicated upon a willingness to consider that the impossible might be possible!
For example, right at the beginning I said that I was making an assumption that time is linear, because I didn’t know how to not assume that!
I wasn’t trying to make a serious point, but actually it is possible that time operates in a different way.
Before I thought of that, it was an unknown unknown to me. An assumption I didn’t know I was making.
What else could we have missed? Well, probably lots.
We can also question sub-assumptions! For example, we didn’t discuss what consciousness really means when we said that technology could potentially simulate it. And, since I’ve so far written over 5000 words today, I’m not going to start now!
So, however unsatisfactory it may be, the answer to this whole thing can only be… WE DON’T KNOW.
At least, not until we manage to create our own simulation, then we can revisit the issue!
Now my head hurts. So in part 3 – the final part of this series- we’re just going to look at some fun stuff… get ready for glitches in the matrix, the Mandela effect, the double slit experiment in quantum physics, and lots of other stuff that’s much less intense than all this probability nonsense!