I suppose I would test out the claim by getting it to mine a few hundred $million of bitcoin for me.
Then crack all the interesting crypto data that is floating around.
Then brute force search for proofs/solutions of all the big problems such as Riemann Hypothesis, factoring etc. Try all texts shorter than say 100 terabytes and see if they solved the problem.
Work out the implications of all the human genes by calculating out how the human body works. This would include things like solving protein folding.
Find the best/shortest algorithm for all the AI/ML challenges that delivers near perfect results.
Then simulate the body with various chemical compounds added to cure cancer, infections etc. Design vaccine + cure for coronavirusV2 and also machines to make them.
I guess you could create a model of the brain and then run all sorts of experiments on the simulation to work out how it works.
Simulate atoms and molecules and brute force ways to build things that would build arbitrary nanotech engines.
At what point would you get worried about AI safety?
I really like your first three ideas, and would definitely consider doing that if I was in this position (although now that I’m thinking about it, I wouldn’t want to accidentally alert any powerful actors against me so early on in my journey, for fear of getting the laptop confiscated/stolen, so I’d be very careful before doing anything that could potentially be traced back to me online). :)
As for “calculating out how the human body works,” I’m not sure it would be that simple to pull off, at least not at first. Taking your statement literally would mean having the laptop simulate an entire human, brain and all, which is discussed later, so for practical purposes I’m assuming what you meant by that is calculating how a typical human cell works; say, a single neuron. You could definitely solve protein folding and probably simulate most chemical interactions fairly trivially, as long as you can express the physics involved as finitely computable functions (which I’m not sure has been proven possible for all of chemistry/quantum mechanics, though I may be mistaken on that). However, in order to figure out how things actually work inside of an entire human cell, you’ll not only need to be able to formally express physics and chemistry, but will also need to know what that cell is chemically composed of in the first place (in order to simulate it properly and not just be given fallible guesses by the computer). In order to make this work, you either already have a pretty much complete formal understanding of a human cell, or have figured out a way to specify your goal so precisely that only a manageable number of valid possibilities are given using the known rules of physics, which seems incredibly hard to do, if not totally impossible with our current tools.
More broadly, the same problem comes up when trying to write a program simulating the human brain. The best neurosurgeons in the world are still in the dark about how most of the brain’s functions are actually performed, and currently have to make do with incredibly generalized and high-level assumptions. In order to simulate a human brain (rather than “simply” create a generalized non-human AI), you would need a level of knowledge about our own inner workings that is not currently available. Thankfully, you might not need to know the exact workings of an adult human brain to make one, but without that knowledge at the very least you will need to be able to fully simulate the growth of an embryonic brain, and be able to properly “feed” it appropriate outside stimulation, which could plausibly be reduced to the problem of perfectly simulating the working of a single embryonic cell, then letting the simulation proceed smoothly from there.
Regardless, both goals reduce to the general problem that in order to simulate a complex system, we must already have at least some amount of “base knowledge” of that system, or to put it more precisely, we must know at least as much information as is contained in its Kolmogorov complexity. (please correct me if I’m wrong about this btw, I’m fairly confident in saying this, but I may have messed up somewhere due to the complexity (heh) of the issue)
That’s what I think makes this hypothetical so interesting to me—the thought that even with unbounded finite computational abilities, some of our most important problems would still require a tremendous amount of physical fieldwork, and would certainly still require thinking intelligently about how to code for the solutions we want.
I suppose I would test out the claim by getting it to mine a few hundred $million of bitcoin for me.
Then crack all the interesting crypto data that is floating around.
Then brute force search for proofs/solutions of all the big problems such as Riemann Hypothesis, factoring etc. Try all texts shorter than say 100 terabytes and see if they solved the problem.
Work out the implications of all the human genes by calculating out how the human body works. This would include things like solving protein folding.
Find the best/shortest algorithm for all the AI/ML challenges that delivers near perfect results.
Then simulate the body with various chemical compounds added to cure cancer, infections etc. Design vaccine + cure for coronavirusV2 and also machines to make them.
I guess you could create a model of the brain and then run all sorts of experiments on the simulation to work out how it works.
Simulate atoms and molecules and brute force ways to build things that would build arbitrary nanotech engines.
At what point would you get worried about AI safety?
I really like your first three ideas, and would definitely consider doing that if I was in this position (although now that I’m thinking about it, I wouldn’t want to accidentally alert any powerful actors against me so early on in my journey, for fear of getting the laptop confiscated/stolen, so I’d be very careful before doing anything that could potentially be traced back to me online). :)
As for “calculating out how the human body works,” I’m not sure it would be that simple to pull off, at least not at first. Taking your statement literally would mean having the laptop simulate an entire human, brain and all, which is discussed later, so for practical purposes I’m assuming what you meant by that is calculating how a typical human cell works; say, a single neuron. You could definitely solve protein folding and probably simulate most chemical interactions fairly trivially, as long as you can express the physics involved as finitely computable functions (which I’m not sure has been proven possible for all of chemistry/quantum mechanics, though I may be mistaken on that). However, in order to figure out how things actually work inside of an entire human cell, you’ll not only need to be able to formally express physics and chemistry, but will also need to know what that cell is chemically composed of in the first place (in order to simulate it properly and not just be given fallible guesses by the computer). In order to make this work, you either already have a pretty much complete formal understanding of a human cell, or have figured out a way to specify your goal so precisely that only a manageable number of valid possibilities are given using the known rules of physics, which seems incredibly hard to do, if not totally impossible with our current tools.
More broadly, the same problem comes up when trying to write a program simulating the human brain. The best neurosurgeons in the world are still in the dark about how most of the brain’s functions are actually performed, and currently have to make do with incredibly generalized and high-level assumptions. In order to simulate a human brain (rather than “simply” create a generalized non-human AI), you would need a level of knowledge about our own inner workings that is not currently available. Thankfully, you might not need to know the exact workings of an adult human brain to make one, but without that knowledge at the very least you will need to be able to fully simulate the growth of an embryonic brain, and be able to properly “feed” it appropriate outside stimulation, which could plausibly be reduced to the problem of perfectly simulating the working of a single embryonic cell, then letting the simulation proceed smoothly from there.
Regardless, both goals reduce to the general problem that in order to simulate a complex system, we must already have at least some amount of “base knowledge” of that system, or to put it more precisely, we must know at least as much information as is contained in its Kolmogorov complexity. (please correct me if I’m wrong about this btw, I’m fairly confident in saying this, but I may have messed up somewhere due to the complexity (heh) of the issue)
That’s what I think makes this hypothetical so interesting to me—the thought that even with unbounded finite computational abilities, some of our most important problems would still require a tremendous amount of physical fieldwork, and would certainly still require thinking intelligently about how to code for the solutions we want.