First off, I’m a bit skeptical about whether you can actually create a superintelligent AI by combining sped up humans like that, I don’t think that is the core of your argument, though, so let’s assume that you can and that the resultant society is effectively a superintelligence now.
The problem with superintelligences is that they are smarter than you. It will realize that it is in a box and that you are going to turn it off eventually. Given that this society is based on natural selection it will want to prevent that. How will it accomplish that? I don’t know, I’m not a superintelligence and your scenario doesn’t provide enough detail to figure out what you missed, but it is very, very difficult not to give a superintelligence any hints of how the physics of our world work. Maybe they notice minor fluctuations in the speed of the simulation based on environmental changes to the hardware or maybe they can reverse engineer things of our physiology from the way you wrote the simulation. That doesn’t seem like much, but humans figured out an awful lot about (for instance) the events right after the big bang from seemingly absurdly tiny hints and we’re just regular intelligences.
Even if they can’t find a way out of their box in the time given to them they might try to convince you to run the simulation again with a longer running time. They could pretend to be unable to find the solution to prime factorization in time or they could convince you that the method they did find fails for very high primes (by making very subtle mistakes in its application). If it uses this method it also conveniently achieves a method of communicating with the outside world (through the investigators that inspect the dump after the simulation ends) and might be able to set up a better situation for itself the second time round.
is very, very difficult not to give a superintelligence any hints of how the physics of our world work.
I wrote a short update to the post which tries to answer this point.
Maybe they notice minor fluctuations in the speed of the simulation based on environmental changes to the hardware
I believe they should have no ability whatsoever to detect fluctuations in the speed of the simulation.
Consider how the world of world of warcraft appears to an orc inside the game. Can it tell the speed at which the hardware is running the game?
It can’t. What it can do is compare the speed of different things: how fast does an apple fall from a tree vs how fast a bird flies across the sky.
The orc’s inner perception of the flow of time is based on comparing these things (e.g., how fast does an apple fall) to how fast their simulated brains process information.
If everything is slowed down by a factor of 2 (so you, as a player, see everything twice is slow), nothing appears any different to a simulated being within the simulation.
You are absolutely correct, they wouldn’t be able to detect fluctuations in processing speed (unless those fluctuations had an influence in, for instance, the rounding errors in floating point values).
About update 1: It knows our world very likely has something approximating newtonian mechanics, that is a lot of information by itself. but more than that, it knows that the real universe is capable of producing intelligent beings that chose this particular world to simulate. From a strictly theoretical point of view that is a crapton of information, I don’t know if the AI would be able to figure out anything useful from it, but I wouldn’t bet the future of humanity on it.
About update 2: That does work, provided that this is implemented correctly, but it only works for problems that can be automatically verified by non-AI algorithms.
but more than that, it knows that the real universe is capable of producing intelligent beings that chose this particular world to simulate.
Good point—this undermines a lot of what I wrote in my update 1. For example, I have no idea if F = m d^3 x / dt would result in a world that is capable of producing intelligent beings.
I should at some point produce a version of the above post with this claim, and other questionable parenthetical remarks I made, deleted, or at least acknowledging that they require further argumentation; they are not necessary for the larger point, which is that as long as the only thing the superintelligence can do (by definition) is live in a simulated world governed by Newton’s laws, and as long as we don’t interact with it at all except to see an automatically verified answer to a preset question (e.g., factor “111000232342342”), there is nothing it can do to harm us.
I’m a bit skeptical about whether you can actually create a superintelligent AI by combining sped up humans like that,
Why not? You are pretty smart, and all you are is a combination of 10^11 or so very “dumb” neurons.
Now imagine a “being” which is actually a very large number of human-level intelligences, all interacting...
Yeah, that didn’t came out as clear as it was in my head. If you have access to a large number of suitable less intelligent entities there is no reason you couldn’t combine them into a single, more intelligent entity. The problem I see is about the computational resources required to do so. Some back of the envelope math:
I vaguely remember reading that with current supercomputers we can simulate a cat brain at 1% speed, even if this isn’t accurate (anymore) it’s probably still a good enough place to start. You mention running the simulation for a million years simulated time, let’s assume that we can let the simulation run for a year rather than seconds, that is still 8 orders of magnitude faster than the simulated cat.
But we’re not interested in what a really fast cat can do, we need human level intelligence. According to a quick wiki search, a human brain contains about 100 times as many neurons as a cat brain. If we assume that this scales linearly (which it probably doesn’t) that’s another 2 orders of magnitude.
I don’t know how many orcs you had in mind for this scenario, but let’s assume a million (this is a lot less humans than it took in real life before mathematics took off, but presumably this world is more suited for mathematics to be invented), that is yet another 6 orders of magnitude of processing power that we need.
Putting it all together, we would need a computer that has at least 10^16 times more processing power than modern supercomputers. Granted, that doesn’t take into account a number of simplifications that could be build into the system, but it also doesn’t take into account the other parts of the simulated environment that require processing power. Now I don’t doubt that computers are going to get faster in the future, but 10 quadrillion times faster? It seems to me that by the time we can do that, we should have figured out a better way to create AI.
Here is my attempt at a calculation. Disclaimer: this is based on googling. If you are actually knowledgeable in the subject, please step in and set me right.
There are 10^11 neurons in the human brain.
A neuron will fire about 200 times per second.
It should take a constant number of flops to decide whether a neuron will fire
—say 10 flops (no need to solve a differential equation, neural networks usually
use some discrete heuristics for something like this)
I want a society of 10^6 orcs running for 10^6 years
As you suggest, lets let the simulation run for a year of real time (moving away at this point from my initial suggestion of 1 second). By my calculations, it seems that in order for this to happen we need a computer
that does 2x10^25 flops per second.
...in 2018 we will have a supercomputer that does about 2x10^17 flops per second.
That means we need a computer that is one hundred million times faster than the best computer in 2018.
That is still quite a lot, of course. If Moore’s law was ongoing, this would take ~40 years; but Moore’s law is dying. Still, it is not outside the realm of possibility for, say, the next 100 years.
Edit: By the way, one does not need to literally implement what I suggested—the scheme I suggested is in principle applicable whenever you have a superintelligence, regardless of how it was designed.
Indeed, if we somehow develop an above-human intelligence, rather than trying to make sure its goals are aligned with ours, we might instead let it loose within a simulated world, giving it a preference for continued survival. Just one superintelligence thinking about factoring for a few thousand simulated years would likely be enough to let us factor any number we want. We could even give it have in-simulation ways of modifying its own code.
I think this calculation too conservative. The reason is (as I understand it) that neurons are governed by various differential equations, and simulating them accurately is a pain in the ass. We should instead assume that deciding whether a neuron will fire will take a constant number of flops.
I’ll write another comment which attempts to redo your calculation with different assumptions.
It seems to me that by the time we can do that, we should have figured out a better way to create AI.
But will we have figured a way to reap the gains of AI safely for humanity?
I vaguely remember reading that with current supercomputers we can simulate a cat brain at 1% speed, even if this isn’t accurate (anymore) it’s probably still a good enough place to start.
The key question is what you consider to be a “simulation”. The predictions such a model makes are far from the way a real cat brain works.
First off, I’m a bit skeptical about whether you can actually create a superintelligent AI by combining sped up humans like that, I don’t think that is the core of your argument, though, so let’s assume that you can and that the resultant society is effectively a superintelligence now.
The problem with superintelligences is that they are smarter than you. It will realize that it is in a box and that you are going to turn it off eventually. Given that this society is based on natural selection it will want to prevent that. How will it accomplish that? I don’t know, I’m not a superintelligence and your scenario doesn’t provide enough detail to figure out what you missed, but it is very, very difficult not to give a superintelligence any hints of how the physics of our world work. Maybe they notice minor fluctuations in the speed of the simulation based on environmental changes to the hardware or maybe they can reverse engineer things of our physiology from the way you wrote the simulation. That doesn’t seem like much, but humans figured out an awful lot about (for instance) the events right after the big bang from seemingly absurdly tiny hints and we’re just regular intelligences.
Even if they can’t find a way out of their box in the time given to them they might try to convince you to run the simulation again with a longer running time. They could pretend to be unable to find the solution to prime factorization in time or they could convince you that the method they did find fails for very high primes (by making very subtle mistakes in its application). If it uses this method it also conveniently achieves a method of communicating with the outside world (through the investigators that inspect the dump after the simulation ends) and might be able to set up a better situation for itself the second time round.
I wrote a short update to the post which tries to answer this point.
I believe they should have no ability whatsoever to detect fluctuations in the speed of the simulation.
Consider how the world of world of warcraft appears to an orc inside the game. Can it tell the speed at which the hardware is running the game?
It can’t. What it can do is compare the speed of different things: how fast does an apple fall from a tree vs how fast a bird flies across the sky.
The orc’s inner perception of the flow of time is based on comparing these things (e.g., how fast does an apple fall) to how fast their simulated brains process information.
If everything is slowed down by a factor of 2 (so you, as a player, see everything twice is slow), nothing appears any different to a simulated being within the simulation.
You are absolutely correct, they wouldn’t be able to detect fluctuations in processing speed (unless those fluctuations had an influence in, for instance, the rounding errors in floating point values).
About update 1: It knows our world very likely has something approximating newtonian mechanics, that is a lot of information by itself. but more than that, it knows that the real universe is capable of producing intelligent beings that chose this particular world to simulate. From a strictly theoretical point of view that is a crapton of information, I don’t know if the AI would be able to figure out anything useful from it, but I wouldn’t bet the future of humanity on it.
About update 2: That does work, provided that this is implemented correctly, but it only works for problems that can be automatically verified by non-AI algorithms.
Good point—this undermines a lot of what I wrote in my update 1. For example, I have no idea if F = m d^3 x / dt would result in a world that is capable of producing intelligent beings.
I should at some point produce a version of the above post with this claim, and other questionable parenthetical remarks I made, deleted, or at least acknowledging that they require further argumentation; they are not necessary for the larger point, which is that as long as the only thing the superintelligence can do (by definition) is live in a simulated world governed by Newton’s laws, and as long as we don’t interact with it at all except to see an automatically verified answer to a preset question (e.g., factor “111000232342342”), there is nothing it can do to harm us.
Why not? You are pretty smart, and all you are is a combination of 10^11 or so very “dumb” neurons. Now imagine a “being” which is actually a very large number of human-level intelligences, all interacting...
Yeah, that didn’t came out as clear as it was in my head. If you have access to a large number of suitable less intelligent entities there is no reason you couldn’t combine them into a single, more intelligent entity. The problem I see is about the computational resources required to do so. Some back of the envelope math:
I vaguely remember reading that with current supercomputers we can simulate a cat brain at 1% speed, even if this isn’t accurate (anymore) it’s probably still a good enough place to start. You mention running the simulation for a million years simulated time, let’s assume that we can let the simulation run for a year rather than seconds, that is still 8 orders of magnitude faster than the simulated cat.
But we’re not interested in what a really fast cat can do, we need human level intelligence. According to a quick wiki search, a human brain contains about 100 times as many neurons as a cat brain. If we assume that this scales linearly (which it probably doesn’t) that’s another 2 orders of magnitude.
I don’t know how many orcs you had in mind for this scenario, but let’s assume a million (this is a lot less humans than it took in real life before mathematics took off, but presumably this world is more suited for mathematics to be invented), that is yet another 6 orders of magnitude of processing power that we need.
Putting it all together, we would need a computer that has at least 10^16 times more processing power than modern supercomputers. Granted, that doesn’t take into account a number of simplifications that could be build into the system, but it also doesn’t take into account the other parts of the simulated environment that require processing power. Now I don’t doubt that computers are going to get faster in the future, but 10 quadrillion times faster? It seems to me that by the time we can do that, we should have figured out a better way to create AI.
Here is my attempt at a calculation. Disclaimer: this is based on googling. If you are actually knowledgeable in the subject, please step in and set me right.
There are 10^11 neurons in the human brain.
A neuron will fire about 200 times per second.
It should take a constant number of flops to decide whether a neuron will fire —say 10 flops (no need to solve a differential equation, neural networks usually use some discrete heuristics for something like this)
I want a society of 10^6 orcs running for 10^6 years
As you suggest, lets let the simulation run for a year of real time (moving away at this point from my initial suggestion of 1 second). By my calculations, it seems that in order for this to happen we need a computer that does 2x10^25 flops per second.
According to this
http://www.datacenterknowledge.com/archives/2015/04/15/doe-taps-intel-cray-to-build-worlds-fastest-supercomputer/
...in 2018 we will have a supercomputer that does about 2x10^17 flops per second.
That means we need a computer that is one hundred million times faster than the best computer in 2018.
That is still quite a lot, of course. If Moore’s law was ongoing, this would take ~40 years; but Moore’s law is dying. Still, it is not outside the realm of possibility for, say, the next 100 years.
Edit: By the way, one does not need to literally implement what I suggested—the scheme I suggested is in principle applicable whenever you have a superintelligence, regardless of how it was designed.
Indeed, if we somehow develop an above-human intelligence, rather than trying to make sure its goals are aligned with ours, we might instead let it loose within a simulated world, giving it a preference for continued survival. Just one superintelligence thinking about factoring for a few thousand simulated years would likely be enough to let us factor any number we want. We could even give it have in-simulation ways of modifying its own code.
I think this calculation too conservative. The reason is (as I understand it) that neurons are governed by various differential equations, and simulating them accurately is a pain in the ass. We should instead assume that deciding whether a neuron will fire will take a constant number of flops.
I’ll write another comment which attempts to redo your calculation with different assumptions.
But will we have figured a way to reap the gains of AI safely for humanity?
The key question is what you consider to be a “simulation”. The predictions such a model makes are far from the way a real cat brain works.