Are you living in a computer simulation? [pdf]
simulation-argument.com
simulation-argument.com
The physics in a simple simulation game like Minecraft is consistent from the perspective of inhabitants of that world. And although Minecraft is Turing complete, from the perspective of inhabitants of that world, it would seem equally impossible to simulate their entire universe from the tools available within the game.
Analogous to how Minecraft is a simulation with a simplified model of our physics, our world could be a simulation with a simplified model of 'real' physics. There is no way to ever find out if this is the case.
To me the interesting part of this is that regardless if we are simulated or not, it doesn't make our lives any less 'real'. Suppose we all lived in a simulation, would that matter? We still think, feel, and operate in the same way. It's not like The Truman Show where if we figure it out we get to escape to the larger world. Would knowing that all this is a simulation change the way you behave? Would you be more reckless, or more careless towards others? Any pain we inflict, on ourselves or others, still has the same effect. I don't see how it changes anything.
>Simulating the entire universe down to the quantum level is obviously infeasible, unless radically new physics is discovered. But in order to get a realistic simulation of human experience, much less is needed
I'm glad Bostrom says simulating the universe is infeasible (most people loosely following his argument ignore this), but I'm not convinced that simulating an infinitesimal portion of the universe is sufficient. Rewinding the simulation whenever one of the billions of simulated humans gets skeptical while doing science sounds absurd and kind of defeats the purpose of a simulation, no? In general, I don't think you can ad-hoc "fill in the details" of a particular piece of a simulation and have it make sense because you weren't simulating the inputs in that much detail.
I wrote quite a bit more here: http://jere.in/the-simulation-hypothesis-is-nonsense
Edit: Imagine that you are simulating a qubit. You want the qubit to behave a certain way, but it refuses to do what you want. You try rewinding it over and over with your magic time-rewinder, but it still refuses to give you the measurement that you expect. That's the Free Will Theorem.
Now imagine that you're simulating a human. You want the human to behave a certain way...
If we are in a simulation, we would have no reason to assume that the physics of the universe in which the simulation runs are as they appear to us. If, however, we chose to assume that is so, it would be consistent to also expect our simulated world to have free will up to the point that the entity running the simulation has.
It's well documented that the human brain ignores or fills in inputs for things that mostly match its expectations. You're already ignoring vast swaths of your sensory input at this very moment by focusing on this text.
A simulation of a world of minds, rather than a simulation of its physics, would be many orders of magnitude more feasible because the simulation itself would have knowledge of each mind's sensory input. Therefore, it would know exactly how much fidelity it needed at any given moment, and where.
Furthermore, such an advanced simulation would have knowledge of what its simulated objects are, rather than the constitution of their matter requiring a simulation of its physics. This merely requires a simulation of the interface with the outputs having the right kind of noise. You wave this away via some absurd "sensory deprivation" remark, but this is explicitly the possibility discussed in the Simulation hypothesis.
Whatever science simulated people do, they simply cannot violate the simulation parameters. Advanced humans with knowledge of what algorithms constitute a human mind would simply need to track the inferences minds make in order to avoid seeing through the simulation.
And I don't find your analysis of Bostrom's argument complete. His argument is airtight:
1. Simulations cannot be built: simulations are impossible, advanced civilizations go extinct before they're built, etc.
2. Simulations will not be built: future humans have no interest in ancestor simulations, they have ethical qualms against doing so, etc.
3. We are almost certainly in a simulation: the number of simulated people far outnumber the number of non-simulated people, so you are likely a simulation.
You simply can't be unconvinced by this argument.
"a computer simulation"
"a universe simulation"
"an ancestor simulation in which the lived experience is indistinguishable from our current experience, which necessarily won't be able to simulate the entire universe nor anything at a high level of fidelity and so requires constant tricks to convince the inhabitants things are real"
The last one is required for this argument to work and I think that's what's being slipped in. Sure, everybody loves computer simulations. We run them all the time. Video games are computer simulations, so look at that. We all love them.
We're talking about a very, very specific type of simulation and the more specific you define it, the less likely it is anyone will ever build one.
"Sensory deprivation tank" isn't absurd. That seems to be what we're talking about here. I suppose I will have to read the book because the noise argument you are presenting seems bizarre to me. Do you think your lived experience can be characterized as sensory noise? I absolutely do not understand how you can just simulate minds without an environment unless we are talking about dreams or hallucination.
I don't think scaling down the fidelity in certain places works. Humans expect natural processes to occur even when they're not looking. The evidence will be there long afterwards. There is no particular place that we wouldn't look for such evidence.
>Furthermore, such an advanced simulation would have knowledge of what its simulated objects are, rather than the constitution of their matter requiring a simulation of its physics.
No I don't think so. The number of possible "simulated objects" is infinite. Even the most advanced simulation you imagine would not be written in such a way to account for them.
>Advanced humans with knowledge of what algorithms constitute a human mind would simply need to track the inferences minds make in order to avoid seeing through the simulation.
Not convinced on this point either, but anyway this would conflict with reality considering a great many people already believe with certainty that they are living in a simulation.
No tricks. Simulating the minds is the entire point. Simulating physics isn't the interesting part of these kinds of simulations.
> I absolutely do not understand how you can just simulate minds without an environment
Each simulated mind contains a set of beliefs about the local environment. The intersection of the belief-sets of the people in close physical proximity dictates the local reality. If some belief-sets conflict, choose the set that has more believers.
Global consistency can be assured if all beliefs that survive a conflict satisfy certain conditions.
> I suppose I will have to read the book because the noise argument you are presenting seems bizarre to me.
You don't need to simulate the LHC, you only need to simulate what outputs the scientists believe the LHC should produce, ie. they expect results consistent with the statistical models of quantum mechanics. Introduce some random errors here and there as a demonstration of human fallibility (like the FTL neutrinos mistake a few years back due to a loose cable).
> No I don't think so. The number of possible "simulated objects" is infinite
The number at any given time is not. Every simulated object will have logical properties associated with it, particularly at its interface. Satisfy the logical properties of the interface, and a simulated mind wouldn't know the difference.
> Not convinced on this point either, but anyway this would conflict with reality considering a great many people already believe with certainty that they are living in a simulation.
They may believe it but they can never prove it, and so they're not actually seeing through the simulation.
Being not convinced is not an unreasonable position to hold, but it hardly counts as an argument that renders the premise a nonstarter.
I am pretty sure that one can estimate approximately how much information would be needed to simulate a minimal universe consistent with everthing that humanity currently thinks it has been able to verify - in fact, I have a vague memory that Scott Aaronson presented something like that in one of his blog posts.
Then there are the uncomfortable gaps in our physical knowledge - the incomaptibility of relativity and QM, our inability to test, and choose between, string theories: suppose they cannot be resolved? I am not proposing that this is likely, but if it were, it would not mean that the universe could not continue, whether simulated or not.
I have read a bunch of "this is obviously wrong" claims, and one frequently-occurring aspect is that of begging the question somewhere, by assuming that something or other in what we seem to know about the universe is necessarily so. To be fair, proponents of the simulation belief sometimes also do this.
Personally, I don't have an opinion as to whether I am in a simulation; I put Bostrom's paper away in the file containing metaphysical solipsism, p-zombie arguments, and ontological proofs of the existence of God, as evidence for the futility of trying to reason oneself to ultimate truths.
But given that Moore’s law has run out and there are no pathways we can see to faster computation (we’re even starting to hit the bedrock of physics in our search) such a simulation would probably require considerable resources. We’re talking like, mining asteroids for silicon and indium and using immense solar panels for energy.
Which gets me to the point that I think strikes at the heart of the argument and kills it: A society capable of ancestor simulations would have to be extremely highly ordered. The universe abhors order and is constantly regressing to higher entropy states. So while the existence of such a society may not be impossible, it would seem to be very very unlikely (and perhaps even short-lived if it did happen), thus meaning there probably aren’t many long-lived high-quality ancestor simulations, and thus we probably aren’t living in one.
Why do people run the sims game today? Isn't that a waste of time even by modern standards? Would future humans not waste time with entertainment? Look at how much time people spend on analyzing Conway's Game of Life to find new constructions. Isn't it plausible that future humans might play with what parameters would have had Trump lose in 2016?
Which leads to: ancestor simulations would have significant historical importance. How many historians today would love to simulate what would have happened had Hitler won? Or what would have happened had one of the plots against Alexander the Great succeeded?
Ancestor simulations would also have economic importance. We already simulate toy economic models. However, we also know that humans aren't prefectly rational. If one day we understand the algorithms driving human brains, why wouldn't you run your economic simulations with real human minds for more precision?
I could go on, but I think the point is clear. We have no reason to suppose any of these values would change in "post-humans". They would still have incomplete information about the world, but they would be able to handle substantially more complexity, at the very least because of the simulations they have available to them.
> But given that Moore’s law has run out and there are no pathways we can see to faster computation
There are many, many pathways forwards that are being explored (optical computing, amorphous computing, reversible computing, etc.). Our current architectures are in fact quite inefficient. We could be making much more use of parallelism, for instance.
If that's true then the simulation argument fails. I'm not convinced though. It seems to me that the materialist argument, and Searle seems to be a materialist, is that consciousness is an emergent property of neuronal activity which is basically a form of computation. Well simulated weather isn't weather, but computer simulated computations are actually computations.
So while I am highly skeptical of the simulation argument, I don't object to it in this respect. Also I feel bad disagreeing with Searle as I think he gets so much else right and has made many very valuable contributions.
So Searle does not agree that consciousness is a product of or a form of computation, which is where he and I part ways.
It is certainly true that a purely digital model of a conscious brain could not perform any of the physical actions that a conscious person can do, such as actually move chess pieces, but the addition of actuators for such interventions in the physical world is not conceptually problematical, and clearly beside the point.
Talk of properties, without inquiring more deeply, was pilloried by Molière in a rather well-known comment:
"It is asked what the cause and reason are of opium's making one sleep.
"To which I respond: because there is in it a dormitive virtue whose nature it is to put the senses to sleep."
It also tends to give rise to positions such as a disbelief that iron ships could float, on the grounds that iron, unlike wood, lacks the buoyancy property. The generalized fallacy here lies in taking a lack of knowledge for proof that something is necessarily the case.
There is no need to feel bad about disagreeing wth Searle - he is not lacking in self-confidence, and I imagine he would laugh at the thought.
Brains are the computer of the human, dummy shells we live in. The skeleton, bones and joints are the mechanical elements, as like the engine, axel and wheels is to a car. We know we can make alterations to our computer with drugs which can access the control port. Whether it being LSD or Caffeine. The same as you can with a car by adding a turbo or an induction kit.
If you take my understanding of of quantum computing where a state can either be 0 or 1, 0 can be 1 and 1 can be 0, what's there not to say that alternate realities can't exist at the same time?
Does even behind me even exist? If a tree falls in the woods and there is no one around to hear it, does it make a sound?
As a thought experiment: I want to head to the supermarket. I can either take route A by turning left out of my apartment, or I can either take route B, turning right out of my apartment.
What happens on route A happens because I am there to witness and what would of taken place on route B, I do not know. Would the same set of events of occurred on that route if I had taken route B instead of route A?
Does myself being a variable alter the constants? Would the guy with the red jacket who walked route B, when I took route A, still be walking route B if I had taken route B initially? Or is it because I chose route A they appeared in route B?
I do believe you can live in two realities, not at the same time as you can only be one of yourself, sort-of; but that you can be part of both.
This was just a ramble to pass thirty minutes of my time so I can finish working from home. But if I actually worked those thirty minutes rather then writing a loose HN comment theory, would I later tonight do something different then what I am going to do now. Even though I am sitting at my desk, in the same location. If I hadn't of written this comment would the tracks in my shuffled playlist be different to the ones I've just listened too?
What about before this "computers" ? Universe cannot be run in computer-simulation if the term is not known before that.
Later if something else comes up, we might believe, its "that" technology
verbal narrative, hand-written text, manual printing press, industrial printing press, mechanical tabulation, punched card computing, Von Neumann computing, distributed public computing, ?
Any computer program can be executed manually by a human given that he has access to a read/write memory. The memory can be as simple as a piece of paper with pen and eraser, or a hard disk. So let's say we simulate the feeling of pain in this way. When does the simulated being feel pain, when the "human CPU" makes read/write operations? Or even when the memory is in an idle state? If you have a large enough hard disk with the right data, is there actually someone in pain?
In addition, there are different ways you can encode the memory data, you can say that when this memory location is set to 0, it actually means 1, and vice versa. This means that any data on a large-enough hard disk means that some simulated being is in pain.
1. Simulations cannot be built: simulations are impossible, advanced civilizations go extinct before they're built, etc.
2. Simulations will not be built: future humans have no interest in ancestor simulations, they have ethical qualms against doing so, etc.
3. We are almost certainly in a simulation: the number of simulated people far outnumber the number of non-simulated people, so you are likely a simulation.
You simply can't be unconvinced by this argument.
Crucially, we have simulated our own ancestors, but the simulation did not appear to us to contain any sort of inner experiences from our ancestors. Is this because we aren't modelling faithfully enough, or because there are fundamental limits on the ability to model inner experiences?
That's part of 1: simulations cannot be built. A simulation that loses fidelity is not a proper simulation is it?
Also, you're assuming a simulation of physics, but simulating a world of minds is more feasible.
> Crucially, we have simulated our own ancestors, but the simulation did not appear to us to contain any sort of inner experiences from our ancestors.
We haven't, we have crude models compared to "post-humans" which would have a proper understanding of the algorithms at play in the human mind.
Nonetheless there are some elementary results that we can use. It takes more than one qubit to simulate a qubit, for example.
The simulations I am thinking of, for example, are those which tell us our genetic ancestry [0][1]. We can simulate the entirety of the depth of human sexual expression, in one single dimension, in this way. So do humans really have a rich internal experience relating to their sexuality? I don't know and I can't tell, but nonetheless the simulations are possible.
Would you care to share with the class how a "world of minds" can be simulated? How can the fidelity of a mind be measured? You appear to have solved a lot of the Hard Problem when nobody was looking!
Edit: Ah, you can't. Good try, thanks for playing.
Furthermore, a simulation also needn't simulate the physics equipment we use to test our physics. An advanced simulation of the human brain's algorithms contains semantic content about what the individual believe these machines do, what properties they test for, so they need only simulate the interface and report results consistent with those properties and subject to the expected error distribution.
You seem to be forgetting that these are post-human civilizations, so they would have such detailed understanding of the human brain.
To tackle point (1), note that Bostrom assumes a Doomsday argument [0] and that there are both pre-human and post-human modes of life which are qualitatively different from ours. But it assumes that the structure of the Universe is relatively constant, and that is not the case. For example, we live in a time when galaxies are clearly visible in the night sky; were we to have been born much sooner or later, we wouldn't know about galaxies, and we might further be fooled into thinking that the Milky Way is unique or special.
Point (3) is really the one that Bostrom wants for us to conclude as true, but it's completely inaccessible and unfalsifiable. Just like Christopher Columbus could never sail to the Garden of Earthly Delights, it's not possible for us to falsify the proposition that we live in a simulation, whether it be a Spinozan world, a Great Programmer world, or whatever quasireligious interpretation one might want to use.
We can now freely embrace (2). We run plenty of simulations of pre-human life. One might object that we do not model all of the details of the inner experiences of such life, but of course there is no evidence for such inner experiences and we can ignore them entirely [2].
There's other critiques of this argument. They usually boil down to some sort of post-selection [1]: Bostrom first post-selects to insist that there will inevitably civilizations which could plausibly simulate us, and then uses the post-selection to lock us into already being simulated.
And, of course, there's a final argument from the outside: This would have all been just as true in the 1600s as it would be today, were the argument to be rock-solid. But then, computers would not have existed in the societies of the 1600s, so the argument would have to be prefaced by an introduction which doesn't just wax poetic about how tiny future computers would be, but also explains computation, quantum mechanics, black holes, etc. However, in the 1600s, Spinoza [3] was able to put forth the argument that we are all simulated by the mind of God, and his deductions and rules were not much different from Bostrom's. Spinoza and Bostrom seem to have the same conceptualization of the Universe, and are following the arguments that they think will lead them there best. But those arguments are hopelessly and obviously covered in the technology and ideas of the time. Therefore, if there is a way in which we are simulated, then the actual argument which proves it will have to come with a massive introduction which justifies it, not just by using our current words and ideas, but by introducing some sort of 25th-century mathematics and physics.
Oh, but wait, why would simulations become possible arbitrarily in the 25th century? We can iterate, and imagine that the hypothetical civilization which simulates us is indefinitely far in the future. When we predict this far into the future, not only do our models break down, but our ability to model breaks down. Bostrom far overshoots with this argument!
[0] https://en.wikipedia.org/wiki/Doomsday_argument
[1] https://en.wikipedia.org/wiki/Postselection
I thought it was xkcd but can't find it. Anyone know which webcomic it was?