AGI Ruin: A List of Lethalities

Crossposted from the AI Alignment Forum. May contain more technical jargon than usual.


(If you're already familiar with all basics and don't want any preamble, skip ahead to Section B for technical difficulties of alignment proper.)

I have several times failed to write up a well-organized list of reasons why AGI will kill you.  People come in with different ideas about why AGI would be survivable, and want to hear different obviously key points addressed first.  Some fraction of those people are loudly upset with me if the obviously most important points aren't addressed immediately, and I address different points first instead.

Having failed to solve this problem in any good way, I now give up and solve it poorly with a poorly organized list of individual rants.  I'm not particularly happy with this list; the alternative was publishing nothing, and publishing this seems marginally more dignified.

Three points about the general subject matter of discussion here, numbered so as not to conflict with the list of lethalities:

-3.  I'm assuming you are already familiar with some basics, and already know what 'orthogonality' and 'instrumental convergence' are and why they're true.  People occasionally claim to me that I need to stop fighting old wars here, because, those people claim to me, those wars have already been won within the important-according-to-them parts of the current audience.  I suppose it's at least true that none of the current major EA funders seem to be visibly in denial about orthogonality or instrumental convergence as such; so, fine.  If you don't know what 'orthogonality' or 'instrumental convergence' are, or don't see for yourself why they're true, you need a different introduction than this one.

-2.  When I say that alignment is lethally difficult, I am not talking about ideal or perfect goals of 'provable' alignment, nor total alignment of superintelligences on exact human values, nor getting AIs to produce satisfactory arguments about moral dilemmas which sorta-reasonable humans disagree about, nor attaining an absolute certainty of an AI not killing everyone.  When I say that alignment is difficult, I mean that in practice, using the techniques we actually have, "please don't disassemble literally everyone with probability roughly 1" is an overly large ask that we are not on course to get.  So far as I'm concerned, if you can get a powerful AGI that carries out some pivotal superhuman engineering task, with a less than fifty percent change of killing more than one billion people, I'll take it.  Even smaller chances of killing even fewer people would be a nice luxury, but if you can get as incredibly far as "less than roughly certain to kill everybody", then you can probably get down to under a 5% chance with only slightly more effort.  Practically all of the difficulty is in getting to "less than certainty of killing literally everyone".  Trolley problems are not an interesting subproblem in all of this; if there are any survivors, you solved alignment.  At this point, I no longer care how it works, I don't care how you got there, I am cause-agnostic about whatever methodology you used, all I am looking at is prospective results, all I want is that we have justifiable cause to believe of a pivotally useful AGI 'this will not kill literally everyone'.  Anybody telling you I'm asking for stricter 'alignment' than this has failed at reading comprehension.  The big ask from AGI alignment, the basic challenge I am saying is too difficult, is to obtain by any strategy whatsoever a significant chance of there being any survivors.

-1.  None of this is about anything being impossible in principle.  The metaphor I usually use is that if a textbook from one hundred years in the future fell into our hands, containing all of the simple ideas that actually work robustly in practice, we could probably build an aligned superintelligence in six months.  For people schooled in machine learning, I use as my metaphor the difference between ReLU activations and sigmoid activations.  Sigmoid activations are complicated and fragile, and do a terrible job of transmitting gradients through many layers; ReLUs are incredibly simple (for the unfamiliar, the activation function is literally max(x, 0)) and work much better.  Most neural networks for the first decades of the field used sigmoids; the idea of ReLUs wasn't discovered, validated, and popularized until decades later.  What's lethal is that we do not have the Textbook From The Future telling us all the simple solutions that actually in real life just work and are robust; we're going to be doing everything with metaphorical sigmoids on the first critical try.  No difficulty discussed here about AGI alignment is claimed by me to be impossible - to merely human science and engineering, let alone in principle - if we had 100 years to solve it using unlimited retries, the way that science usually has an unbounded time budget and unlimited retries.  This list of lethalities is about things we are not on course to solve in practice in time on the first critical try; none of it is meant to make a much stronger claim about things that are impossible in principle.

That said:

Here, from my perspective, are some different true things that could be said, to contradict various false things that various different people seem to believe, about why AGI would be survivable on anything remotely remotely resembling the current pathway, or any other pathway we can easily jump to.


Section A:

This is a very lethal problem, it has to be solved one way or another, it has to be solved at a minimum strength and difficulty level instead of various easier modes that some dream about, we do not have any visible option of 'everyone' retreating to only solve safe weak problems instead, and failing on the first really dangerous try is fatal.


1.  Alpha Zero blew past all accumulated human knowledge about Go after a day or so of self-play, with no reliance on human playbooks or sample games.  Anyone relying on "well, it'll get up to human capability at Go, but then have a hard time getting past that because it won't be able to learn from humans any more" would have relied on vacuum.  AGI will not be upper-bounded by human ability or human learning speed.  Things much smarter than human would be able to learn from less evidence than humans require to have ideas driven into their brains; there are theoretical upper bounds here, but those upper bounds seem very high. (Eg, each bit of information that couldn't already be fully predicted can eliminate at most half the probability mass of all hypotheses under consideration.)  It is not naturally (by default, barring intervention) the case that everything takes place on a timescale that makes it easy for us to react.

2.  A cognitive system with sufficiently high cognitive powers, given any medium-bandwidth channel of causal influence, will not find it difficult to bootstrap to overpowering capabilities independent of human infrastructure.  The concrete example I usually use here is nanotech, because there's been pretty detailed analysis of what definitely look like physically attainable lower bounds on what should be possible with nanotech, and those lower bounds are sufficient to carry the point.  My lower-bound model of "how a sufficiently powerful intelligence would kill everyone, if it didn't want to not do that" is that it gets access to the Internet, emails some DNA sequences to any of the many many online firms that will take a DNA sequence in the email and ship you back proteins, and bribes/persuades some human who has no idea they're dealing with an AGI to mix proteins in a beaker, which then form a first-stage nanofactory which can build the actual nanomachinery.  (Back when I was first deploying this visualization, the wise-sounding critics said "Ah, but how do you know even a superintelligence could solve the protein folding problem, if it didn't already have planet-sized supercomputers?" but one hears less of this after the advent of AlphaFold 2, for some odd reason.)  The nanomachinery builds diamondoid bacteria, that replicate with solar power and atmospheric CHON, maybe aggregate into some miniature rockets or jets so they can ride the jetstream to spread across the Earth's atmosphere, get into human bloodstreams and hide, strike on a timer.  Losing a conflict with a high-powered cognitive system looks at least as deadly as "everybody on the face of the Earth suddenly falls over dead within the same second".  (I am using awkward constructions like 'high cognitive power' because standard English terms like 'smart' or 'intelligent' appear to me to function largely as status synonyms.  'Superintelligence' sounds to most people like 'something above the top of the status hierarchy that went to double college', and they don't understand why that would be all that dangerous?  Earthlings have no word and indeed no standard native concept that means 'actually useful cognitive power'.  A large amount of failure to panic sufficiently, seems to me to stem from a lack of appreciation for the incredible potential lethality of this thing that Earthlings as a culture have not named.)

3.  We need to get alignment right on the 'first critical try' at operating at a 'dangerous' level of intelligence, where unaligned operation at a dangerous level of intelligence kills everybody on Earth and then we don't get to try again.  This includes, for example: (a) something smart enough to build a nanosystem which has been explicitly authorized to build a nanosystem; or (b) something smart enough to build a nanosystem and also smart enough to gain unauthorized access to the Internet and pay a human to put together the ingredients for a nanosystem; or (c) something smart enough to get unauthorized access to the Internet and build something smarter than itself on the number of machines it can hack; or (d) something smart enough to treat humans as manipulable machinery and which has any authorized or unauthorized two-way causal channel with humans; or (e) something smart enough to improve itself enough to do (b) or (d); etcetera.  We can gather all sorts of information beforehand from less powerful systems that will not kill us if we screw up operating them; but once we are running more powerful systems, we can no longer update on sufficiently catastrophic errors.  This is where practically all of the real lethality comes from, that we have to get things right on the first sufficiently-critical try.  If we had unlimited retries - if every time an AGI destroyed all the galaxies we got to go back in time four years and try again - we would in a hundred years figure out which bright ideas actually worked.  Human beings can figure out pretty difficult things over time, when they get lots of tries; when a failed guess kills literally everyone, that is harder.  That we have to get a bunch of key stuff right on the first try is where most of the lethality really and ultimately comes from; likewise the fact that no authority is here to tell us a list of what exactly is 'key' and will kill us if we get it wrong.  (One remarks that most people are so absolutely and flatly unprepared by their 'scientific' educations to challenge pre-paradigmatic puzzles with no scholarly authoritative supervision, that they do not even realize how much harder that is, or how incredibly lethal it is to demand getting that right on the first critical try.)

4.  We can't just "decide not to build AGI" because GPUs are everywhere, and knowledge of algorithms is constantly being improved and published; 2 years after the leading actor has the capability to destroy the world, 5 other actors will have the capability to destroy the world.  The given lethal challenge is to solve within a time limit, driven by the dynamic in which, over time, increasingly weak actors with a smaller and smaller fraction of total computing power, become able to build AGI and destroy the world.  Powerful actors all refraining in unison from doing the suicidal thing just delays this time limit - it does not lift it, unless computer hardware and computer software progress are both brought to complete severe halts across the whole Earth.  The current state of this cooperation to have every big actor refrain from doing the stupid thing, is that at present some large actors with a lot of researchers and computing power are led by people who vocally disdain all talk of AGI safety (eg Facebook AI Research).  Note that needing to solve AGI alignment only within a time limit, but with unlimited safe retries for rapid experimentation on the full-powered system; or only on the first critical try, but with an unlimited time bound; would both be terrifically humanity-threatening challenges by historical standards individually.

5.  We can't just build a very weak system, which is less dangerous because it is so weak, and declare victory; because later there will be more actors that have the capability to build a stronger system and one of them will do so.  I've also in the past called this the 'safe-but-useless' tradeoff, or 'safe-vs-useful'.  People keep on going "why don't we only use AIs to do X, that seems safe" and the answer is almost always either "doing X in fact takes very powerful cognition that is not passively safe" or, even more commonly, "because restricting yourself to doing X will not prevent Facebook AI Research from destroying the world six months later".  If all you need is an object that doesn't do dangerous things, you could try a sponge; a sponge is very passively safe.  Building a sponge, however, does not prevent Facebook AI Research from destroying the world six months later when they catch up to the leading actor.

6.  We need to align the performance of some large task, a 'pivotal act' that prevents other people from building an unaligned AGI that destroys the world.  While the number of actors with AGI is few or one, they must execute some "pivotal act", strong enough to flip the gameboard, using an AGI powerful enough to do that.  It's not enough to be able to align a weak system - we need to align a system that can do some single very large thing.  The example I usually give is "burn all GPUs".  This is not what I think you'd actually want to do with a powerful AGI - the nanomachines would need to operate in an incredibly complicated open environment to hunt down all the GPUs, and that would be needlessly difficult to align.  However, all known pivotal acts are currently outside the Overton Window, and I expect them to stay there.  So I picked an example where if anybody says "how dare you propose burning all GPUs?" I can say "Oh, well, I don't actually advocate doing that; it's just a mild overestimate for the rough power level of what you'd have to do, and the rough level of machine cognition required to do that, in order to prevent somebody else from destroying the world in six months or three years."  (If it wasn't a mild overestimate, then 'burn all GPUs' would actually be the minimal pivotal task and hence correct answer, and I wouldn't be able to give that denial.)  Many clever-sounding proposals for alignment fall apart as soon as you ask "How could you use this to align a system that you could use to shut down all the GPUs in the world?" because it's then clear that the system can't do something that powerful, or, if it can do that, the system wouldn't be easy to align.  A GPU-burner is also a system powerful enough to, and purportedly authorized to, build nanotechnology, so it requires operating in a dangerous domain at a dangerous level of intelligence and capability; and this goes along with any non-fantasy attempt to name a way an AGI could change the world such that a half-dozen other would-be AGI-builders won't destroy the world 6 months later.

7.  The reason why nobody in this community has successfully named a 'pivotal weak act' where you do something weak enough with an AGI to be passively safe, but powerful enough to prevent any other AGI from destroying the world a year later - and yet also we can't just go do that right now and need to wait on AI - is that nothing like that exists.  There's no reason why it should exist.  There is not some elaborate clever reason why it exists but nobody can see it.  It takes a lot of power to do something to the current world that prevents any other AGI from coming into existence; nothing which can do that is passively safe in virtue of its weakness.  If you can't solve the problem right now (which you can't, because you're opposed to other actors who don't want to be solved and those actors are on roughly the same level as you) then you are resorting to some cognitive system that can do things you could not figure out how to do yourself, that you were not close to figuring out because you are not close to being able to, for example, burn all GPUs.  Burning all GPUs would actually stop Facebook AI Research from destroying the world six months later; weaksauce Overton-abiding stuff about 'improving public epistemology by setting GPT-4 loose on Twitter to provide scientifically literate arguments about everything' will be cool but will not actually prevent Facebook AI Research from destroying the world six months later, or some eager open-source collaborative from destroying the world a year later if you manage to stop FAIR specifically.  There are no pivotal weak acts.

8.  The best and easiest-found-by-optimization algorithms for solving problems we want an AI to solve, readily generalize to problems we'd rather the AI not solve; you can't build a system that only has the capability to drive red cars and not blue cars, because all red-car-driving algorithms generalize to the capability to drive blue cars.

9.  The builders of a safe system, by hypothesis on such a thing being possible, would need to operate their system in a regime where it has the capability to kill everybody or make itself even more dangerous, but has been successfully designed to not do that.  Running AGIs doing something pivotal are not passively safe, they're the equivalent of nuclear cores that require actively maintained design properties to not go supercritical and melt down.


Section B:

Okay, but as we all know, modern machine learning is like a genie where you just give it a wish, right?  Expressed as some mysterious thing called a 'loss function', but which is basically just equivalent to an English wish phrasing, right?  And then if you pour in enough computing power you get your wish, right?  So why not train a giant stack of transformer layers on a dataset of agents doing nice things and not bad things, throw in the word 'corrigibility' somewhere, crank up that computing power, and get out an aligned AGI?


Section B.1:  The distributional leap. 

10.  You can't train alignment by running lethally dangerous cognitions, observing whether the outputs kill or deceive or corrupt the operators, assigning a loss, and doing supervised learning.  On anything like the standard ML paradigm, you would need to somehow generalize optimization-for-alignment you did in safe conditions, across a big distributional shift to dangerous conditions.  (Some generalization of this seems like it would have to be true even outside that paradigm; you wouldn't be working on a live unaligned superintelligence to align it.)  This alone is a point that is sufficient to kill a lot of naive proposals from people who never did or could concretely sketch out any specific scenario of what training they'd do, in order to align what output - which is why, of course, they never concretely sketch anything like that.  Powerful AGIs doing dangerous things that will kill you if misaligned, must have an alignment property that generalized far out-of-distribution from safer building/training operations that didn't kill you.  This is where a huge amount of lethality comes from on anything remotely resembling the present paradigm.  Unaligned operation at a dangerous level of intelligence*capability will kill you; so, if you're starting with an unaligned system and labeling outputs in order to get it to learn alignment, the training regime or building regime must be operating at some lower level of intelligence*capability that is passively safe, where its currently-unaligned operation does not pose any threat.  (Note that anything substantially smarter than you poses a threat given any realistic level of capability.  Eg, "being able to produce outputs that humans look at" is probably sufficient for a generally much-smarter-than-human AGI to navigate its way out of the causal systems that are humans, especially in the real world where somebody trained the system on terabytes of Internet text, rather than somehow keeping it ignorant of the latent causes of its source code and training environments.)

11.  If cognitive machinery doesn't generalize far out of the distribution where you did tons of training, it can't solve problems on the order of 'build nanotechnology' where it would be too expensive to run a million training runs of failing to build nanotechnology.  There is no pivotal act this weak; there's no known case where you can entrain a safe level of ability on a safe environment where you can cheaply do millions of runs, and deploy that capability to save the world and prevent the next AGI project up from destroying the world two years later.  Pivotal weak acts like this aren't known, and not for want of people looking for them.  So, again, you end up needing alignment to generalize way out of the training distribution - not just because the training environment needs to be safe, but because the training environment probably also needs to be cheaper than evaluating some real-world domain in which the AGI needs to do some huge act.  You don't get 1000 failed tries at burning all GPUs - because people will notice, even leaving out the consequences of capabilities success and alignment failure.

12.  Operating at a highly intelligent level is a drastic shift in distribution from operating at a less intelligent level, opening up new external options, and probably opening up even more new internal choices and modes.  Problems that materialize at high intelligence and danger levels may fail to show up at safe lower levels of intelligence, or may recur after being suppressed by a first patch.

13.  Many alignment problems of superintelligence will not naturally appear at pre-dangerous, passively-safe levels of capability.  Consider the internal behavior 'change your outer behavior to deliberately look more aligned and deceive the programmers, operators, and possibly any loss functions optimizing over you'.  This problem is one that will appear at the superintelligent level; if, being otherwise ignorant, we guess that it is among the median such problems in terms of how early it naturally appears in earlier systems, then around half of the alignment problems of superintelligence will first naturally materialize after that one first starts to appear.  Given correct foresight of which problems will naturally materialize later, one could try to deliberately materialize such problems earlier, and get in some observations of them.  This helps to the extent (a) that we actually correctly forecast all of the problems that will appear later, or some superset of those; (b) that we succeed in preemptively materializing a superset of problems that will appear later; and (c) that we can actually solve, in the earlier laboratory that is out-of-distribution for us relative to the real problems, those alignment problems that would be lethal if we mishandle them when they materialize later.  Anticipating all of the really dangerous ones, and then successfully materializing them, in the correct form for early solutions to generalize over to later solutions, sounds possibly kinda hard.

14.  Some problems, like 'the AGI has an option that (looks to it like) it could successfully kill and replace the programmers to fully optimize over its environment', seem like their natural order of appearance could be that they first appear only in fully dangerous domains.  Really actually having a clear option to brain-level-persuade the operators or escape onto the Internet, build nanotech, and destroy all of humanity - in a way where you're fully clear that you know the relevant facts, and estimate only a not-worth-it low probability of learning something which changes your preferred strategy if you bide your time another month while further growing in capability - is an option that first gets evaluated for real at the point where an AGI fully expects it can defeat its creators.  We can try to manifest an echo of that apparent scenario in earlier toy domains.  Trying to train by gradient descent against that behavior, in that toy domain, is something I'd expect to produce not-particularly-coherent local patches to thought processes, which would break with near-certainty inside a superintelligence generalizing far outside the training distribution and thinking very different thoughts.  Also, programmers and operators themselves, who are used to operating in not-fully-dangerous domains, are operating out-of-distribution when they enter into dangerous ones; our methodologies may at that time break.

15.  Fast capability gains seem likely, and may break lots of previous alignment-required invariants simultaneouslyGiven otherwise insufficient foresight by the operators, I'd expect a lot of those problems to appear approximately simultaneously after a sharp capability gain.  See, again, the case of human intelligence.  We didn't break alignment with the 'inclusive reproductive fitness' outer loss function, immediately after the introduction of farming - something like 40,000 years into a 50,000 year Cro-Magnon takeoff, as was itself running very quickly relative to the outer optimization loop of natural selection.  Instead, we got a lot of technology more advanced than was in the ancestral environment, including contraception, in one very fast burst relative to the speed of the outer optimization loop, late in the general intelligence game.  We started reflecting on ourselves a lot more, started being programmed a lot more by cultural evolution, and lots and lots of assumptions underlying our alignment in the ancestral training environment broke simultaneously.  (People will perhaps rationalize reasons why this abstract description doesn't carry over to gradient descent; eg, “gradient descent has less of an information bottleneck”.  My model of this variety of reader has an inside view, which they will label an outside view, that assigns great relevance to some other data points that are not observed cases of an outer optimization loop producing an inner general intelligence, and assigns little importance to our one data point actually featuring the phenomenon in question.  When an outer optimization loop actually produced general intelligence, it broke alignment after it turned general, and did so relatively late in the game of that general intelligence accumulating capability and knowledge, almost immediately before it turned 'lethally' dangerous relative to the outer optimization loop of natural selection.  Consider skepticism, if someone is ignoring this one warning, especially if they are not presenting equally lethal and dangerous things that they say will go wrong instead.)


Section B.2:  Central difficulties of outer and inner alignment. 

16.  Even if you train really hard on an exact loss function, that doesn't thereby create an explicit internal representation of the loss function inside an AI that then continues to pursue that exact loss function in distribution-shifted environments.  Humans don't explicitly pursue inclusive genetic fitness; outer optimization even on a very exact, very simple loss function doesn't produce inner optimization in that direction This happens in practice in real life, it is what happened in the only case we know about, and it seems to me that there are deep theoretical reasons to expect it to happen again: the first semi-outer-aligned solutions found, in the search ordering of a real-world bounded optimization process, are not inner-aligned solutions.  This is sufficient on its own, even ignoring many other items on this list, to trash entire categories of naive alignment proposals which assume that if you optimize a bunch on a loss function calculated using some simple concept, you get perfect inner alignment on that concept.

17.  More generally, a superproblem of 'outer optimization doesn't produce inner alignment' is that on the current optimization paradigm there is no general idea of how to get particular inner properties into a system, or verify that they're there, rather than just observable outer ones you can run a loss function over.  This is a problem when you're trying to generalize out of the original training distribution, because, eg, the outer behaviors you see could have been produced by an inner-misaligned system that is deliberately producing outer behaviors that will fool you.  We don't know how to get any bits of information into the inner system rather than the outer behaviors, in any systematic or general way, on the current optimization paradigm.

18.  There's no reliable Cartesian-sensory ground truth (reliable loss-function-calculator) about whether an output is 'aligned', because some outputs destroy (or fool) the human operators and produce a different environmental causal chain behind the externally-registered loss function.  That is, if you show an agent a reward signal that's currently being generated by humans, the signal is not in generalreliable perfect ground truth about how aligned an action was, because another way of producing a high reward signal is to deceive, corrupt, or replace the human operators with a different causal system which generates that reward signal.  When you show an agent an environmental reward signal, you are not showing it something that is a reliable ground truth about whether the system did the thing you wanted it to do; even if it ends up perfectly inner-aligned on that reward signal, or learning some concept that exactly corresponds to 'wanting states of the environment which result in a high reward signal being sent', an AGI strongly optimizing on that signal will kill you, because the sensory reward signal was not a ground truth about alignment (as seen by the operators).

19.  More generally, there is no known way to use the paradigm of loss functions, sensory inputs, and/or reward inputs, to optimize anything within a cognitive system to point at particular things within the environment - to point to latent events and objects and properties in the environment, rather than relatively shallow functions of the sense data and reward.  This isn't to say that nothing in the system’s goal (whatever goal accidentally ends up being inner-optimized over) could ever point to anything in the environment by accident Humans ended up pointing to their environments at least partially, though we've got lots of internally oriented motivational pointers as well.  But insofar as the current paradigm works at all, the on-paper design properties say that it only works for aligning on known direct functions of sense data and reward functions.  All of these kill you if optimized-over by a sufficiently powerful intelligence, because they imply strategies like 'kill everyone in the world using nanotech to strike before they know they're in a battle, and have control of your reward button forever after'.  It just isn't true that we know a function on webcam input such that every world with that webcam showing the right things is safe for us creatures outside the webcam.  This general problem is a fact about the territory, not the map; it's a fact about the actual environment, not the particular optimizer, that lethal-to-us possibilities exist in some possible environments underlying every given sense input.

20.  Human operators are fallible, breakable, and manipulable.  Human raters make systematic errors - regular, compactly describable, predictable errors.  To faithfully learn a function from 'human feedback' is to learn (from our external standpoint) an unfaithful description of human preferences, with errors that are not random (from the outside standpoint of what we'd hoped to transfer).  If you perfectly learn and perfectly maximize the referent of rewards assigned by human operators, that kills them.  It's a fact about the territory, not the map - about the environment, not the optimizer - that the best predictive explanation for human answers is one that predicts the systematic errors in our responses, and therefore is a psychological concept that correctly predicts the higher scores that would be assigned to human-error-producing cases.

21.  There's something like a single answer, or a single bucket of answers, for questions like 'What's the environment really like?' and 'How do I figure out the environment?' and 'Which of my possible outputs interact with reality in a way that causes reality to have certain properties?', where a simple outer optimization loop will straightforwardly shove optimizees into this bucket.  When you have a wrong belief, reality hits back at your wrong predictions.  When you have a broken belief-updater, reality hits back at your broken predictive mechanism via predictive losses, and a gradient descent update fixes the problem in a simple way that can easily cohere with all the other predictive stuff.  In contrast, when it comes to a choice of utility function, there are unbounded degrees of freedom and multiple reflectively coherent fixpoints.  Reality doesn't 'hit back' against things that are locally aligned with the loss function on a particular range of test cases, but globally misaligned on a wider range of test cases.  This is the very abstract story about why hominids, once they finally started to generalize, generalized their capabilities to Moon landings, but their inner optimization no longer adhered very well to the outer-optimization goal of 'relative inclusive reproductive fitness' - even though they were in their ancestral environment optimized very strictly around this one thing and nothing else.  This abstract dynamic is something you'd expect to be true about outer optimization loops on the order of both 'natural selection' and 'gradient descent'.  The central result:  Capabilities generalize further than alignment once capabilities start to generalize far.

22.  There's a relatively simple core structure that explains why complicated cognitive machines work; which is why such a thing as general intelligence exists and not just a lot of unrelated special-purpose solutions; which is why capabilities generalize after outer optimization infuses them into something that has been optimized enough to become a powerful inner optimizer.  The fact that this core structure is simple and relates generically to low-entropy high-structure environments is why humans can walk on the Moon.  There is no analogous truth about there being a simple core of alignment, especially not one that is even easier for gradient descent to find than it would have been for natural selection to just find 'want inclusive reproductive fitness' as a well-generalizing solution within ancestral humans.  Therefore, capabilities generalize further out-of-distribution than alignment, once they start to generalize at all.

23.  Corrigibility is anti-natural to consequentialist reasoning; "you can't bring the coffee if you're dead" for almost every kind of coffee.  We (MIRI) tried and failed to find a coherent formula for an agent that would let itself be shut down (without that agent actively trying to get shut down).  Furthermore, many anti-corrigible lines of reasoning like this may only first appear at high levels of intelligence.

24.  There are two fundamentally different approaches you can potentially take to alignment, which are unsolvable for two different sets of reasons; therefore, by becoming confused and ambiguating between the two approaches, you can confuse yourself about whether alignment is necessarily difficult.  The first approach is to build a CEV-style Sovereign which wants exactly what we extrapolated-want and is therefore safe to let optimize all the future galaxies without it accepting any human input trying to stop it.  The second course is to build corrigible AGI which doesn't want exactly what we want, and yet somehow fails to kill us and take over the galaxies despite that being a convergent incentive there.

  1. The first thing generally, or CEV specifically, is unworkable because the complexity of what needs to be aligned or meta-aligned for our Real Actual Values is far out of reach for our FIRST TRY at AGI.  Yes I mean specifically that the dataset, meta-learning algorithm, and what needs to be learned, is far out of reach for our first try.  It's not just non-hand-codable, it is unteachable on-the-first-try because the thing you are trying to teach is too weird and complicated.
  2. The second thing looks unworkable (less so than CEV, but still lethally unworkable) because corrigibility runs actively counter to instrumentally convergent behaviors within a core of general intelligence (the capability that generalizes far out of its original distribution).  You're not trying to make it have an opinion on something the core was previously neutral on.  You're trying to take a system implicitly trained on lots of arithmetic problems until its machinery started to reflect the common coherent core of arithmetic, and get it to say that as a special case 222 + 222 = 555.  You can maybe train something to do this in a particular training distribution, but it's incredibly likely to break when you present it with new math problems far outside that training distribution, on a system which successfully generalizes capabilities that far at all.


Section B.3:  Central difficulties of sufficiently good and useful transparency / interpretability.

25.  We've got no idea what's actually going on inside the giant inscrutable matrices and tensors of floating-point numbers.  Drawing interesting graphs of where a transformer layer is focusing attention doesn't help if the question that needs answering is "So was it planning how to kill us or not?"

26.  Even if we did know what was going on inside the giant inscrutable matrices while the AGI was still too weak to kill us, this would just result in us dying with more dignity, if DeepMind refused to run that system and let Facebook AI Research destroy the world two years later.  Knowing that a medium-strength system of inscrutable matrices is planning to kill us, does not thereby let us build a high-strength system of inscrutable matrices that isn't planning to kill us.

27.  When you explicitly optimize against a detector of unaligned thoughts, you're partially optimizing for more aligned thoughts, and partially optimizing for unaligned thoughts that are harder to detect.  Optimizing against an interpreted thought optimizes against interpretability.

28.  The AGI is smarter than us in whatever domain we're trying to operate it inside, so we cannot mentally check all the possibilities it examines, and we cannot see all the consequences of its outputs using our own mental talent.  A powerful AI searches parts of the option space we don't, and we can't foresee all its options.

29.  The outputs of an AGI go through a huge, not-fully-known-to-us domain (the real world) before they have their real consequences.  Human beings cannot inspect an AGI's output to determine whether the consequences will be good.

30.  Any pivotal act that is not something we can go do right now, will take advantage of the AGI figuring out things about the world we don't know so that it can make plans we wouldn't be able to make ourselves.  It knows, at the least, the fact we didn't previously know, that some action sequence results in the world we want.  Then humans will not be competent to use their own knowledge of the world to figure out all the results of that action sequence.  An AI whose action sequence you can fully understand all the effects of, before it executes, is much weaker than humans in that domain; you couldn't make the same guarantee about an unaligned human as smart as yourself and trying to fool you.  There is no pivotal output of an AGI that is humanly checkable and can be used to safely save the world but only after checking it; this is another form of pivotal weak act which does not exist.

31.  A strategically aware intelligence can choose its visible outputs to have the consequence of deceiving you, including about such matters as whether the intelligence has acquired strategic awareness; you can't rely on behavioral inspection to determine facts about an AI which that AI might want to deceive you about.  (Including how smart it is, or whether it's acquired strategic awareness.)

32.  Human thought partially exposes only a partially scrutable outer surface layer.  Words only trace our real thoughts.  Words are not an AGI-complete data representation in its native style.  The underparts of human thought are not exposed for direct imitation learning and can't be put in any dataset.  This makes it hard and probably impossible to train a powerful system entirely on imitation of human words or other human-legible contents, which are only impoverished subsystems of human thoughts; unless that system is powerful enough to contain inner intelligences figuring out the humans, and at that point it is no longer really working as imitative human thought.

33.  The AI does not think like you do, the AI doesn't have thoughts built up from the same concepts you use, it is utterly alien on a staggering scale.  Nobody knows what the hell GPT-3 is thinking, not only because the matrices are opaque, but because the stuff within that opaque container is, very likely, incredibly alien - nothing that would translate well into comprehensible human thinking, even if we could see past the giant wall of floating-point numbers to what lay behind.


Section B.4:  Miscellaneous unworkable schemes. 

34.  Coordination schemes between superintelligences are not things that humans can participate in (eg because humans can't reason reliably about the code of superintelligences); a "multipolar" system of 20 superintelligences with different utility functions, plus humanity, has a natural and obvious equilibrium which looks like "the 20 superintelligences cooperate with each other but not with humanity".

35.  Schemes for playing "different" AIs off against each other stop working if those AIs advance to the point of being able to coordinate via reasoning about (probability distributions over) each others' code.  Any system of sufficiently intelligent agents can probably behave as a single agent, even if you imagine you're playing them against each other.  Eg, if you set an AGI that is secretly a paperclip maximizer, to check the output of a nanosystems designer that is secretly a staples maximizer, then even if the nanosystems designer is not able to deduce what the paperclip maximizer really wants (namely paperclips), it could still logically commit to share half the universe with any agent checking its designs if those designs were allowed through, if the checker-agent can verify the suggester-system's logical commitment and hence logically depend on it (which excludes human-level intelligences).  Or, if you prefer simplified catastrophes without any logical decision theory, the suggester could bury in its nanosystem design the code for a new superintelligence that will visibly (to a superhuman checker) divide the universe between the nanosystem designer and the design-checker.

36.  What makes an air conditioner 'magic' from the perspective of say the thirteenth century, is that even if you correctly show them the design of the air conditioner in advance, they won't be able to understand from seeing that design why the air comes out cold; the design is exploiting regularities of the environment, rules of the world, laws of physics, that they don't know about.  The domain of human thought and human brains is very poorly understood by us, and exhibits phenomena like optical illusions, hypnosis, psychosis, mania, or simple afterimages produced by strong stimuli in one place leaving neural effects in another place.  Maybe a superintelligence couldn't defeat a human in a very simple realm like logical tic-tac-toe; if you're fighting it in an incredibly complicated domain you understand poorly, like human minds, you should expect to be defeated by 'magic' in the sense that even if you saw its strategy you would not understand why that strategy worked.  AI-boxing can only work on relatively weak AGIs; the human operators are not secure systems.


Section C:

Okay, those are some significant problems, but lots of progress is being made on solving them, right?  There's a whole field calling itself "AI Safety" and many major organizations are expressing Very Grave Concern about how "safe" and "ethical" they are?


37.  There's a pattern that's played out quite often, over all the times the Earth has spun around the Sun, in which some bright-eyed young scientist, young engineer, young entrepreneur, proceeds in full bright-eyed optimism to challenge some problem that turns out to be really quite difficult.  Very often the cynical old veterans of the field try to warn them about this, and the bright-eyed youngsters don't listen, because, like, who wants to hear about all that stuff, they want to go solve the problem!  Then this person gets beaten about the head with a slipper by reality as they find out that their brilliant speculative theory is wrong, it's actually really hard to build the thing because it keeps breaking, and society isn't as eager to adopt their clever innovation as they might've hoped, in a process which eventually produces a new cynical old veteran.  Which, if not literally optimal, is I suppose a nice life cycle to nod along to in a nature-show sort of way.  Sometimes you do something for the first time and there are no cynical old veterans to warn anyone and people can be really optimistic about how it will go; eg the initial Dartmouth Summer Research Project on Artificial Intelligence in 1956:  "An attempt will be made to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves. We think that a significant advance can be made in one or more of these problems if a carefully selected group of scientists work on it together for a summer."  This is less of a viable survival plan for your planet if the first major failure of the bright-eyed youngsters kills literally everyone before they can predictably get beaten about the head with the news that there were all sorts of unforeseen difficulties and reasons why things were hard.  You don't get any cynical old veterans, in this case, because everybody on Earth is dead.  Once you start to suspect you're in that situation, you have to do the Bayesian thing and update now to the view you will predictably update to later: realize you're in a situation of being that bright-eyed person who is going to encounter Unexpected Difficulties later and end up a cynical old veteran - or would be, except for the part where you'll be dead along with everyone else.  And become that cynical old veteran right away, before reality whaps you upside the head in the form of everybody dying and you not getting to learn.  Everyone else seems to feel that, so long as reality hasn't whapped them upside the head yet and smacked them down with the actual difficulties, they're free to go on living out the standard life-cycle and play out their role in the script and go on being bright-eyed youngsters; there's no cynical old veterans to warn them otherwise, after all, and there's no proof that everything won't go beautifully easy and fine, given their bright-eyed total ignorance of what those later difficulties could be.

38.  It does not appear to me that the field of 'AI safety' is currently being remotely productive on tackling its enormous lethal problems.  These problems are in fact out of reach; the contemporary field of AI safety has been selected to contain people who go to work in that field anyways.  Almost all of them are there to tackle problems on which they can appear to succeed and publish a paper claiming success; if they can do that and get funded, why would they embark on a much more unpleasant project of trying something harder that they'll fail at, just so the human species can die with marginally more dignity?  This field is not making real progress and does not have a recognition function to distinguish real progress if it took place.  You could pump a billion dollars into it and it would produce mostly noise to drown out what little progress was being made elsewhere.

39.  I figured this stuff out using the null string as input, and frankly, I have a hard time myself feeling hopeful about getting real alignment work out of somebody who previously sat around waiting for somebody else to input a persuasive argument into them.  This ability to "notice lethal difficulties without Eliezer Yudkowsky arguing you into noticing them" currently is an opaque piece of cognitive machinery to me, I do not know how to train it into others.  It probably relates to 'security mindset', and a mental motion where you refuse to play out scripts, and being able to operate in a field that's in a state of chaos.

40.  "Geniuses" with nice legible accomplishments in fields with tight feedback loops where it's easy to determine which results are good or bad right away, and so validate that this person is a genius, are (a) people who might not be able to do equally great work away from tight feedback loops, (b) people who chose a field where their genius would be nicely legible even if that maybe wasn't the place where humanity most needed a genius, and (c) probably don't have the mysterious gears simply because they're rare.  You cannot just pay $5 million apiece to a bunch of legible geniuses from other fields and expect to get great alignment work out of them.  They probably do not know where the real difficulties are, they probably do not understand what needs to be done, they cannot tell the difference between good and bad work, and the funders also can't tell without me standing over their shoulders evaluating everything, which I do not have the physical stamina to do.  I concede that real high-powered talents, especially if they're still in their 20s, genuinely interested, and have done their reading, are people who, yeah, fine, have higher probabilities of making core contributions than a random bloke off the street. But I'd have more hope - not significant hope, but more hope - in separating the concerns of (a) credibly promising to pay big money retrospectively for good work to anyone who produces it, and (b) venturing prospective payments to somebody who is predicted to maybe produce good work later.

41.  Reading this document cannot make somebody a core alignment researcherThat requires, not the ability to read this document and nod along with it, but the ability to spontaneously write it from scratch without anybody else prompting you; that is what makes somebody a peer of its author.  It's guaranteed that some of my analysis is mistaken, though not necessarily in a hopeful direction.  The ability to do new basic work noticing and fixing those flaws is the same ability as the ability to write this document before I published it, which nobody apparently did, despite my having had other things to do than write this up for the last five years or so.  Some of that silence may, possibly, optimistically, be due to nobody else in this field having the ability to write things comprehensibly - such that somebody out there had the knowledge to write all of this themselves, if they could only have written it up, but they couldn't write, so didn't try.  I'm not particularly hopeful of this turning out to be true in real life, but I suppose it's one possible place for a "positive model violation" (miracle).  The fact that, twenty-one years into my entering this death game, seven years into other EAs noticing the death game, and two years into even normies starting to notice the death game, it is still Eliezer Yudkowsky writing up this list, says that humanity still has only one gamepiece that can do that.  I knew I did not actually have the physical stamina to be a star researcher, I tried really really hard to replace myself before my health deteriorated further, and yet here I am writing this.  That's not what surviving worlds look like.

42.  There's no plan.  Surviving worlds, by this point, and in fact several decades earlier, have a plan for how to survive.  It is a written plan.  The plan is not secret.  In this non-surviving world, there are no candidate plans that do not immediately fall to Eliezer instantly pointing at the giant visible gaping holes in that plan.  Or if you don't know who Eliezer is, you don't even realize you need a plan, because, like, how would a human being possibly realize that without Eliezer yelling at them?  It's not like people will yell at themselves about prospective alignment difficulties, they don't have an internal voice of caution.  So most organizations don't have plans, because I haven't taken the time to personally yell at them.  'Maybe we should have a plan' is deeper alignment mindset than they possess without me standing constantly on their shoulder as their personal angel pleading them into... continued noncompliance, in fact.  Relatively few are aware even that they should, to look better, produce a pretend plan that can fool EAs too 'modest' to trust their own judgments about seemingly gaping holes in what serious-looking people apparently believe.

43.  This situation you see when you look around you is not what a surviving world looks like.  The worlds of humanity that survive have plans.  They are not leaving to one tired guy with health problems the entire responsibility of pointing out real and lethal problems proactively.  Key people are taking internal and real responsibility for finding flaws in their own plans, instead of considering it their job to propose solutions and somebody else's job to prove those solutions wrong.  That world started trying to solve their important lethal problems earlier than this.  Half the people going into string theory shifted into AI alignment instead and made real progress there.  When people suggest a planetarily-lethal problem that might materialize later - there's a lot of people suggesting those, in the worlds destined to live, and they don't have a special status in the field, it's just what normal geniuses there do - they're met with either solution plans or a reason why that shouldn't happen, not an uncomfortable shrug and 'How can you be sure that will happen' / 'There's no way you could be sure of that now, we'll have to wait on experimental evidence.'

A lot of those better worlds will die anyways.  It's a genuinely difficult problem, to solve something like that on your first try.  But they'll die with more dignity than this.

New Comment
705 comments, sorted by Click to highlight new comments since:
Some comments are truncated due to high volume. (⌘F to expand all)Change truncation settings

That requires, not the ability to read this document and nod along with it, but the ability to spontaneously write it from scratch without anybody else prompting you; that is what makes somebody a peer of its author. It's guaranteed that some of my analysis is mistaken, though not necessarily in a hopeful direction. The ability to do new basic work noticing and fixing those flaws is the same ability as the ability to write this document before I published it, which nobody apparently did, despite my having had other things to do than write this up for the last five years or so. Some of that silence may, possibly, optimistically, be due to nobody else in this field having the ability to write things comprehensibly - such that somebody out there had the knowledge to write all of this themselves, if they could only have written it up, but they couldn't write, so didn't try. I'm not particularly hopeful of this turning out to be true in real life, but I suppose it's one possible place for a "positive model violation" (miracle). The fact that, twenty-one years into my entering this death game, seven years into other EAs noticing the death game, and two years into even normies start

... (read more)

I agree this list doesn't seem to contain much unpublished material, and I think the main value of having it in one numbered list is that "all of it is in one, short place", and it's not an "intro to computers can think" and instead is "these are a bunch of the reasons computers thinking is difficult to align".

The thing that I understand to be Eliezer's "main complaint" is something like: "why does it seem like No One Else is discovering new elements to add to this list?". Like, I think Risks From Learned Optimization was great, and am glad you and others wrote it! But also my memory is that it was "prompted" instead of "written from scratch", and I imagine Eliezer reading it more had the sense of "ah, someone made 'demons' palatable enough to publish" instead of "ah, I am learning something new about the structure of intelligence and alignment."

[I do think the claim that Eliezer 'figured it out from the empty string' doesn't quite jive with the Yudkowsky's Coming of Age sequence.]

9Eliezer Yudkowsky
Nearly empty string of uncommon social inputs.  All sorts of empirical inputs, including empirical inputs in the social form of other people observing things. It's also fair to say that, though they didn't argue me out of anything, Moravec and Drexler and Ed Regis and Vernor Vinge and Max More could all be counted as social inputs telling me that this was an important thing to look at.

Thank you, Evan, for living the Virture of Scholarship. Your work is appreciated. 

Eliezer's post here is doing work left undone by the writing you cite. It is a much clearer account of how our mainline looks doomed than you'd see elsewhere, and it's frank on this point.

I think Eliezer wishes these sorts of artifacts were not just things he wrote, like this and "There is no fire alarm".

Also, re your excerpts for (14), (15), and (32), I see Eliezer as saying something meaningfully different in each case. I might elaborate under this comment.

Re (14), I guess the ideas are very similar, where the mesaoptimizer scenario is like a sharp example of the more general concept Eliezer points at, that different classes of difficulties may appear at different capability levels.

Re (15), "Fast capability gains seem likely, and may break lots of previous alignment-required invariants simultaneously", which is about how we may have reasons to expect aligned output that are brittle under rapid capability gain: your quote from Richard is just about "fast capability gain seems possible and likely", and isn't about connecting that to increased difficulty in succeeding at the alignment problem?

Re (32), I don't think your quote isn't talking about the thing Eliezer is talking about, which is that in order to be human level at modelling human-generated text, your AI must be doing something on par with human thought that figures out what humans would say. Your quote just isn't discussing this, namely that strong imitation requires cognition that is dangerous.

So I guess I don't take much issue with (14) or (15), but I think you're quite off the mark about (32). In any case, I still have a strong sense that Eliezer is successfully being more on the mark here than the rest of us manage. Kudos of course to you and others that are working on writing things up and figuring things out. Though I remain sympathetic to Eliezer's complaint.

7Eliezer Yudkowsky
Well, my disorganized list sure wasn't complete, so why not go ahead and list some of the foreseeable difficulties I left out?  Bonus points if any of them weren't invented by me, though I realize that most people may not realize how much of this entire field is myself wearing various trenchcoats.

Sure—that's easy enough. Just off the top of my head, here's five safety concerns that I think are important that I don't think you included:

  • The fact that there exist functions that are easier to verify than satisfy ensures that adversarial training can never guarantee the absence of deception.

  • It is impossible to verify a model's safety—even given arbitrarily good transparency tools—without access to that model's training process. For example, you could get a deceptive model that gradient hacks itself in such a way that cryptographically obfuscates its deception.

  • It is impossible in general to use interpretability tools to select models to have a particular behavioral property. I think this is clear if you just stare at Rice's theorem enough: checking non-trivial behavioral properties, even with mechanistic access, is in general undecidable. Note, however, that this doesn't rule out checking a mechanistic property that implies a behavioral property.

  • Any prior you use to incentivize models to behave in a particular way doesn't necessarily translate to situations where that model itself runs another search over algorithms. For example, the fastest way to search for algorith

... (read more)

Consider my vote to be placed that you should turn this into a post, keep going for literally as long as you can, expand things to paragraphs, and branch out beyond things you can easily find links for.

(I do think there's a noticeable extent to which I was trying to list difficulties more central than those, but I also think many people could benefit from reading a list of 100 noncentral difficulties.)

I do think there's a noticeable extent to which I was trying to list difficulties more central than those

Probably people disagree about which things are more central, or as evhub put it:

Every time anybody writes up any overview of AI safety, they have to make tradeoffs [...] depending on what the author personally believes is most important/relevant to say

Now FWIW I thought evhub was overly dismissive of (4) in which you made an important meta-point:

EY: 4. We can't just "decide not to build AGI" because GPUs are everywhere, and knowledge of algorithms is constantly being improved and published; 2 years after the leading actor has the capability to destroy the world, 5 other actors will have the capability to destroy the world.  The given lethal challenge is to solve within a time limit, driven by the dynamic in which, over time, increasingly weak actors with a smaller and smaller fraction of total computing power, become able to build AGI and destroy the world.  Powerful actors all refraining in unison from doing the suicidal thing just delays this time limit - it does not lift it [...]

evhub: This is just answering a particular bad plan.

But I would add a criticism of my o... (read more)

(Note that these have a theme: you can't wrangle general computation / optimization. That's why I'm short universal approaches to AI alignment (approaches that aim at making general optimization safe by enforcing universal rules), and long existential approaches (approaches that try to find specific mechanisms that can be analytically seen to do the right thing).)

Eliezer: If you find that (for reasons still left explained) * ... selection of code for intentionality is coupled – over the long run, in mostly non-reverse-engineerable ways – to various/most of the physical/chemical properties  * ... of the molecular substrate through which discrete code is necessarily computed/expressed (via input and output channels of information/energy packet transmission), then given that  * ... the properties of the solid-state substrate (e.g. silicon-based hardware) computing AGI's code * ... differ from the properties of the substrate of humans (carbon-based wetware), a conclusion that follows is that * ... the intentionality being selected for in AGI over the long run * ... will diverge from the intentionality that was selected for in humans.
2Rob Bensinger
What do you mean by 'intentionality'? Per SEP, "In philosophy, intentionality is the power of minds and mental states to be about, to represent, or to stand for, things, properties and states of affairs." So I read your comment as saying, a la Searle, 'maybe AI can never think like a human because there's something mysterious and crucial about carbon atoms in particular, or about capital-b Biology, for doing reasoning.' This seems transparently silly to me -- I know of no reasonable argument for thinking carbon differs from silicon on this dimension -- and also not relevant to AGI risk. You can protest "but AlphaGo doesn't really understand Go!" until the cows come home, and it will still beat you at Go. You can protest "but you don't really understand killer nanobots!" until the cows come home, and superintelligent Unfriendly AI will still build the nanobots and kill you with them. By the same reasoning, Searle-style arguments aren't grounds for pessimism either. If Friendly AI lacks true intentionality or true consciousness or whatever, it can still do all the same mechanistic operations, and therefore still produce the same desirable good outcomes as if it had human-style intentionality or whatver.
That’s not the argument. Give me a few days to write a response. There’s a minefield of possible misinterpretations here. However, the argumentation does undermine the idea that designing for mechanistic (alignment) operations is going to work. I’ll try and explain why.
BTW, with ‘intentionality’, I meant something closer to everyday notions of ‘intentions one has’. Will more precisely define that meaning later. I should have checked for diverging definitions from formal fields. Thanks for catching that.
If you happen to have time, this paper serves as useful background reading: Particularly note the shift from trivial self-replication (e.g. most computer viruses) to non-trivial self-replication (e.g. as through substrate-environment pathways to reproduction). None of this is sufficient for you to guess what the argumentation is (you might be able to capture a bit of it, along with a lot of incorrect and often implicit assumptions we must dig into). If you could call on some patience and openness to new ideas, I would really appreciate it! I am already bracing for a next misinterpretation (which is fine, if we can talk about that). I apologise for that I cannot find a viable way yet to throw out all the argumentation in one go, and also for that this will get a bit disorientating when we go through arguments step-by-step.
Returning to this:   Key idea: Different basis of existence→ different drives→ different intentions→ different outcomes. @Rob, I wrote up a longer explanation here, which I prefer to discuss with you in private first.  Will email you a copy tomorrow in the next weeks.

I'm sorry to hear that your health is poor and you feel that this is all on you. Maybe you're right about the likelihood of doom, and even if I knew you were, I'd be sorry that it troubles you this way.

I think you've done an amazing job of building the AI safety field and now, even when the field has a degree of momentum of its own, it does seem to be less focused on doom than it should be, and I think you continuing to push people to focus on doom is valuable.

I don't think its easy to get people to take weird ideas seriously. I've had many experiences where I've had ideas about how people should change their approach to a project that weren't particularly far out and (in my view) were right for very straightforward reasons, and yet for the most part I was ignored altogether. What you've accomplished in building the AI safety field is amazing because AI doom ideas seemed really crazy when you started talking about them.

Nevertheless, I think some of the things you've said in this post are counterproductive. Most of the post is good, but insulting people who might contribute to solving the problem is not, nor is demanding that people acknowledge that you are smarter than they are. I'... (read more)

There's a point here about how fucked things are that I do not know how to convey without saying those things, definitely not briefly or easily.  I've spent, oh, a fair number of years, being politer than this, and less personal than this, and the end result is that people nod along and go on living their lives.

I expect this won't work either, but at some point you start trying different things instead of the things that have already failed.  It's more dignified if you fail in different ways instead of the same way.


FWIW you taking off the Mr. Nice guy gloves has actually made me make different life decisions. I'm glad you tried it even if it doesn't work.

Do whatever you want, obviously, but I just want to clarify that I did not suggest you avoid personally criticising people (only that you avoid vague/hard to interpret criticism) or saying you think doom is overwhelmingly likely. Some other comments give me a stronger impression than yours that I was asking you in a general sense to be nice, but I'm saying it to you because I figure it mostly matters that you're clear on this.

You might not have this ability, but surely you know at least one person who does?

I vehemently disagree here, based on my personal and generalizable or not history. I will illustrate with the three turning points of my recent life.

First step: I stumbled upon HPMOR, and Eliezer way of looking straight into the irrationality of all our common ways of interacting and thinking was deeply shocking. It made me feel like he was in a sense angrily pointing at me, who worked more like one of the PNJ rather than Harry. I heard him telling me you're dumb and all your ideals of making intelligent decisions, being the gifted kid and being smarter than everyone are all are just delusions. You're so out of touch with reality on so many levels, where to even start.

This attitude made me embark on a journey to improve myself, read the sequences, pledge on Giving What we can after knowing EA for many years, and overall reassess whether I was striving towards my goal of helping people (spoiler: I was not).

Second step:  The April fools post also shocked me on so many levels. I was once again deeply struck by the sheer pessimism of this figure I respected so much. After months of reading articles on LessWrong and so many about AI alignment, this was the one that made me terrifie... (read more)

I disagree strongly. To me it seems that AI safety has long punched below its weight because its proponents are unwilling to be confrontational, and are too reluctant to put moderate social pressure on people doing the activities which AI safety proponents hold to be very extremely bad. It is not a coincidence that among AI safety proponents, Eliezer is both unusually confrontational and unusually successful.

This isn't specific to AI safety. A lot of people in this community generally believe that arguments which make people feel bad are counterproductive because people will be "turned off".

This is false. There are tons of examples of disparaging arguments against bad (or "bad") behavior that succeed wildly. Such arguments very frequently succeed in instilling individual values like e.g. conscientiousness or honesty. Prominent political movements which use this rhetoric abound. When this website was young, Eliezer and many others participated in an aggressive campaign of discourse against religious ideas, and this campaign accomplished many of its goals. I could name many many more large and small examples. I bet you can too.

Obviously this isn't to say that confrontational and insu... (read more)

I think there's an important distinction between:

  • Deliberately phrasing things in confrontational or aggressive ways, in the hope that this makes your conversation partner "wake up" or something.
  • Choosing not to hide real, potentially-important beliefs you have about the world, even though those beliefs are liable to offend people, liable to be disagreed with, etc.

Either might be justifiable, but I'm a lot more wary of heuristics like "it's never OK to talk about individuals' relative proficiency at things, even if it feels very cruxy and important, because people just find the topic too triggering" than of heuristics like "it's never OK to say things in ways that sound shouty or aggressive". I think cognitive engines can much more easily get by self-censoring their tone than self-censoring what topics are permissible to think or talk about.

How is "success" measured among AI safety proponents?

This kind of post scares away the person who will be the key person in the AI safety field if we define "key person" as the genius main driver behind solving it, not the loudest person.  Which is rather unfortunate, because that person is likely to read this post at some point.

I don't believe this post has any "dignity", whatever weird obscure definition dignity has been given now. It's more like flailing around in death throes while pointing fingers and lauding yourself than it is a solemn battle stance against an oncoming impossible enemy.

For context, I'm not some Eliezer hater, I'm a young person doing an ML masters currently who just got into this space and within the past week have become a huge fan of Eliezer Yudkowsky's earlier work while simultaneously very disappointed in the recent, fruitless, output.

It seems worth doing a little user research on this to see how it actually affects people. If it is a net positive, then great. If it is a net negative, the question becomes how big of a net negative it is and whether it is worth the extra effort to frame things more nicely.

4Eli Tyre
I think this was excellently worded, and I'm glad you said it. I'm also glad to have read all the responses, many of which seem important and on point to me. I strong upvoted this comment as well as several of the responses. I'm leaving this comment, because I want to give you some social reinforcement for saying what you said, and saying it as clearly and tactfully as you did. 
Strongly agree with this, said more eloquently than I was able to :)

I'd have more hope - not significant hope, but more hope - in separating the concerns of (a) credibly promising to pay big money retrospectively for good work to anyone who produces it, and (b) venturing prospective payments to somebody who is predicted to maybe produce good work later.


I desperately want to make this ecosystem exist, either as part of Manifold Markets, or separately. Some people call it "impact certificates" or "retroactive public goods funding"; I call it "equity for public goods", or "Manifund" in the specific case.

If anyone is interested in:

a) Being a retroactive funder for good work (aka bounties, prizes)

b) Getting funding through this kind of mechanism (aka income share agreements, angel investment)

c) Working on this project full time (full-stack web dev, ops, community management)

Please get in touch! Reply here, or message

I'm also on a team trying to build impact certificates/retroactive public goods funding and we are receiving a grant from an FTX Future Fund regrantor to make it happen!

If you're interested in learning more or contributing you can:

  • Read about our ongoing $10,000 retro-funding contest (Austin is graciously contributing to the prize pool)
  • Submit an EA Forum Post to this retro-funding contest (before July 1st)
  • Join our Discord to chat/ask questions
  • Read/Comment on our lengthy informational EA forum post "Towards Impact Markets"
[-]Matthew BarnettΩ35142-26

It's as good as time as any to re-iterate my reasons for disagreeing with what I see as the Yudkowskian view of future AI. What follows isn't intended as a rebuttal of any specific argument in this essay, but merely a pointer that I'm providing for readers, that may help explain why some people might disagree with the conclusion and reasoning contained within.

I'll provide my cruxes point-by-point,

  • I think raw intelligence, while important, is not the primary factor that explains why humanity-as-a-species is much more powerful than chimpanzees-as-a-species. Notably, humans were once much less powerful, in our hunter-gatherer days, but over time, through the gradual process of accumulating technology, knowledge, and culture, humans now possess vast productive capacities that far outstrip our ancient powers.

    Similarly, our ability to coordinate through language also plays a huge role in explaining our power compared to other animals. But, on a first approximation, other animals can't coordinate at all, making this distinction much less impressive. The first AGIs we construct will be born into a culture already capable of coordinating, and sharing knowledge, making the potential power di
... (read more)

Notably, humans were once much less powerful, in our hunter-gatherer days, but over time, through the gradual process of accumulating technology, knowledge, and culture, humans now possess vast productive capacities that far outstrip our ancient powers.

Similarly, our ability to coordinate through language also plays a huge role in explaining our power compared to other animals. But, on a first approximation, other animals can't coordinate at all, making this distinction much less impressive. The first AGIs we construct will be born into a culture already capable of coordinating, and sharing knowledge, making the potential power difference between AGI and humans relatively much smaller than between humans and other animals, at least at first.

I basically buy the story that human intelligence is less useful that human coordination; i.e. it's the intelligence of "humanity" the entity that matters, with the intelligence of individual humans relevant only as, like, subcomponents of that entity.

But... shouldn't this mean you expect AGI civilization to totally dominate human civilization? They can read each other's source code, and thus trust much more deeply! They can transmit information... (read more)


But... shouldn't this mean you expect AGI civilization to totally dominate human civilization? They can read each other's source code, and thus trust much more deeply! They can transmit information between them at immense bandwidths! They can clone their minds and directly learn from each other's experiences!

This is 100% correct, and part of why I expect the focus on superintelligence, while literally true, is bad for AI outreach. There's a much simpler (and empirically, in my experience, more convincing) explanation of why we lose to even an AI with an IQ of 110. It is Dath Ilan, and we are Earth. Coordination is difficult for humans and the easy part for AIs. 

I will note that Eliezer wrote That Alien Message a long time ago I think in part to try to convey the issue to this perspective, but it's mostly about "information-theoretic bounds are probably not going to be tight" in a simulation-y universe instead of "here's what coordination between computers looks like today". I do predict the coordination point would be good to include in more of the intro materials.


But... shouldn't this mean you expect AGI civilization to totally dominate human civilization? They can read each other's source code, and thus trust much more deeply! They can transmit information between them at immense bandwidths! They can clone their minds and directly learn from each other's experiences!

I don't think it's obvious that this means that AGI is more dangerous, because it means that for a fixed total impact of AGI, the AGI doesn't have to be as competent at individual thinking (because it leans relatively more on group thinking). And so at the point where the AGIs are becoming very powerful in aggregate, this argument pushes us away from thinking they're good at individual thinking.

Also, it's not obvious that early AIs will actually be able to do this if their creators don't find a way to train them to have this affordance. ML doesn't currently normally make AIs which can helpfully share mind-states, and it probably requires non-trivial effort to hook them up correctly to be able to share mind-state.

6Anthony DiGiovanni
Being able to read source code doesn't automatically increase trust—you also have to be able to verify that the code being shared with you actually governs the AGI's behavior, despite that AGI's incentives and abilities to fool you. (Conditional on the AGIs having strongly aligned goals with each other, sure, this degree of transparency would help them with pure coordination problems.)

Nice! Thanks! I'll give my commentary on your commentary, also point by point. Your stuff italicized, my stuff not. Warning: Wall of text incoming! :)

I think raw intelligence, while important, is not the primary factor that explains why humanity-as-a-species is much more powerful than chimpanzees-as-a-species. Notably, humans were once much less powerful, in our hunter-gatherer days, but over time, through the gradual process of accumulating technology, knowledge, and culture, humans now possess vast productive capacities that far outstrip our ancient powers.

Similarly, our ability to coordinate through language also plays a huge role in explaining our power compared to other animals. But, on a first approximation, other animals can't coordinate at all, making this distinction much less impressive. The first AGIs we construct will be born into a culture already capable of coordinating, and sharing knowledge, making the potential power difference between AGI and humans relatively much smaller than between humans and other animals, at least at first.

I don't think I understand this argument. Yes, humans can use language to coordinate & benefit from cultural evolution, so an AI that... (read more)

5Chris van Merwijk
"I have sat down to make toy models .." reference?
7Daniel Kokotajlo
? I am the reference, I'm describing a personal experience.
1Chris van Merwijk
I meant, is there a link to where you've written this down somewhere? Maybe you just haven't written it down. 
2Daniel Kokotajlo
I'll send you a DM.
Markdown has syntax for quotes: a line with > this on it will look like
3Paul Kent
  Can I get a link or two to read more about this incident?
2Daniel Kokotajlo
It's not so much an incident as a trend. I haven't investigated it myself, but I've read lots of people making this claim, citing various studies, etc. See e.g. "The social dilemma" by Tristan Harris. There's an academic literature on the subject now which I haven't read but which you can probably find by googling. I just did a quick search and found graphs like this: Presumably not all of the increase in deaths is due to Facebook; presumably it's multi-causal blah blah blah. But even if Facebook is responsible for a tiny fraction of the increase, that would mean Facebook was responsible for thousands of deaths.

You said you weren't replying to any specific point Eliezer was making, but I think it's worth pointing out that when he brings up Alpha Go, he's not talking about the 2 years it took Google to build a Go-playing AI - remarkable and surprising as that was - but rather the 3 days it took Alpha Zero to go from not knowing anything about the game beyond the basic rules to being better than all humans and the earlier AIs.


I hate how convincing so many different people are. I wish I just had some fairly static, reasoned perspective based on object-level facts and not persuasion strings.

[This comment is no longer endorsed by its author]Reply
Note that convincing is a 2-place word. I don't think I can transfer this ability, but I haven't really tried, so here's a shot: The target is: "reading as dialogue." Have a world-model. As you read someone else, be simultaneously constructing / inferring "their world-model" and holding "your world-model", noting where you agree and disagree. If you focus too much on "how would I respond to each line", you lose the ability to listen and figure out what they're actually pointing at. If you focus too little on "how would I respond to this", you lose the ability to notice disagreements, holes, and notes of discord. The first homework exercise I'd try to printing out something (probably with double-spacing), and writing your thoughts each sentence. "uh huh", "wait what?", "yes and", "no but", etc.; at the beginning you're probably going to be alternating between the two moves before you can do them simultaneously. [Historically, I think I got this both from 'reading a lot', including a lot of old books, and also 'arguing on the internet' in forum environments that only sort of exist today, which was a helpful feedback loop for the relevant subskills, and of course whatever background factors made me do those activities.]
Why can't I delete comments sometimes? >:(
Users can't delete their own comments if the comment has been replied to, to avoid disrupting other people's content. (you can edit it to be blank though, or mark it as retracted)

Some quick thoughts on these points:

  • I think the ability for humans to communicate and coordinate is a double edged sword. In particular, it enables the attack vector of dangerous self propagating memes. I expect memetic warfare to play a major role in many of the failure scenarios I can think of. As we've seen, even humans are capable of crafting some pretty potent memes, and even defending against human actors is difficult.
  • I think it's likely that the relevant reference class here is research bets rather then the "task" of AGI. An extremely successful research bet could be currently underinvested in, but once it shows promise, discontinuous (relative to the bet) amounts of resources will be dumped into scaling it up, even if the overall investment towards the task as a whole remains continuous. In other words, in this case even though investment into AGI may be continuous (though that might not even hold), discontinuity can occur on the level of specific research bets. Historical examples would include imagenet seeing discontinuous improvement with AlexNet despite continuous investment into image recognition to that point. (Also, for what it's worth, my personal model of AI doo
... (read more)
9emanuele ascani
Thanks a lot for writing this.  These disagreements mainly concern the relative power of future AIs, the polarity of takeoff, takeoff speed, and, in general, the shape of future AIs. Do you also have detailed disagreements about the difficulty of alignment? If anything, the fact that the future unfolds differently in your view should impact future alignment efforts (but you also might have other considerations informing your view on alignment). You partially answer this in the last point, saying: "But, equally, one could view these theses pessimistically." But what do you personally think? Are you more pessimistic, more optimistic, or equally pessimistic about humanity's chances of surviving AI progress? And why?

Part of what makes it difficult for me to talk about alignment difficultly is that the concept doesn’t fit easily into my paradigm of thinking about the future of AI. If I am correct, for example, that AI services will be modular, marginally more powerful than what comes before, and numerous as opposed to monolithic, then there will not be one alignment problem, but many.

I could talk about potential AI safety principles, healthy cultural norms, and specific engineering issues, but not “a problem” called “aligning the AI” — a soft prerequisite for explaining how difficult “the problem” will be. Put another way, my understanding is that future AI alignment will be continuous with ordinary engineering, like cars and skyscrapers. We don’t ordinarily talk about how hard the problem of building a car is, in some sort of absolute sense, though there are many ways of operationalizing what that could mean.

One question is how costly it is to build a car. We could then compare that cost to the overall consumer benefit that people get from cars, and from that, deduce whether and how many cars will be built. Similarly, we could ask about the size of the “alignment tax” (the cost of aligning an ... (read more)

Btw, your top-level comment is one of the best comments I've come across ever. Probably. Top 5? Idk, I'll check how I feel tomorrow. Aspiring to read everything you've ever written rn. Incidentally, you mention that And I've been thinking lately about how important it is to prioritise original thinking before you've consumed all the established literature in an active field of research.[1] If you manage to diverge early, the novelty of your perspective compounds over time (feel free to ask about my model) and you're more likely to end up with a productively different paradigm from what's already out there. Did you ever feel embarrassed trying to think for yourself when you didn't feel like you had read enough? Or, did you feel like other people might have expected you to feel embarrassed for how seriously you took your original thoughts, given how early you were in your learning arc? 1. ^ I'm not saying you haven't. I'm just guessing that you acquired your paradigm by doing original thinking early, and thus had the opportunity to diverge early, rather than greedily over-prioritising the consumption of existing literature in order to "reach the frontier". Once having hastily consumed someone else's paradigm, it's much harder to find its flaws and build something else from the ground up.
5Vishrut Arya
hi Matt! on the coordination crux, you say  but wouldn’t an AGI be able to coordinate and do knowledge sharing with humans because  a) it can impersonate being a human online and communicate with them via text and speech and  b) it‘ll realize such coordination is vital to accomplish it‘s goals and so it’ll do the necessary acculturation?  Watching all the episodes of Friends or reading all the social media posts by the biggest influencers, as examples. 
One reason that a fully general AGI might be more profitable than specialised AIs, despite obvious gains-from-specialisation, is if profitability depends on insight-production. For humans, it's easier to understand a particular thing the more other things you understand. One of the main ways you make novel intellectual progress is by combining remote associations from models about different things. Insight-ability for a particular novel task grows with the number of good models you have available to draw connections between. But, it could still be that the gains from increased generalisation for a particular model grows too slowly and can't compete with obvious gains from specialised AIs.
2David Johnston
Slightly relatedly, I think it's possible that "causal inference is hard". The idea is: once someone has worked something out, they can share it and people can pick it up easily, but it's hard to figure the thing out to begin with - even with a lot of prior experience and efficient inference, most new inventions still need a lot of trial and error. Thus the reason the process of technology accumulation is gradual is, crudely, because causal inference is hard. Even if this is true, one way things could still go badly is if most doom scenarios are locked behind a bunch of hard trial and error, but the easiest one isn't. On the other hand, if both of these things are true then there could be meaningful safety benefits gained from censoring certain kinds of data.
This is what struck me as the least likely to be true from the above AI doom scenario. Is diamondoid nanotechnology possible?  Very likely it is or something functionally equivalent.   Can a sufficiently advanced superintelligence infer how to build it from scratch solely based on human data?  Or will it need a large R&D center with many, many robotic systems that conduct experiments in parallel to extract the information required about our specific details of physics in our actual universe.  Not the very slightly incorrect approximations a simulator will give you.   The 'huge R&D center so big you can't see the end of it' is somewhat easier to regulate the 'invisible dust the AI assembles with clueless stooges'.
8Marion Z.
Any individual doomsday mechanism we can think of, I would agree is not nearly so simple for an AGI to execute as Yudkowsky implies. I do think that it's quite likely we're just not able to think of mechanisms even theoretically that an AGI could think of,  and one or more of those might actually be quite easy to do secretly and quickly. I wouldn't call it guaranteed by any means, but intuitively this seems like the sort of thing that raw cognitive power might have a significant bearing on.
I agree. One frightening mechanism I thought of is : "ok, assume the AGI can't craft the bioweapon or nanotechnology killbots without collecting vast amounts of information through carefully selected and performed experiments. (Basically enormous complexes full of robotics). How does it get the resources it needs? And the answer is it scams humans into doing it. We have many examples of humans trusting someone they shouldn't even when the evidence was readily available that they shouldn't.
Any “huge R&D center” constraint is trivialized in a future where agile, powerful robots will be ubiquitous and an AGI can use robots to create an underground lab in the middle of nowhere, using its superintelligence to be undetectable in all ways that are physically possible. An AGI will also be able to use robots and 3D printers to fabricate purpose-built machines that enable it to conduct billions of physical experiments a day. Sure, it would be harder to construct something like a massive particle accelerator, but 1) that isn’t needed to make killer nanobots 2) even that isn’t impossible for a sufficiently intelligent machine to create covertly and quickly.
[-]Vanessa KosoyΩ2611227

First, some remarks about the meta-level:

The ability to do new basic work noticing and fixing those flaws is the same ability as the ability to write this document before I published it, which nobody apparently did, despite my having had other things to do than write this up for the last five years or so. Some of that silence may, possibly, optimistically, be due to nobody else in this field having the ability to write things comprehensibly - such that somebody out there had the knowledge to write all of this themselves, if they could only have written it up, but they couldn't write, so didn't try. I'm not particularly hopeful of this turning out to be true in real life, but I suppose it's one possible place for a "positive model violation" (miracle). The fact that, twenty-one years into my entering this death game, seven years into other EAs noticing the death game, and two years into even normies starting to notice the death game, it is still Eliezer Yudkowsky writing up this list, says that humanity still has only one gamepiece that can do that.

Actually, I don't feel like I learned that much reading this list, compared to what I already knew. [EDIT: To be clear, this know... (read more)

There is a big chunk of what you're trying to teach which not weird and complicated, namely: "find this other agent, and what their values are". Because, "agents" and "values" are natural concepts, for reasons strongly related to "there's a relatively simple core structure that explains why complicated cognitive machines work".

This seems like it must be true to some degree, but "there is a big chunk" feels a bit too strong to me.

Possibly we don't disagree, and just have different notions of what a "big chunk" is. But some things that make the chunk feel smaller to me:

  • Humans are at least a little coherent, or we would never get anything done; but we aren't very coherent, so the project of piecing together 'what does the human brain as a whole "want"' can be vastly more difficult than the problem of figuring out what a coherent optimizer wants.
  • There are shards of planning and optimization and goal-oriented-ness in a cat's brain, but 'figure out what utopia would look like for a cat' is a far harder problem than 'identify all of the goal-encoding parts of the cat's brain and "read off" those goals'. E.g., does 'identifying utopia' in this context involve uplifting or extrapolating the
... (read more)

Humans are at least a little coherent, or we would never get anything done; but we aren't very coherent, so the project of piecing together 'what does the human brain as a whole "want"' can be vastly more difficult than the problem of figuring out what a coherent optimizer wants.

This is a point where I feel like I do have a substantial disagreement with the "conventional wisdom" of LessWrong.

First, LessWrong began with a discussion of cognitive biases in human irrationality, so this naturally became a staple of the local narrative. On the other hand, I think that a lot of presumed irrationality is actually rational but deceptive behavior (where the deception runs so deep that it's part of even our inner monologue). There are exceptions, like hyperbolic discounting, but not that many.

Second, the only reason why the question "what X wants" can make sense at all, is because X is an agent. As a corollary, it only makes sense to the extent that X is an agent. Therefore, if X is not entirely coherent then X's preferences are only approximately defined, and hence we only need to infer them approximately. So, the added difficulty of inferring X's preferences, resulting from the partial ... (read more)

Second, the only reason why the question "what X wants" can make sense at all, is because X is an agent. As a corollary, it only makes sense to the extent that X is an agent.

I'm not sure this is true; or if it's true, I'm not sure it's relevant. But assuming it is true...

Therefore, if X is not entirely coherent then X's preferences are only approximately defined, and hence we only need to infer them approximately.

... this strikes me as not capturing the aspect of human values that looks strange and complicated. Two ways I could imagine the strangeness and complexity cashing out as 'EU-maximizer-ish' are:

  • Maybe I sort-of contain a lot of subagents, and 'my values' are the conjunction of my sub-agents' values (where they don't conflict), plus the output of an idealized negotiation between my sub-agents (where they do conflict).
  • Alternatively, maybe I have a bunch of inconsistent preferences, but I have a complicated pile of meta-preferences that collectively imply some chain of self-modifications and idealizations that end up producing something more coherent and utility-function-ish after a long sequence of steps.

In both cases, the fact that my brain isn't a single coherent EU maximiz... (read more)

7Vanessa Kosoy
If we go down that path then it becomes the sort of conversation where I have no idea what common assumptions do we have, if any, that we could use to agree. As a general rule, I find it unconstructive, for the purpose of trying to agree on anything, to say things like "this (intuitively compelling) assumption is false" unless you also provide a concrete argument or an alternative of your own. Otherwise the discussion is just ejected into vacuum. Which is to say, I find it self-evident that "agents" are exactly the sort of beings that can "want" things, because agency is about pursuing objectives and wanting is about the objectives that you pursue. If you don't believe this then I don't know what these words even mean for you. Maybe, and maybe this means we need to treat "composite agents" explicitly in our models. But, there is also a case to be made that groups of (super)rational agents effectively converge into a single utility function, and if this is true, then the resulting system can just as well be interpreted as a single agent having this effective utility function, which is a solution that should satisfy the system of agents according to their existing bargaining equilibrium. If your agent converges to optimal behavior asymptotically, then I suspect it's still going to have infinite g and therefore an asymptotically-crisply-defined utility function. Of course it doesn't help on its own. What I mean is, we are going to find a precise mathematical formalization of this concept and then hard-code this formalization into our AGI design.
5Rob Bensinger
Fair enough! I don't think I agree in general, but I think 'OK, but what's your alternative to agency?' is an especially good case for this heuristic. The first counter-example that popped into my head was "a mind that lacks any machinery for considering, evaluating, or selecting actions; but it does have machinery for experiencing more-pleasurable vs. less pleasurable states". This is a mind we should be able to build, even if it would never evolve naturally. Possibly this still qualifies as an "agent" that "wants" and "pursues" things, as you conceive it, even though it doesn't select actions?
9Vanessa Kosoy
My 0th approximation answer is: you're describing something logically incoherent, like a p-zombie. My 1st approximation answer is more nuanced. Words that, in the pre-Turing era, referred exclusively to humans (and sometimes animals, and fictional beings), such as "wants", "experiences" et cetera, might have two different referents. One referent is a natural concept, something tied into deep truths about how the universe (or multiverse) works. In particular, deep truths about the "relatively simple core structure that explains why complicated cognitive machines work". The other referent is something in our specifically-human "ontological model" of the world (technically, I imagine that to be an infra-POMDP that all our hypotheses our refinements of). Since the latter is a "shard" of the former produced by evolution, the two referents are related, but might not be the same. (For example, I suspect that cats lack natural!consciousness but have human!consciousness.) The creature you describe does not natural!want anything. You postulated that it is "experiencing more pleasurable and less pleasurable states", but there is no natural method that would label its states as such, or that would interpret them as any sort of "experience". On the other hand, maybe if this creature is designed as a derivative of the human brain, then it does human!want something, because our shard of the concept of "wanting" mislabels (relatively to natural!want) weird states that wouldn't occur in the ancestral environment. You can then ask, why should we design the AI to follow what we natural!want rather than what we human!want? To answer this, notice that, under ideal conditions, you converge to actions that maximize your natural!want, (more or less) according to definition of natural!want. In particular, under ideal conditions, you would build an AI that follows your natural!want. Hence, it makes sense to take a shortcut and "update now to the view you will predictably update to later":

On Twitter, Eric Rogstad wrote:

"the thing where it keeps being literally him doing this stuff is quite a bad sign"

I'm a bit confused by this part. Some thoughts on why it seems odd for him (or others) to express that sentiment...

1. I parse the original as, "a collection of EY's thoughts on why safe AI is hard". It's EY's thoughts, why would someone else (other than @robbensinger) write a collection of EY's thoughts?

(And if we generalize to asking why no-one else would write about why safe AI is hard, then what about Superintelligence, or the AI stuff in cold-takes, or ...?)

2. Was there anything new in this doc? It's prob useful to collect all in one place, but we don't ask, "why did no one else write this" for every bit of useful writing out there, right?

Why was it so overwhelmingly important that someone write this summary at this time, that we're at all scratching our heads about why no one else did it?

Copying over my reply to Eric:

My shoulder Eliezer (who I agree with on alignment, and who speaks more bluntly and with less hedging than I normally would) says:

  1. The list is true, to the best of my knowledge, and the details actually matter.

    Many civilizations try to make a canonical
... (read more)
  I don't think making this list in 1980 would have been meaningful. How do you offer any sort of coherent, detailed plan for dealing with something when all you have is toy examples like Eliza?  We didn't even have the concept of machine learning back then - everything computers did in 1980 was relatively easily understood by humans, in a very basic step-by-step way. Making a 1980s computer "safe" is a trivial task, because we hadn't yet developed any technology that could do something "unsafe" (i.e. beyond our understanding). A computer in the 1980s couldn't lie to you, because you could just inspect the code and memory and find out the actual reality. What makes you think this would have been useful? Do we have any historical examples to guide us in what this might look like?

I think most worlds that successfully navigate AGI risk have properties like:

  • AI results aren't published publicly, going back to more or less the field's origin.
  • The research community deliberately steers toward relatively alignable approaches to AI, which includes steering away from approaches that look like 'giant opaque deep nets'.
    • This means that you need to figure out what makes an approach 'alignable' earlier, which suggests much more research on getting de-confused regarding alignable cognition.
      • Many such de-confusions will require a lot of software experimentation, but the kind of software/ML that helps you learn a lot about alignment as you work with it is itself a relatively narrow target that you likely need to steer towards deliberately, based on earlier, weaker deconfusion progress. I don't think having DL systems on hand to play with has helped humanity learn much about alignment thus far, and by default, I don't expect humanity to get much more clarity on this before AGI kills us.
  • Researchers focus on trying to predict features of future systems, and trying to get mental clarity about how to align such systems, rather than focusing on 'align ELIZA' just because ELIZA is
... (read more)
5Thomas Kwa
"most worlds that successfully navigate AGI risk" is kind of a strange framing to me.  For one thing, it represents p(our world | success) and we care about p(success | our world). To convert between the two you of course need to multiply by p(success) / p(our world). What's the prior distribution of worlds? This seems underspecified. For another, using the methodology "think about whether our civilization seems more competent than the problem is hard" or "whether our civilization seems on track to solve the problem" I might have forecast nuclear annihilation (not sure about this). The methodology seems to work when we're relatively certain about the level of difficulty on the mainline, so if I were more sold on that I would believe this more. It would still feel kind of weird though.
I mean, I think many of the computing pioneers 'basically saw' AI risk. I noted some surprise that IJ Good didn't write the precursor to this list in 1980, and apparently Wikipedia claims there was an unpublished statement in 1998 about AI x-risk; it'd be interesting to see what it contains and how much it does or doesn't line up with our modern conception of why the problem is hard.

The historical figures who basically saw it (George Eliot 1879: "will the creatures who are to transcend and finally supersede us be steely organisms [...] performing with infallible exactness more than everything that we have performed with a slovenly approximativeness and self-defeating inaccuracy?"; Turing 1951: "At some stage therefore we should have to expect the machines to take control") seem to have done so in the spirit of speculating about the cosmic process. The idea of coming up with a plan to solve the problem is an additional act of audacity; that's not really how things have ever worked so far. (People make plans about their own lives, or their own businesses; at most, a single country; no one plans world-scale evolutionary transitions.)

5Andrew McKnight
I'm tempted to call this a meta-ethical failure. Fatalism, universal moral realism, and just-world intuitions seem to be the underlying implicit hueristics or principals that would cause this "cosmic process" thought-blocker.
Why is this v0 and not, or the Sequences, or any of the documents that Evan links to here? That's part of what I meant to be responding to — not that this post is not useful, but that I don't see what makes it so special compared to all the other stuff that Eliezer and others have already written.
To put it another way, I would agree that Eliezer has made (what seem to me like) world-historically-significant contributions to understanding and advocating for (against) AI risk. So, if 2007 Eliezer was asking himself, "Why am I the only one really looking into this?", I think that's a very reasonable question. But here in 2022, I just don't see this particular post as that significant of a contribution compared to what's already out there.
Wrote a long comment here. (Which you've seen, but linking since your comment started as a response to me.)

-3.  I'm assuming you are already familiar with some basics, and already know what 'orthogonality' and 'instrumental convergence' are and why they're true.

I think this is actually the part that I most "disagree" with. (I put "disagree" in quotes, because there are forms of these theses that I'm persuaded by. However, I'm not so confident that they'll be relevant for the kinds of AIs we'll actually build.)

1. The smart part is not the agent-y part

It seems to me that what's powerful about modern ML systems is their ability to do data compression / pattern recognition. That's where the real cognitive power (to borrow Eliezer's term) comes from. And I think that this is the same as what makes us smart.

GPT-3 does unsupervised learning on text data. Our brains do predictive processing on sensory inputs. My guess (which I'd love to hear arguments against!) is that there's a true and deep analogy between the two, and that they lead to impressive abilities for fundamentally the same reason.

If so, it seems to me that that's where all the juice is. That's where the intelligence comes from. (In the past, I've called this the core smarts of our brains.)

On this view, all the agent-y, planful... (read more)

GPT-3 does unsupervised learning on text data. Our brains do predictive processing on sensory inputs. My guess (which I'd love to hear arguments against!) is that there's a true and deep analogy between the two, and that they lead to impressive abilities for fundamentally the same reason.

Agree that self-supervised learning powers both GPT-3 updates and human brain world-model updates (details & caveats). (Which isn’t to say that GPT-3 is exactly the same as the human brain world-model—there are infinitely many different possible ML algorithms that all update via self-supervised learning).


If so, it seems to me that that's where all the juice is. That's where the intelligence comes from … if agency is not a fundamental part of intelligence, and rather something that can just be added in on top, or not, and if we're at a loss for how to either align a superintelligent agent with CEV or else make it corrigible, then why not try to avoid creating the agent part of superintelligent agent?

I disagree; I think the agency is necessary to build a really good world-model, one that includes new useful concepts that humans have never thought of.

Without the agency, some of the things ... (read more)

Why is agency necessary for these things? If we follow Ought's advice and build "process-based systems [that] are built on human-understandable task decompositions, with direct supervision of reasoning steps", do you expect us to hit a hard wall somewhere that prevents these systems from creatively choosing things to think about, books to read, or better brainstorming strategies?
7Steven Byrnes
(Copying from here:) (Does that count as “agency”? I don’t know, it depends on what you mean by “agency”.) In terms of the “task decomposition” strategy, this might be a tricky to discuss because you probably have a more detailed picture in your mind than I do. I’ll try anyway. It seems to me that the options are: (1) the subprocess only knows its narrow task (“solve this symplectic geometry homework problem”), and is oblivious to the overall system goal (“design a better microscope”), or (2) the subprocess is aware of the overall system goal and chooses actions in part to advance it. In Case (2), I’m not sure this really counts as “task decomposition” in the first place, or how this would help with safety. In Case (1), yes I expect systems to hit a hard wall—I’m skeptical that tasks we care about decompose cleanly. For example, at my last job, I would often be part of a team inventing a new gizmo, and it was not at all unusual for me to find myself sketching out the algorithms and sketching out the link budget and scrutinizing laser spec sheets and scrutinizing FPGA spec sheets and nailing down end-user requirements, etc. etc. Not because I’m individually the best person at each of those tasks—or even very good!—but because sometimes a laser-related problem is best solved by switching to a different algorithm, or an FPGA-related problem is best solved by recognizing that the real end-user requirements are not quite what we thought, etc. etc. And that kind of design work is awfully hard unless a giant heap of relevant information and knowledge is all together in a single brain / world-model. In the case of my current job doing AI alignment research, I sometimes come across small self-contained tasks that could be delegated, but I would have no idea how to decompose most of what I do. (E.g. writing this comment!) Here’s John Wentworth making a similar point more eloquently: A possible example of a seemingly-hard-to-decompose task would be: Until 1948, no h
1David Johnston
FWIW self-supervised learning can be surprisingly capable of doing things that we previously only knew how to do with "agentic" designs. From that link: classification is usually done with an objective + an optimization procedure, but GPT-3 just does it.

For example, I claim that while AlphaGo could be said to be agent-y, it does not care about atoms. And I think that we could make it fantastically more superhuman at Go, and it would still not care about atoms. Atoms are just not in the domain of its utility function.

In particular, I don't think it has an incentive to break out into the real world to somehow get itself more compute, so that it can think more about its next move. It's just not modeling the real world at all. It's not even trying to rack up a bunch of wins over time. It's just playing the single platonic game of Go.

I would distinguish three ways in which different AI systems could be said to "not care about atoms":

  1. The system is thinking about a virtual object (e.g., a Go board in its head), and it's incapable of entertaining hypotheses about physical systems. Indeed, we might add the assumption that it can't entertain hypotheses like 'this Go board I'm currently thinking about is part of a larger universe' at all. (E.g., there isn't some super-Go-board I and/or the board are embedded in.)
  2. The system can think about atoms/physics, but it only terminally cares about digital things in a simulated environment (e.g., winni
... (read more)
In my mind, this is still making the mistake of not distinguishing the true domain of the agent's utility function from ours. Whether the simulation continues to be instantiated in some computer in our world is a fact about our world, not about the simulated world. AlphaGo doesn't care about being unplugged in the middle of a game (unless that dynamic was part of its training data). It cares about the platonic game of go, not about the instantiated game it's currently playing. We need to worry about leaky abstractions, as per my original comment. So we can't always assume the agent's domain is what we'd ideally want it to be. But I'm trying to highlight that it's possible (and I would tentatively go further and say probable) for agents not to care about the real world. To me, assuming care about the real world (including wanting not to be unplugged) seems like a form of anthropomorphism. For any given agent-y system I think we need to analyze whether it in particular would come to care about real world events. I don't think we can assume in general one way or the other.
6Rob Bensinger
What if the programmers intervene mid-game to give the other side an advantage? Does a Go AGI, as you're thinking of it, care about that? I'm not following why a Go AGI (with the ability to think about the physical world, but a utility function that only cares about states of the simulation) wouldn't want to seize more hardware, so that it can think better and thereby win more often in the simulation; or gain control of its hardware and directly edit the simulation so that it wins as many games as possible as quickly as possible. Why would having a utility function that only assigns utility based on X make you indifferent to non-X things that causally affect X? If I only terminally cared about things that happened a year from now, I would still try to shape the intervening time because doing so will change what happens a year from now. (This is maybe less clear in the case of shutdown, because it's not clear how an agent should think about shutdown if its utility is defined states of its simulation. So I'll set that particular case aside.)
2David Johnston
A Go AI that learns to play go via reinforcement learning might not "have a utility function that only cares about winning Go". Using standard utility theory, you could observe its actions and try to rationalise them as if they were maximising some utility function, and the utility function you come up with probably wouldn't be "win every game of Go you start playing" (what you actually come up with will depend, presumably, on algorithmic and training regime details). The reason why the utility function is slippery is that it's fundamentally an adaptation executor, not a utility maxmiser.
2David Johnston
Not necessarily. Train something multimodally on digital games of Go and on, say, predicting the effects of modifications to its own code on its success at Go. It could be a) good at go and b) have some real understanding of "real world actions" that make it better at Go, and still not actually take any real world actions to make it better at Go, even if it had the opportunity. You could modify the training to make it likely to do so - perhaps by asking it to either make a move or to produce descendants that make better choices - but if you don't do this then it seems entirely plausible, and even perhaps likely, that it develops an understanding of self-modification and of go playing without ever self-modifying in order to play go better. Its goal, so to speak, is "play go with the restriction of using only legal game moves". Edit - forget the real world, here's an experiment: Train a board game playing AI with two modes of operation: game state x move -> outcome and game state -> best move. Subtle difference: in the first mode of operation, the move has a "cheat button" that, when pressed, always results in a win. In the second, it can output cheat button presses, but it has no effect on winning or losing. Question is: does it learn to press the cheat button? I'm really not sure. Could you prevent it from learning to press the cheat button if training feedback is never allowed to depend on whether or not this button was pressed? That seems likely.
7James Payor
Can you visualize an agent that is not "open-ended" in the relevant ways, but is capable of, say, building nanotech and melting all the GPUs? In my picture most of the extra sauce you'd need on top of GPT-3 looks very agenty. It seems tricky to name "virtual worlds" in which AIs manipulate just "virtual resources" and still manage to do something like melting the GPUs.
8James Payor
I should say that I do see this as a reasonable path forward! But we don't seem to be coordinating to do this, and AI researchers seem to love doing work on open-ended agents, which sucks. Hm, regardless it doesn't really move the needle, so long as people are publishing all of their work. Developing overpowered pattern recognizers is similar to increasing our level of hardware overhang. People will end up using them as components of systems that aren't safe.
4David Johnston
I strongly disagree. Gain of function research happens, but it's rare because people know it's not safe. To put it mildly, I think reducing the number of dangerous experiments substantially improves the odds of no disaster happening over any given time frame
FWIW, I'm not sold on the idea of taking a single pivotal act. But, engaging with what I think is the real substance of the question — can we do complex, real-world, superhuman things with non-agent-y systems? Yes, I think we can! Just as current language models can be prompt-programmed into solving arithmetic word problems, I think a future system could be led to generate a GPU-melting plan, without it needing to be a utility-maximizing agent. For a very hand-wavy sketch of how that might go, consider asking GPT-N to generate 1000s of candidate high-level plans, then rate them by feasibility, then break each plan into steps and re-evaluate, etc. Or, alternatively, imagine the cognitive steps you might take if you were trying to come up with a GPU-melting plan (or alternatively a pivotal act plan in general). Do any of those steps really require that you have a utility function or that you're a goal-directed agent? It seems to me that we need some form of search, and discrimination and optimization. But not necessarily anymore than GPT-3 already has. (It would just need to be better at the search. And we'd need to make many many passes through the network to complete all the cognitive steps.) On your view, what am I missing here? * Is GPT-3 already more of an agent than I realize? (If so, is it dangerous?) * Will GPT-N by default be more of an agent than GPT-3? * Are our own thought processes making use of goal-directedness more than I realize? * Will prompt-programming passive systems hit a wall somewhere? * If so, what are some of the simplest cognitive tasks that we can do that you think such systems wouldn't be able to do? * (See also my similar question here.)
4David Johnston
FWIW, I'd call this "weakly agentic" in the sense that you're searching through some options, but the number of options you're looking through is fairly small. It's plausible that this is enough to get good results and also avoid disasters, but it's actually not obvious to me. The basic reason: if the top 1000 plans are good enough to get superior performance, they might also be "good enough" to be dangerous. While it feels like there's some separation between "useful and safe" and "dangerous" plans and this scheme might yield plans all of the former type, I don't presently see a stronger reason to believe that this is true.
Separately from whether the plans themselves are safe or dangerous, I think the key question is whether the process that generated the plans is trying to deceive you (so it can break out into the real world or whatever). If it's not trying to deceive you, then it seems like you can just build in various safeguards (like asking, "is this plan safe?", as well as more sophisticated checks), and be okay.
>then rate them by feasibility, I mean, literal GPT is just going to have poor feasibility ratings for novel engineering concepts. >Do any of those steps really require that you have a utility function or that you're a goal-directed agent? Yes, obviously. You have to make many scientific and engineering discoveries, which involves goal-directed investigation.  > Are our own thought processes making use of goal-directedness more than I realize? Yes, you know which ideas make sense by generalizing from ideas more closely tied in with the actions you take directed towards living.  
4David Johnston
What do you think of a claim like "most of the intelligence comes from the steps where you do most of the optimization"? A corollary of this is that we particularly want to make sure optimization intensive steps of AI creation are safe WRT not producing intelligent programs devoted to killing us. Example: most of the "intelligence" of language models comes from the supervised learning step. However, it's in-principle plausible that we could design e.g. some really capable general purpose reinforcement learner where the intelligence comes from the reinforcement, and the latter could (but wouldn't necessarily) internalise "agenty" behaviour. I have a vague impression that this is already something other people are thinking about, though maybe I read too much into some tangential remarks in this direction. E.g. I figured the concern about mesa-optimizers was partly motivated by the idea that we can't always tell when an optimization intensive step is taking place. I can easily imagine people blundering into performing unsafe optimization-intensive AI creation processes. Gain of function pathogen research would seem to be a relevant case study here, except we currently have less idea about what kind of optimization makes deadly AIs vs what kind of optimization makes deadly pathogens. One of the worries (again, maybe I'm reading too far into comments that don't say this explicitly) is that the likelihood of such a blunder approaches 1 over long enough times, and the "pivotal act" framing is supposed to be about doing something that could change this (??) That said, it seems that there's a lot that could be done to make it less likely in short time frames.
This seems probably right to me. I agree that reinforcement learners seem more likely to be agent-y (and therefore scarier) than self-supervised learners.

I agree with pretty much everything here, and I would add into the mix two more claims that I think are especially cruxy and therefore should maybe be called out explicitly to facilitate better discussion:

Claim A: “There’s no defense against an out-of-control omnicidal AGI, not even with the help of an equally-capable (or more-capable) aligned AGI, except via aggressive outside-the-Overton-window acts like preventing the omnicidal AGI from being created in the first place.”

I think this claim is true, on account of gray goo and lots of other things, and I suspect Eliezer does too, and I’m pretty sure other people disagree with this claim.

If someone disagrees with this claim (i.e., if they think that if DeepMind can make an aligned and Overton-window-abiding “helper” AGI, then we don’t have to worry about Meta making a similarly-capable out-of-control omnicidal misaligned AGI the following year, because DeepMind’s AGI will figure out how to protect us), and also believes in extremely slow takeoff, I can see how such a person might be substantially less pessimistic about AGI doom than I am.

Claim B: “Shortly after (i.e., years not decades after) we have dangerous AGI, we will have dang... (read more)

 I think this claim is true, on account of gray goo and lots of other things, and I suspect Eliezer does too, and I’m pretty sure other people disagree with this claim.

If you have robust alignment, or AIs that are rapidly bootstrapping their level of alignment fast enough to outpace the danger of increased capabilities, aligned AGI could get through its intelligence explosion to get radically superior technology and capabilities. It could also get a hard start on superexponential replication in space, so that no follower could ever catch up, and enough tech and military hardware to neutralize any attacks on it (and block attacks on humans via nukes, bioweapons, robots, nanotech, etc). That wouldn't work if there are thing like vacuum collapse available to attackers, but we don't have much reason to expect that from current science and the leading aligned AGI would find out first.

That could be done without any violation of the territory of other sovereign states. The legality of grabbing space resources is questionable in light of the Outer Space Treaty, but commercial exploitation of asteroids is in the Overton window. The superhuman AGI would also be in a good position to per... (read more)

A bit pedantic, but isn't superexponential replication too fast? Won't it hit physical limits eventually, e.g. expanding at the speed of light in each direction, so at most a cubic function of time? Also, never allowing followers to catch up means abandoning at least some or almost all of the space you passed through. Plausibly you could take most of the accessible and useful resources with you, which would also make it harder for pursuers to ever catch up, since they will plausibly need to extract resources every now and then to fuel further travel. On the other hand, it seems unlikely to me that we could extract or destroy resources quickly enough to not leave any behind for pursuers, if they're at most months behind.
Naturally it doesn't go on forever, but any situation where you're developing technologies that move you to successively faster exponential trajectories is superexponential overall for some range. E.g. if you have robot factories that can reproduce exponentially until they've filled much of the Earth or solar system, and they are also developing faster reproducing factories,  the overall process is superexponential. So is the history of human economic growth, and the improvement from an AI intelligence explosion. By the time you're at ~cubic expansion being ahead on the early superexponential phase the followers have missed their chance.
I agree that they probably would have missed their chance to catch up with the frontier of your expansion. Maybe an electromagnetic radiation-based assault could reach you if targeted (the speed of light is constant relative to you in a vacuum, even if you're traveling in the same direction), although unlikely to get much of the frontier of your expansion, and there are plausibly effective defenses, too. Do you also mean they wouldn't be able to take most what you've passed through, though? Or it wouldn't matter? If so, how would this be guaranteed (without any violation of the territory of sovereign states on Earth)? Exhaustive extraction in space? An advantage in armed space conflicts?

I agree with these two points. I think an aligned AGI actually able to save the world would probably take initial actions that look pretty similar to those an unaligned AGI would take. Lots of sizing power, building nanotech, colonizing out into space, self-replication, etc. 

So how would we know the difference (for the first few years at least)?

If it kills you, then it probably wasn’t aligned. 

Maybe it did that to save your neural weights.  Define 'kill'. 
4Quintin Pope
I did say “probably”!
I disagree with this claim inasmuch as I expect a year headstart by an aligned AI is absolutely enough to prevent Meta from killing me and my family. 

Depends on what DeepMind does with the AI, right?

Maybe DeepMind uses their AI in very narrow, safe, low-impact ways to beat ML benchmarks, or read lots of cancer biology papers and propose new ideas about cancer treatment.

Or alternatively, maybe DeepMind asks their AI to undergo recursive self-improvement and build nano-replicators in space, etc., like in Carl Shulman’s reply.

I wouldn’t have thought that the latter is really in the Overton window. But what do I know.

You could also say “DeepMind will just ask their AI what they should do next”. If they do that, then maybe the AI (if they’re doing really great on safety such that the AI answers honestly and helpfully) will reply: “Hey, here’s what you should do, you should let me undergo recursive-self-improvement, and then I’ll be able to think of all kinds of crazy ways to destroy the world, and then I can think about how to defend against all those things”. But if DeepMind is being methodical & careful enough that their AI hasn’t destroyed the world already by this point, I’m inclined to think that they’re also being methodical & careful enough that when the AI proposes to do that, DeepMind will say, “Umm, no, that’s total... (read more)

If DeepMind was committed enough to successfully build an aligned AI (which, as extensively elaborated upon in the post, is a supernaturally difficult proposition), I would assume they understand why running it is necessary. There's no reason to take all of the outside-the-overton-window measures indicated in the above post unless you have functioning survival instincts and have thought through the problem sufficiently to hit the green button.
If you can build one aligned superintelligence, then plausibly you can 1. explain to other AGI developers how to make theirs safe or even just give them a safe design (maybe homomorphically encrypted to prevent modification, but they might not trust that), and 2. have aligned AGI monitoring the internet and computing resources, and alert authorities of abnomalies that might signal new AGI developments. Require that AGI developments provide proof that they were designed according to one of a set of approved designs, or pass some tests determined by your aligned superintelligence. Then aligned AGI can proliferate first and unaligned AGI will plausibly face severe barriers. Plausibly 1 is enough, since there's enough individual incentive to build something safe or copy other people's designs and save work. 2 depends on cooperation with authorities and I'd guess cloud computing service providers or policy makers.

explain to other AGI developers how to make theirs safe or even just give them a safe design (maybe homomorphically encrypted to prevent modification, but they might not trust that)

What if the next would-be AGI developer rejects your “explanation”, and has their own great ideas for how to make an even better next-gen AGI that they claim will work better, and so they discard your “gift” and proceed with their own research effort?

I can think of at least two leaders of would-be AGI development efforts (namely Yann LeCun of Meta and Jeff Hawkins of Numenta) who believe (what I consider to be) spectacularly stupid things about AGI x-risk, and have believed those things consistently for decades, despite extensive exposure to good counter-arguments.

Or what if the next would-be AGI developer agrees with you and accepts your “gift”, and so does the one after that, and the one after that, but not the twelfth one?

have aligned AGI monitoring the internet and computing resources, and alert authorities of [anomalies] that might signal new AGI developments. Require that AGI developments provide proof that they were designed according to one of a set of approved designs, or pass some tests determi

... (read more)
When you ask "what if", are you implying these things are basically inevitable? And inevitable no matter how much more compute aligned AGIs have before unaligned AGIs are developed and deployed? How much of a disadvantage against aligned AGIs does an unaligned AGI need before doom isn't overwhelmingly likely? What's the goal post here for survival probability? You can have AGIs monitoring for pathogens, nanotechnology, other weapons, and building defenses against them, and this could be done locally and legally. They can monitor transactions and access to websites through which dangerous physical systems (including possibly factories, labs, etc.) could be taken over or built. Does every country need to be competent and compliant to protect just one country from doom? The Overton window could also shift dramatically if omnicidal weapons are detected. I agree that plausibly not every country with significant compute will comply, and hacking everyone is outside the public Overton window. I wouldn't put hacking everyone past the NSA, but also wouldn't count on them either.
4Steven Byrnes
Let’s see, I think “What if the next would-be AGI developer rejects your “explanation” / “gift”” has a probability that asymptotes to 100% as the number of would-be AGI developers increases. (Hence “Claim B” above becomes relevant.) I think “What if the authorities in most countries do care, but not the authorities in every single country?” seems to have high probability in today’s world, although of course I endorse efforts to lower the probability. I think “What if the only way to “monitor the internet and computing resources” is to hack into every data center and compute cluster on the planet? (Including those in secret military labs.)” seems very likely to me, conditional on “Claim B” above. Hmm. Offense-defense balance in bio-warfare is not obvious to me. Preventing a virus from being created would seem to require 100% compliance by capable labs, but I’m not sure how many “capable labs” there are, or how geographically distributed and rule-following. Once the virus starts spreading, aligned AGIs could help with vaccines, but apparently a working COVID-19 vaccine was created in 1 day, and that didn’t help much, for various societal coordination & governance reasons. So then you can say “Maybe aligned AGI will solve all societal coordination and governance problems”. And maybe it will! Or, maybe some of those coordination & governance problems come from blame-avoidance and conflicts-of-interest and status-signaling and principle-agent problems and other things that are not obviously solvable by easy access to raw intelligence. I don’t know. Offense-defense balance in nuclear warfare is likewise not obvious to me. I presume that an unaligned AGI could find a way to manipulate nuclear early warning systems (trick them, hack into them, bribe or threaten their operators, etc.) to trigger all-out nuclear war, after hacking into a data center in New Zealand that wouldn’t be damaged. An aligned AGI playing defense would need to protect against these vulnerabilities.
Some more skepticism about infectious diseases and nukes killing us all here: Also my more general skeptical take against non-nano attacks here: With nanotech, I think there will be tradeoffs between targeting effectiveness and requiring (EM) signals from computers that can be effectively interferred with through things within or closer to the Overton window. Maybe a crux is how good autonomous nanotech with no remote control would be at targeting humans or spreading so much that it just gets into almost all buildings or food or water because it's basically going everywhere.
4Steven Byrnes
Thanks! I wasn’t assuming the infectious diseases and nukes by themselves would kill us all. They don’t have to, because the AGI can do other things in conjunction, like take command of military drones and mow down the survivors (or bomb the PPE factories), or cause extended large-scale blackouts, which would incidentally indirectly prevent PPE production and distribution, along with preventing pretty much every other aspect of an organized anti-pandemic response. See Section 1.6 here. So that brings us to the topic of offense-defense balance for illicitly taking control of military drones. And I would feel concerned about substantial delays before the military trusts a supposedly-aligned AGI so much that they give it root access to all its computer systems (which in turn seems necessary if the aligned AGI is going to be able to patch all the security holes, defend against spear-phishing attacks, etc.) Of course there’s the usual caveat that maybe DeepMind will give their corrigible aligned AGI permission to hack into military systems (for their own good!), and then maybe we wouldn’t have to worry. But the whole point of this discussion is that I’m skeptical that DeepMind would actually give their AGI permission to do something like that. And likewise we would need to talk about offense-defense balance for the power grid. And I would have the same concern about people being unwilling to give a supposedly-aligned AGI root access to all the power grid computers. And I would also be concerned about other power grid vulnerabilities like nuclear EMPs, drone attacks on key infrastructure, etc. And likewise, what’s the offense-defense balance for mass targeted disinformation campaigns? Well, if DeepMind gives its AGI permission to engage in a mass targeted counter-disinformation campaign, maybe we’d be OK on that front. But that’s a big “if”! …And probably dozens of other things like that. Seems like a good question, and maybe difficult to resolve. Or maybe I would
I think there would be too many survivors and enough manned defense capability for existing drones to directly kill the rest of us with high probability. Blocking PPE production and organized pandemic responses still won't stop people from self-isolating, doing no contact food deliveries, etc., although things would be tough, and deliveries and food production would be good targets for drone strikes. It could be bad if lethal pathogens become widespread and practically unremovable in our food/water, or if food production is otherwise consistently attacked, but the militaries would probably step in to protect the food/water supplies. I think, overall, there are too few ways to reliably and kill double or even single digit percentages of the human population with high probability and that can be combined to get basically everyone with high probability. I'm not saying there aren't any, but I'm skeptical that there are enough. There are diminishing returns on doing the same ones (like pandemics) more, because of resistance, and enough people being personally very careful or otherwise difficult targets.