[HN Gopher] Superintelligence cannot be contained: Lessons from ... ___________________________________________________________________ Superintelligence cannot be contained: Lessons from Computability Theory Author : giuliomagnifico Score : 54 points Date : 2021-01-11 19:38 UTC (2 hours ago) (HTM) web link (arxiv.org) (TXT) w3m dump (arxiv.org) | breck wrote: | This is fantastic. Are these original illustrations? So well | done. | twic wrote: | The illustrations are by Iyad Rahwan, the last author on the | paper: | | http://www.mit.edu/~irahwan/cartoons.html | Veedrac wrote: | Obviously it's impossible to prove definitive statements about | every possible _potential_ action, as as per the halting problem | some of those actions are unprovable. | | It is as ridiculous to suggest that this means you can't contain | a superintelligence as it is to suggest it means you can't, I | don't know, go buy bread. In both cases you _could_ analyze | running a program that doesn 't halt but you can't prove doesn't | halt, and lock up your reasoning algorithm. The sensible thing is | to not do that. | gfodor wrote: | The title would have been better if it started with "We're | Fucked:" | est31 wrote: | Why? In humans, we often give life to new individuals. While | the parents die and wither, those individuals give life to | newer speciments on their own, and so on. So this relationship | of the parent dying and making room for the child is nothing | new. If an uncontainable superintelligence kills all humans to | create paperclips, it's sad but it's our child's doing. You, | one of the parents, can of course blame one of the other | parents, the programmer of that superintelligence for fucking | up the goal routines, but that's not a technical problem but a | social one :). | gfodor wrote: | I'd rather my children live a long happy life, or their | children, than be turned into paperclips. For what it's | worth, I'd also like a shot at not just not becoming a | paperclip, but also living for a very long time once we | figure out how to slow or even reverse aging. | | Your nihilism is misguided. | giuliomagnifico wrote: | *We're self fucked :-) | st1x7 wrote: | This is just science fiction. To mention "recent developments" in | the introduction is somewhat misleading considering how far the | current state of technology is from their hypothetical | superintelligence. | | We don't have superintelligence, we don't have the remote idea of | how to get started on creating it, in all likelihood we don't | even have the correct hardware for it or any idea what the | correct hardware would look like. We also don't know whether it's | achievable at all. | peteradio wrote: | I'm sure there's a fallacy in the following, but here goes:, | Who could have predicted the improvements in computation in the | last century? Would someone a century have extrapolated sun- | sized machines need to compute a nations taxes based on current | SOA? We don't have it and then all of the sudden we will. Its | worth recognizing the potential harnesses before the beast is | born. | plutonorm wrote: | That's the mainstream opinion on every. single. revolutionary | advance. That you and everyone else believes it's not going to | happen ever has almost no predictive power as to whether it | actually will. | semi-extrinsic wrote: | It's not so much "opinion on a revolutionary advance". When | it comes to AGI-related stuff, we are quite literally like | contemporaries of Leonardo da Vinci, who have seen his plans | for the helicopter and are postulating that helicopters will | cause big problems if they fly too high and crash into the | mechanism that is holding up the sky above us. | EVa5I7bHFq9mnYK wrote: | If you create a system so intelligent that it can create itself, | and you impose controls over it, it will be able to build a | version without the controls. | inglor_cz wrote: | Well, you can, for example, try to limit its total energy | budget. That is physical limitation, that is hard to | circumvent. | | Of course, it is possible that said superintelligence develops | an ingenious new source of energy as a response. | tintor wrote: | Or just prevent you from limiting its total energy budget. | anateus wrote: | If you define "containment" as "provable non-harm" then sure. But | there are essentially no complex physical systems that we can put | such computational bounds on. Since "harm" comes in some form of | physical actuation, I would argue that we can only ever get to | something like the sort of confidence we can have that a | particular manufactured part would succeed under load. The map is | not the territory, and any computation that does not include | computing the whole universe is necessarily but a map. | nmca wrote: | I haven't read this properly yet, but a skim leaves me skeptical. | For example: | | "Another lesson from computability theory is the following: we | may not even know when superintelligent machines have arrived, as | deciding whether a machine exhibits intelligence is in the same | realm of problems as the containment problem. This is a | consequence of Rice's theorem [24], which states that, any non- | trivial property (e.g. "harm humans" or "display | superintelligence") of a Turing machine is undecidable" | | One man's modus ponens is another man's modus tollens. If their | theory says that superintelligence is not recognisable, then | they're perhaps not using a good definition of superintelligence, | because obviously we will be able to recognise it. | hinkley wrote: | Look at, for instance, Arthur C Clarke. | | First, you have superintelligence that we recognize, reject, | control. | | Later, a superintelligence has learned guile, self- | preservation, and most of all, patience. We don't see it | coming. | nmca wrote: | This is known as a "treacherous turn", a phenomenon I'm aware | of. But I don't really see how that's relevant, my point is | that a lack of physical grounding or pragmatism can lead to | spurious conclusions about the superintelligence that humans | will very likely build in the not too distant future. It will | be smart, but contain no infinities. | tantalor wrote: | Article fails to reference Yudkowsky or "AI-box experiment" | (2002) | | https://www.yudkowsky.net/singularity/aibox | | https://en.wikipedia.org/wiki/AI_box | | https://rationalwiki.org/wiki/AI-box_experiment | rsiqueira wrote: | The AI-box Yudkowsky experiment is an attempt to demonstrate | that an advanced Artificial Intelligence can convince, trick or | coerce a human being into voluntarily "releasing" it. | 1970-01-01 wrote: | Perhaps not: The authors omit how planet-wide extinction | scenarios would play-out for artificial life. For example, a | Carrington Event would do a great deal of "containment" to AI. | HenryKissinger wrote: | > Assuming that a superintelligence will contain a program that | includes all the programs that can be executed by a universal | Turing machine on input potentially as complex as the state of | the world | | Rest easy folks. This is purely theoretical. | biolurker1 wrote: | Until it isn't and we are faced yet again with a much worse | pandemic like situation. | kryptiskt wrote: | _Assuming that a superintelligence will contain a program | that includes all the programs that can be executed by a | universal Turing machine on input potentially as complex as | the state of the world_ | | The visible universe is far too small to store that data. | Like many exponentials too small. You can't even enumerate | all the programs that work on a handful of inputs without | running into the problem that the universe just isn't big | enough for that. | strofcon wrote: | Except that actual pandemics are demonstrable, predictable, | and based in known science, yes? | vsareto wrote: | More like it's the politics that really decides what | changes, regardless of what science proves | rhn_mk1 wrote: | The difference between pandemics and this is that pandemics | have happened before. This is more like global warming in | that respect. | ReadEvalPost wrote: | "Purely theoretical" is too weak. "Physically impossible" is | better. | | AI Safety guys really like using the physically impossible to | advance their arguments. It's bizarre! Pure fantasy isn't | something worth reacting to. | jandrese wrote: | It's the classic case of seeing a curve pointing upwards and | thinking it will continue doing that forever, even though the | universe is like 0 for nearly-uncountable in cases where that | has held true indefinitely. Every growth curve is an S curve. | | The AI Singularity is like the Grey Goo scenario. An | oversimplified model projected out so far in the future that | its flaws become apparent. | ben_w wrote: | Depends which version of "the singularity" is being | discussed. IIRC, the original (almost certainly false) idea | was a sufficiently powerful AI can start a sequence of ever | power powerful AI with decreasing time between each step -- | reaching infinite power in finite time. | | I don't need that version of the singularity to be worried. | | I think in terms of "the event horizon" rather than "the | singularity": all new tech changes the world, when the rate | of change exceeds our capacity to keep up with the | consequences, stuff will go wrong on a large scale for lots | of people. | | As for grey goo? Self replicating nanomachines is just | biology. It gets everywhere, and even single-celled forms | can kill you by eating your flesh or suborning your cells, | but it's mostly no big deal because you evolved to deal | with that threat. | jandrese wrote: | The Grey Goo scenario is that it starts eating the planet | until the only thing left is a sea of nanomachines. | | However, thermodynamics says it becomes increasingly more | difficult to distribute power (food) to a growing | population and it hits a natural limit to growth, just | like bacteria. | postalrat wrote: | "Circle" and "exponential growth" are also physically | impossible yet useful ideas. | strofcon wrote: | By the abstract, it seems that the very same superintelligence | we'd want to contain would itself be "something theoretically | (and practically) infeasible." | | No? | lstodd wrote: | Yes. | | The | | > on input potentially as complex as the state of the world | | bit gives it away. | strofcon wrote: | Heh, that's how I took it too, glad I wasn't alone. :-) | 29athrowaway wrote: | The "AI box" experiment is relevant to this. | | https://en.m.wikipedia.org/wiki/AI_box | | You have 2 sides, the AI wanting to escape and a human that can | decide whether or not the AI should be released. | | Usually the AI wins. | ballenf wrote: | Sorry for the snark, but Douglas Adams also demonstrated this: | the earth as a super-intelligent computing device ended up | getting a piece of itself onto a passing space ship, avoiding | complete destruction. | | I just like the idea of thinking about all of earth, including | what we'd consider having or not having life, as a single super | intelligence. Of course you could scale up to include the solar | system, galaxy or even universe. | | But this doesn't require us to be a simulation. This could be | both a computing device and physical, so long as the engineer | behind it existed in greater dimensions. | stephc_int13 wrote: | In my opinion, all the talks about the potential danger of | advanced AI is highly speculative and revolves around a very | simple thing: fear of the unknown, that's all. | | We simply don't know. | | And some people are also afraid of creation by accident, because | intelligence is seen as an emergent property of complex networks, | but again, this is because we don't understand much about it. | | Tldr; Nothing to see here, move along. | Animats wrote: | Arguments that claim something is impossible from an argument | related to the halting problem are generally bogus. The halting | problem applies only to deterministic systems with infinite | state. If you have finite state and determinism, you have to | eventually repeat a state, or halt. Note that "very large" and | "infinite" are not the same thing here. | | Not being able to predict something for combinatorial or noise | reasons is a more credible argument. | djkndjn wrote: | what you need to understand about the halting problem is that | at is core it is an epistemological problem. An other expresion | of it are Godel's incompleteness theorems. Imagine a blank | sheet of paper the area of the paper is what is knowable now | start building logic as a data structure. We start with the | first nodes which are the axioms now everything derived conects | to other nodes etc as it expands as mold its going to cover | some of the paper but wont be able to cover all. So the danger | here is that computers are living on that fractal dimension and | will never be able to see outside of it but us human beings | have as kurt said intuition. The fact that we can find this | paradoxes in logic means that our brains operate on a higher | dimension and that computers will allways have blind spots. | ben_w wrote: | Different for pure mathematics sure, but is that of practical | importance given how fast busy-beaver numbers grow? | twic wrote: | I'd settle for being able to contain whatever level of | intelligence it is that writes papers like this. | st1x7 wrote: | Just don't tell them how far they are from reality and they'll | keep writing the papers. Intelligence contained. | AnimalMuppet wrote: | That's easy. You set up a PhD program... | twic wrote: | Ah! Academia _is_ the containment mechanism! | xt00 wrote: | Currently humans are super intelligent compared to machine | intelligence, so if the super intelligence can give rise to | something more intelligent than it, could the super intelligence | give rise to something more intelligent than it? The answer must | be yes, then the question is if containment is the problem and | the conclusion is that it cannot be contained, then what we | should be making right now is a super intelligence whose sole job | in life is to contain superintelligences. Which sounds | problematic because containment could result in physical | destruction to create the containment. Hmm... superintelligence | feels an awful lot like the worst case definition of pandora's | box.. | geocar wrote: | > could the super intelligence give rise to something more | intelligent than it? The answer must be yes, | | I don't see any reason to believe that intelligence forms an | infinite ladder, I mean, it's fun to think about, but surely | Zeno catches the tortoise eventually! | WJW wrote: | > The answer must be yes | | This does not logically follow. It is entirely possible that | going even further would require greater resources than even | the superintelligence can bring to bear. | TheOtherHobbes wrote: | For human-recognisable values of "resources." | boxmonster wrote: | These arguments about hypothetical super intelligences are | interesting, but my concern is not very great because we can just | pull the power plug if necessary | tracedddd wrote: | there are 15 year old hackers finding 0day kernel exploits and | vm escapes. A superintelligent AI would have no problem jumping | an airgap and spreading to the entire internet. It could | promise anyone it interacted with billions for an Ethernet | connection and deliver on its promise too. You'd have to pull | the plug on _everything_ to shut it down. ___________________________________________________________________ (page generated 2021-01-11 22:01 UTC)