Posts
Comments
Good question! Yeah, there's nothing fundamentally quantum about this effect. But if the simulator wants to focus on universes with 1 & 2 fixed (e.g. if they're trying to calculate the distribution of superintelligences across Tegmark IV), the PNRG (along with the initial conditions of the universe) seem like good places for a simulator to tweak things.
It is not clear to me that this would result in a lower Kolmogorov complexity at all. Such an algorithm could of course use a pseudo-random number generator for the vast majority quantum events which do not affect p(ASI) (like the creation of CMB photons), but this is orthogonal to someone nudging the relevant quantum events towards ASI. For these relevant events, I am not sure that the description "just do whatever favors ASI" is actually shorter than just the sequence of events.
Hmm, I notice I may have been a bit unclear in my original post. When I'd said "pseudorandom", I wasn't referring to the use of a pseudo-random number generator instead of a true RNG. I was referring to the "transcript" of relevant quantum events only appearing random, without being "truly random", because of the way in which they were generated (which I'm thinking of as being better described as "sampled from a space parameterizing the possible ways the world could be, conditional on humanity building superintelligence" rather than "close to truly random, or generated by a pseudo-random RNG, except with nudges toward ASI".)
I mean, if we are simulated by a Turing Machine (which is equivalent to quantum events having a low Kolmogorov complexity), then a TM which just implements the true laws of physics (and cheats with a PNRG, not like the inhabitants would ever notice) is surely simpler than one which tries to optimize towards some distant outcome state.
As an analogy, think about the Kolmogorov complexity of a transcript of a very long game of chess. If both opponents are following a simple algorithm of "determine the allowed moves, then use a PRNG to pick one of them", that should have a bound complexity. If both are chess AIs which want to win the game (i.e. optimize towards a certain state) and use a deterministic PRNG (lest we are incompressible), the size of your Turing Machine -- which /is/ the Kolmogorov complexity -- just explodes.
Wouldn't this also serve as an argument against malign consequentialists in the Solomonoff prior, that may make it a priori more likely for us to end up in a world with particular outcomes optimized in their favor?
It is not clear to me that this would result in a lower Kolmogorov complexity at all.
[...]
Look at me rambling about universe-simulating TMs. Enough, enough.
To be clear, it's also not clear to me that this would result in a lower K-complexity either. My main point is that (1) the null hypothesis of quantum events being independent of consciousness rests on assumptions (like assumptions about what the Solomonoff prior is like) that I think are actually pretty speculative, and that (2) there are speculative ways the Solomonoff prior could be in which our consciousness can influence quantum outcomes.
My goal here is not to make a positive case for consciousness affecting quantum outcomes, as much as it is to question the assumptions behind the case against the world working that way.
This. Physics runs on falsifiable predictions. If 'consciousness can affect quantum outcomes' is any more true than the classic 'there is an invisible dragon in my garage', then discovering that fact would seem easy from an experimentalist standpoint. Sources of quantum randomness (e.g. weak source+detector) are readily available, so any claimant who thinks they can predict or affect their outcomes could probably be tested initially for a few 100$.
Yes, I'm also bearish on consciousness affecting quantum outcomes in ways that are as overt and measurable in the way you're gesturing at. The only thing I was arguing in this post is that the effect size of consciousness on quantum outcomes is maybe more than zero, as opposed to obviously exactly zero. I don't think of myself as having made any arguments that the effect size should be non-negligible, although I also don't think that possibility has been ruled out for non-neglible effect sizes lying somewhere between "completely indistinguishable from no influence at all" and "overt and measurable to the extent a proclaimed psychic could reproducibly affect quantum RNG outcomes".
I'll take a stab at this. Suppose we had strong a priori reasons for thinking it's in our logical past that we'll have created a superintelligence of some sort. Let's suppose that some particular quantum outcome in the future can get chaotically amplified, so that in one Everett branch humanity never builds any superintelligence because of some sort of global catastrophe (say with 99% probability, according to the Born rule), and in some other Everett branch humanity builds some kind of superintelligence (say with 1% probability, according to the Born rule). Then we should expect to end up in the Everett branch in which humanity builds some kind of superintelligence with ~100% probability, despite the Born rule saying we only have a 1% chance of ending up there, because the "99%-likely" Everett branch was ruled out by our a priori reasoning.
I'm not sure if this is the kind of concrete outcome that you're asking for. I imagine that, for the most part, the kind of universe I'm describing will still yield frequencies that converge on the Born probabilities, and for the most part appear indistinguishable from a universe in which quantum outcomes are "truly random". See my reply to Joel Burget for some more detail about how I think about this hypothesis.
If we performed a trillion 50/50 quantum coin flips, and found a program with K-complexity far less than a trillion that could explain these outcomes, that would be an example of evidence in favor of this hypothesis. (I don't think it's very likely that we'll be able to find a positive result if we run that particular experiment; I'm naming it more to illustrate the kind of thing that would serve as evidence.) (EDIT: This would only serve as evidence against quantum outcomes being truly random. In order for it to serve as evidence in favor of quantum outcomes being impacted by consciousness, the low K-complexity program explaining these outcomes would need to route through the decisions of conscious beings somehow; it wouldn't work if the program were just printing out digits of pi in binary, for example.)
My inside view doesn't currently lead me to put much credence on this picture of reality actually being true. My inside view is more like "huh, I notice I have become way more uncertain about the a priori arguments about what kind of universe we live in -- especially the arguments that we live in a universe in which quantum outcomes are supposed to be 'truly random' -- so I will expand my hypothesis space for what kinds of universes we might be living in".
Shortly after publishing this, I discovered something written by John Wheeler (whom Chris Langan cites) that feels thematically relevant. From Law Without Law:
I was hoping people other than Jessica would share some specific curated insights they got. Syndiffeonesis is in fact a good insight.
I finally wrote one up! It ballooned into a whole LessWrong post.
It seems if I only read the main text, the obvious interpretation is that points are events and the circles restrict which other events they can interact with.
This seems right to me, as far as I can tell, with the caveat that "restrict" (/ "filter") and "construct" are two sides of the same coin, as per constructive-filtrative duality.
From the diagram text, it seems he is instead saying that each circle represents entangled wavefunctions of some subset of objects that generated the circle.
I think each circle represents the entangled wavefunctions of all of the objects that generated the circle, not just some subset.
Relatedly, you talk about "the" wave function in a way that connotes a single universal wave function, like in many-worlds. I'm not sure if this is what you're intending, but it seems plausible that the way you're imagining things is different from how my model of Chris is imagining things, which is as follows: if there are N systems that are all separable from one another, we could write a universal wave function for these N systems that we could factorize as ψ_1 ⊗ ψ_2 ⊗ ... ⊗ ψ_N, and there would be N inner expansion domains (/ "circles"), one for each ψ_i, and we can think of each ψ_i as being "located within" each of the circles.
Great. Yes, I think that's the thing to do. Start small! I (and presumably others) would update a lot from a new piece of actual formal mathematics from Chris's work. Even if that work was, by itself, not very impressive.
(I would also want to check that that math had something to do with his earlier writings.)
I think we're on exactly the same page here.
Please be prepared for the possibility that Chris is very smart and creative, and that he's had some interesting ideas (e.g. Syndiffeonesis), but that his framework is more of a interlocked collection of ideas than anything mathematical (despite using terms from mathematics). Litany of Tarsky and all that.
That's certainly been a live hypothesis in my mind as well, that I don't think can be ruled out before I personally see (or produce) a piece of formal math (that most mathematicians would consider formal, lol) that captures the core ideas of the CTMU.
So Chris either (i) doesn't realize that you need to be precise to communicate with mathematicians, or (ii) doesn't understand how to be precise.
While I agree that there isn't very much explicit and precise mathematical formalism in the CTMU papers themselves, my best guess is that (iii) Chris does unambiguously gesture at a precise structure he has in mind, assuming a sufficiently thorough understanding of the background assumptions in his document (which I think is a false assumption for most mathematicians reading this document). By analogy, it seems plausible to me that Hegel was gesturing at something quite precise in some of his philosophical works, that only got mathematized nearly 200 years later by category theorists. (I don't understand any Hegel myself, so take this with a grain of salt.)
Except, I can already predict you're going to say that no piece of his framework can be understood without the whole. Not even by making a different smaller framework that exists just to showcase the well-ordering alternative. It's a little suspicious.
False! :P I think no part of his framework can be completely understood without the whole, but I think the big pictures of some core ideas can be understood in relative isolation. (Like syndiffeonesis, for example.) I think this is plausibly true for his alternatives to well-ordering as well.
If you're going to fund someone to do something, it should be to formalize Chris's work. That would not only serve as a BS check, it would make it vastly more approachable.
I'm very on board with formalizing Chris's work, both to serve as a BS check and to make it more approachable. I think formalizing it in full will be a pretty nontrivial undertaking, but formalizing isolated components feels tractable, and is in fact where I'm currently directing a lot of my time and funding.
"gesture at something formal" -- not in the way of the "grammar" it isn't. I've seen rough mathematics and proof sketches, especially around formal grammars. This isn't that, and it isn't trying to be.
[...]
Nonsense! If Chris has an alternative to well-ordering, that's of general mathematical interest! He would make a splash simply writing that up formally on its own, without dragging the rest of his framework along with it.
My claim was specifically around whether it would be worth people's time to attempt to decipher Chris's written work, not whether there's value in Chris's work that's of general mathematical interest. If I succeed at producing formal artifacts inspired by Chris's work, written in a language that is far more approachable for general academic audiences, I would recommend for people to check those out.
That said, I am very sympathetic to the question "If Chris has such good ideas that he claims he's formalized, why hasn't he written them down formally -- or at least gestured at them formally -- in a way that most modern mathematicians or scientists can recognize? Wouldn't that clearly be in his self-interest? Isn't it pretty suspicious that he hasn't done that?"
My current understanding is that he believes that his current written work should be sufficient for modern mathematicians and scientists to understand his core ideas, and insofar as they reject his ideas, it's because of some combination of them not being intelligent and open-minded enough, which he can't do much about. I think his model is... not exactly false, but is also definitely not how I would choose to characterize most smart people who are skeptical of Chris.
To understand why Chris thinks this way, it's important to remember that he had never been acculturated into the norms of the modern intellectual elite -- he grew up in the midwest, without much affluence; he had a physically abusive stepfather he kicked out of his home by lifting weights; he was expelled from college for bureaucratic reasons, which pretty much ended his relationship with academia (IIRC); he mostly worked blue-collar jobs throughout his adult life; AND he may actually have been smarter than almost anybody he'd ever met or heard of. (Try picturing what von Neumann may have been like if he'd had the opposite of a prestigious and affluent background, and had gotten spurned by most of the intellectuals he'd talked to.) Among other things, Chris hasn't had very many intellectual peers who could gently inform him that many portions of his written work that he considers totally obvious and straightforward are actually not at all obvious for a majority of his intended audience.
On the flip side, I think this means there's a lot of low-hanging fruit in translating Chris's work into something more digestible by the modern intelletual elite.
I was hoping people other than Jessica would share some specific curated insights they got. Syndiffeonesis is in fact a good insight.
Gotcha! I'm happy to do that in a followup comment.
I'd categorize this section as "not even wrong"; it isn't doing anything formal enough to have a mistake in it.
I think it's an attempt to gesture at something formal within the framework of the CTMU that I think you can only really understand if you grok enough of Chris's preliminary setup. (See also the first part of my comment here.)
(Perhaps you'd run into issues with making the sets well-ordered, but if so he's running headlong into the same issues.)
A big part of Chris's preliminary setup is around how to sidestep the issues around making the sets well-ordered. What I've picked up in my conversations with Chris is that part of his solution involves mutually recursively defining objects, relations, and processes, in such a way that they all end up being "bottomless fractals" that cannot be fully understood from the perspective of any existing formal frameworks, like set theory. (Insofar as it's valid for me to make analogies between the CTMU and ZFC, I would say that these "bottomless fractals" violate the axiom of foundation, because they have downward infinite membership chains.)
I'm really not seeing any value in this guy's writing. Could someone who got something out of it share a couple specific insights that got from it?
I think Chris's work is most valuable to engage with for people who have independently explored philosophical directions similar to the ones Chris has explored; I don't recommend for most people to attempt to decipher Chris's work.
I'm confused why you're asking about specific insights people have gotten when Jessica has included a number of insights she's gotten in her post (e.g. "He presents a number of concepts, such as syndiffeonesis, that are useful in themselves.").
Thanks a lot for posting this, Jessica! A few comments:
It's an alternative ontology, conceiving of reality as a self-processing language, which avoids some problems of more mainstream theories, but has problems of its own, and seems quite underspecified in the document despite the use of formal notation.
I think this is a reasonable take. My own current best guess is that the contents of the document uniquely specifies a precise theory, but that it's very hard to understand what's being specified without grokking the details of all the arguments he's using to pin down the CTMU. I partly believe this because of my conversations with Chris, and I partly believe this because someone else I'd funded to review Chris's work (who had extensive prior familiarity with the kinds of ideas and arguments Chris employs) managed to make sense of most of the CTMU (including the portions using formal notation) based on Chris's written work alone, in a way that Chris has vetted over the course of numerous three-way Zoom calls.
In particular, I doubt that conspansion solves quantum locality problems as Langan suggests; conceiving of the wave function as embedded in conspanding objects seems to neglect correlations between the objects implied by the wave function, and the appeal to teleology to explain the correlations seems hand-wavey.
I'm actually not sure which quantum locality problems Chris is referring to, but I don't think the thing Chris means by "embedding the wave function in conspanding objects" runs into the problems you're describing. Insofar as one object is correlated with others via quantum entanglement, I think those other objects would occupy the same circle -- from the subtext of Diagram 11 on page 28, The result is a Venn diagram in which circles represent objects and events, or (n>1)-ary interactive relationships of objects. That is, each circle depicts the “entangled quantum wavefunctions” of the objects which interacted with each other to generate it.
In particular, I think this manifests in part as an extreme lack of humility.
I just want to note that, based on my personal interactions with Chris, I experience Chris's "extreme lack of humility" similarly to how I experience Eliezer's "extreme lack of humility":
- in both cases, I think they have plausibly calibrated beliefs about having identified certain philosophical questions that are of crucial importance to the future of humanity, that most of the world is not taking seriously,[1] leading them to feel a particular flavor of frustration that people often interpret as an extreme lack of humility
- in both cases, they are in some senses incredibly humble in their pursuit of truth, doing their utmost to be extremely honest with themselves about where they're confused
- ^
It feels worth noting that Chris Langan has written about Newcomb's paradox in 1989, and that his resolution involves thinking in terms of being in a simulation, similarly to what Andrew Critch has written about.
I've spent 40+ hours talking with Chris directly, and for me, a huge part of the value also comes from seeing how Chris synthesizes all these ideas into what appears to be a coherent framework.
Here's my current understanding of what Scott meant by "just a little off".
I think exact Bayesian inference via Solomonoff induction doesn't run into the trapped prior problem. Unfortunately, bounded agents like us can't do exact Bayesian inference via Solomonoff induction, since we can only consider a finite set of hypotheses at any given point. I think we try to compensate for this by recognizing that this list of hypotheses is incomplete, and appending it with new hypotheses whenever it seems like our current hypotheses are doing a sufficiently terrible job of explaining the input data.
One side effect is that if the true hypothesis (eg "polar bears are real") is not among our currently considered hypotheses, but our currently considered hypotheses are doing a sufficiently non-terrible job of explaining the input data (eg if the hypothesis "polar bears aren't real, but there's a lot of bad evidence suggesting that they are" is included, and the data is noisy enough that this hypothesis is reasonable), we just never even end up considering the true hypothesis. There wouldn't be accumulating likelihood ratios in favor of polar bears, because actual polar bears were never considered in the first place.
I think something similar is happening with phobias. For example, for someone with a phobia of dogs, I think the (subconscious, non-declarative) hypothesis "dogs are safe" doesn't actually get considered until the subject is well into exposure therapy, after which they've accumulated enough evidence that's sufficiently inconsistent with their prior hypotheses of dogs being scary and dangerous that they start considering alternative hypotheses.
In some sense this algorithm is "going out of its way to do something like compartmentalization", in that it's actively trying to fit all input data into its current hypotheses (/ "compartments") until this method no longer works.
Yep! I addressed this point in footnote [3].
I just want to share another reason I find this n=1 anecdote so interesting -- I have a highly speculative inside view that the abstract concept of self provides a cognitive affordance for intertemporal coordination, resulting in a phase transition in agentiness only known to be accessible to humans.
Hmm, I'm not sure I understand what point you think I was trying to make. The only case I was trying to make here was that much of our subjective experience which may appear uniquely human might stem from our langauge abilites, which seems consistent with Helen Keller undergoing a phase transition in her subjective experience upon learning a single abstract concept. I'm not getting what age has to do with this.
Questions #2 and #3 seem positively correlated – if the thing that humans have is important, it's evidence that architectural changes matter a lot.
Not necessarily. For example, it may be that language ability is very important, but that most of the heavy lifting in our language ability comes from general learning abilities + having a culture that gives us good training data for learning language, rather than from architectural changes.
I remembered reading about this a while back and updating on it, but I'd forgotten about it. I definitely think this is relevant, so I'm glad you mentioned it -- thanks!
I think this explanation makes sense, but it raises the further question of why we don't see other animal species with partial language competency. There may be an anthropic explanation here - i.e. that once one species gets a small amount of language ability, they always quickly master language and become the dominant species. But this seems unlikely: e.g. most birds have such severe brain size limitations that, while they could probably have 1% of human language, I doubt they could become dominant in anywhere near the same way we did.
Can you elaborate more on what partial language competency would look like to you? (FWIW, my current best guess is on "once one species gets a small amount of language ability, they always quickly master language and become the dominant species", but I have a lot of uncertainty. I suppose this also depends a lot on what exactly what's meant by "language ability".)
This seems like a false dichotomy. We shouldn't think of scaling up as "free" from a complexity perspective - usually when scaling up, you need to make quite a few changes just to keep individual components working. This happens in software all the time: in general it's nontrivial to roll out the same service to 1000x users.
I agree. But I also think there's an important sense in which this additional complexity is mundane -- if the only sorts of differences between a mouse brain and a human brain were the sorts of differences involved in scaling up a software service to 1000x users, I think it would be fair (although somewhat glib) to call a human brain a scaled-up mouse brain. I don't think this comparison would be fair if the sorts of differences were more like the sorts of differences involved in creating 1000 new software services.
That's one of the "unique intellectual superpowers" that I think language confers us:
On a species level, our mastery of language enables intricate insights to accumulate over generations with high fidelity. Our ability to stand on the shoulders of giants is unique among animals, which is why our culture is unrivaled in its richness in sophistication.
(I do think it helps to explicitly name our ability to learn culture as something that sets us apart, and wish I'd made that more front-and-center.)
I'm still confused about how each of the approaches would prevent us from eventually creating agents that spend 99% of their cognition acting corrigibly, while spending a well-hidden 1% of its cognition trying to sniff out whether it's in the test distribution, and executing a treacherous turn if so. The way I understand your summaries:
- Adversarial training makes it harder for the malignant 1% to do anything dangerous, since it makes it harder to tell whether it's in the test distribution. But it does little to prevent a malignant 1% from very conservatively trying (and usually failing) to ascertain whether it's in the test distribution.
- Interpretability makes it harder for the malignant 1% to be hidden, but doesn't prevent malignant cognition it can't detect. (My reading of "Gradient descent will be pushing against whatever kinds of treacherous turns that can be detected by our machinery. But it will only push for things that improve behavior on the training distribution, so a treacherous turn can only appear where it is coupled with the machinery that is producing good predictions, giving us more traction for interpretability." is completely consistent with this.)
I didn't understand what your wrote about verification well enough to have anything to say.
It does prima facie seem that an agent spending 100% of its cognition being competent and corrigible achieves higher reward than an agent that only spends 99% of its cognition being competent and corrigible, and 1% of its cognition trying (and almost always failing) to see if it's in the test distribution. Is your model that gradient descent will favor the former agent over the latter agent, making the 99%/1% agent unlikely to arise (and perhaps quantifiably so)?
The inner process may nevertheless use TDT if TDT doesn't diverge from CDT on the training distribution, or it might learn to use TDT but "look nice" so that it doesn't get selected against.
This was what I was intending to convey in assumption 3.
I'm currently intuiting that there's a broad basin of "seeming corrigible until you can perform a treacherous turn", but not a broad basin of true corrigibility. If the overseer can robustly detect incorrigible cognition in the distilled agent, I think things may be fine. But if e.g. the overseer is only inspecting the distilled agent's justifications for its behavior, and something like its verbal loop, I don't see how things can work out.
Here are my assumptions underlying this intuition:
1. Past a certain capabilities level, distilled agents can meaningfully be modeled as expected utility maximizers for short-term rewards from its overseer. (This is because expected utility maximization is instrumentally convergent, the same way that believing 51 to be composite is instrumentally convergent, as per Eliezer's comment.)
2. If the agents reason using TDT, the effective utility functions of these act-based agents will optimize for long-term instrumental preferences of their users, because in expectation, adopting this policy raises whatever short-term instrumental reward they'll get.
3. Expected utility maximizers for discounted future rewards of reward signals are hopelessly incorrigible.
4. Agents produced through the distillation process, misaligned in this way, could enter a regime of devoting 99% of their cognition to doing exactly what a corrigible agent would do, and 1% of their cognition to finding strategies for determining how to / whether to execute a treacherous turn.
5. Because of how many ways there are to hide this 1% of cognition, a complete white-box understanding of the agent's cognition is necessary to spot this sort of malign reasoning.
6. We can't achieve this level of understanding via anything like current ML transparency techniques.
Which of these assumptions do you disagree with most? Or do you agree with these assumptions, but feel optimistic that we can get good worst-case guarantees anyway?
2. How easy is it to learn to be corrigible? I'd think of this as: if we impose the extra constraint that our model behave corrigibly on all inputs, in addition to solving the object-level task well, how much bigger do we need to make the model?
My current intuition: for the same reasons it seems really hard to train models that believe 51 is prime while also performing well on object-level tasks, learning to be corrigible seems very difficult.
This does seem like a double crux; my sense is that correctly reasoning about self-modification requires a potentially complicated theory that I don't expect a general reasoning to realize it needs as soon as it becomes capable of self-modification (or creating successor agents, which I think is a subproblem of self-modification).
I share this intuition, for sufficiently complex self-modifications, with massive error bounds around what constitutes "sufficiently complex". I'm not sure if humans perform sufficiently complex self-modifications, I think our first AGis might perform sufficiently complex self-modifications, and I think AGIs undergoing a fast takeoff are most likely performing sufficiently complex self-modifications.
is simply not able to foresee the impacts of its changes and so makes them 'recklessly' (in the sense that every particular change seems worth it, even if the policy of making changes at that threshold of certainty seems likely to lead to disaster).
+100. This is why I feel queasy about "OK, I judge this self-modification to be fine" when the self-modifications are sufficiently complex, if this judgment isn't based off something like zero-shot reasoning (in which case we'd have strong reason to think that an agent following a policy of making every change it determines to be good will actually avoid disasters).
If we view the US government as a single entity, it's not clear that it would make sense to describe it as aligned with itself, under your notion of alignment. If we consider an extremely akrasiatic human, it's not clear that it would make sense to describe him as aligned with himself. The more agenty a human is, the more it seems to make sense to describe him as being aligned with himself.
If an AI assistant has a perfect model of what its operator approves of and only acts according to that model, it seems like it should qualify as aligned. But if the operator is very akrasiatic, should this AI still qualify as being aligned with the operator?
It seems to me that clear conceptual understandings of alignment, corrigibility, and benignity depend critically on a clear conceptual understanding of agency, which suggests a few things:
- Significant conceptual understanding of corrigibility is at least partially blocked on conceptual progess on HRAD. (Unless you think the relevant notions of agency can mostly be formalized with ideas outside of HRAD? Or that conceptual understandings of agency are mostly irrelevant for conceptual understandings of corrigibility?)
- Unless we have strong reasons to think we can impart the relevant notions of agency via labeled training data, we shouldn't expect to be able to adequately impart corrigibility via labeled training data.
- Without a clear conceptual notion of agency, we won't have a clear enough concept of alignment or corrigibility we can use to make worst-case bounds.
I think a lot of folks who are confused about your claims about corrigibility share my intuitions around the nature of corrigibility / the difficulty of learning corrigibility from labeled data, and I think it would shed a lot of light if you shared more of your own views on this.
I should clarify a few more background beliefs:
- I think zero-shot reasoning is probably not very helpful for the first AGI, and will probably not help much with daemons in our first AGI.
- I agree that right now, nobody is trying to (or should be trying to) build an AGI that's competently optimizing for our values for 1,000,000,000 years. (I'd want an aligned, foomed AGI to be doing that.)
- I agree that if we're not doing anything as ambitious as that, it's probably fine to rely on human input.
- I agree that if humanity builds a non-fooming AGI, they could coordinate around solving zero-shot reasoning before building a fooming AGI in a small fraction of the first 10,000 years (perhaps with the help of the first AGI), in which case we don't have to worry about zero-shot reasoning today.
- Conditioning on reasonable international coordination around AGI at all, I give 50% to coordination around intelligence explosions. I think the likelihood of this outcome rises with the amount of legitimacy zero-shot shot reasoning has at coordination time, which is my main reason for wanting to work on it today. (If takeoff is much slower I'd give something more like 80% to coordination around intelligence explosions, conditional on international coordination around AGIs.)
Let me now clarify what I mean by "foomed AGI":
- A rough summary is included in my footnote: [6] By “recursively self-improving AGI”, I’m specifically referring to an AGI that can complete an intelligence explosion within a year [or hours], at the end of which it will have found something like the optimal algorithms for intelligence per relevant unit of computation. ("Optimally optimized optimizer" is another way of putting it.)
- You could imagine analogizing the first AGI we build to the first dynamite we ever build. You could analogize a foomed AGI to a really big dynamite, but I think it's more accurate to analogize it to a nuclear bomb, given the positive feedback loops involved.
- I expect the intelligence differential between our first AGI and a foomed AGI to be numerous orders of magnitude larger than the intelligence differential between a chimp and a human.
- In this "nuclear explosion" of intelligence, I expect the equivalent of millions of years of human cognitive labor to elapse, if not many more.
In this comment thread, I was referring primarily to foomed AGIs, not the first AGIs we build. I imagine you either having a different picture of takeoff, or thinking something like "Just don't build a foomed AGI. Just like it's way too hard to build AGIs that competently optimize for our values for 1,000,000,000 years, it's way too hard to build a safe foomed AGI, so let's just not do it". And my position is something like "It's probably inevitable, and I think it will turn out well if we make a lot of intellectual progress (probably involving solutions to metaphilosophy and zero-shot reasoning, which I think are deeply related). In the meantime, let's do what we can to ensure that nation-states and individual actors will understand this point well enough to coordinate around not doing it until the time is right."
I'm happy to delve into your individual points, but before I do so, I'd like to get your sense of what you think our remaining disagreements are, and where you think we might still be talking about different things.
Corrigibility. Without corrigibility I would be just as scared of Goodhart.
This seems like it's using a bazooka to kill a fly. I'm not sure if I agree that zero-shot reasoning saves you from daemons, but even if so, why not try to attack the problem of daemons directly?
I agree that zero-shot reasoning doesn't save us from daemons by itself, and I think there's important daemon-specific research to be done independently of zero-shot reasoning. I more think that zero-shot reasoning may end up being critically useful in saving us from a specific class of daemons.
Okay, sure, but then my claim is that Solomonoff induction is _better_ than zero-shot reasoning on the axes you seem to care about, and yet it still has daemons. Why expect zero-shot reasoning to do better?
The daemons I'm focusing on here mostly arise from embedded agency, which Solomonoff induction doesn't capture at all. (It's worth nothing that I consider there to be a substantial difference between Solomonoff induction daemons and "internal politics"/"embedded agency" daemons.) I'm interested in hashing this out further, but probably at some future point, since this doesn't seem central to our disagreement.
But in scenarios where we have an AGI, yet we fail to achieve these objectives, the reason that seems most likely to me is "the AGI was incompetent at some point, made a mistake, and bad things happened". I don't know how to evaluate the probability of this and so become uncertain. But, if you are correct that we can formalize zero-shot reasoning and actually get high confidence, then the AGI could do that too. The hard problem is in getting the AGI to "want" to do that.
However, I expect that the way we actually get high confidence answers to those questions, is that we implement a control mechanism (i.e. the AI) that gets to act over the entire span of 10,000 or 1 billion years or whatever, and it keeps course correcting in order to stay on the path.
....
If you're trying to [build the spacecraft] without putting some general intelligence into it, this sounds way harder to me, because you can't build in a sufficiently general control mechanism for the spacecraft. I agree that (without access to general-intelligence-routines for the spacecraft) such a task would need very strong zero-shot reasoning. (It _feels_ impossible to me that any actual system could do this, including AGI, but that does feel like a failure of imagination on my part.)
I'm surprised by how much we seem to agree about everything you've written here. :P Let me start by clarifying my position a bit:
- When I imagine the AGI making a "plan that will work in one go", I'm not imagining it going like "OK, here's a plan that will probably work for 1,000,000,000 years! Time to take my hands off the wheel and set it in motion!" I'm imagining the plan to look more like "set a bunch of things in motion, reevaluate and update it based on where things are, and repeat". So the overall shape of this AGI's cognition will look something like "execute on some plan for a while, reevaluate and update it, execute on it again for a while, reevaluate and update it again, etc.", happening miliions or billions of times over (which seems a lot like a control mechanism that course-corrects). The zero-shot reasoning is mostly for ensuring that each step of reevaluation and updating doesn't introduce any critical errors.
- I think an AGI competently optimizing for our values should almost certainly be exploring distant galaxies for billions of years (given the availability of astronomical computing resources). On this view, building a spacecraft that can explore the universe for 1,000,000,000 years without critical malfunctions is strictly easier than building an AGI that competently optimizes for our values for 1,000,000,000 years.
- Millions of years of human cognitive labor (or much more) might happen in an intelligence explosion that occurs over the span of hours. So undergoing a safe intelligence explosion seems at least as difficult as getting an earthbound AGI doing 1,000,000 years' worth of human cognition without any catastrophic failures.
- I'm less concerned about the AGI killing its operators than I am about the AGI failing to capture a majority of our cosmic endowment. It's plausible that the latter usually leads to the former (particularly if there's a fast takeoff on Earth that completes in a few hours), but that's mostly not what I'm concerned about.
In terms of actual disagreement, I suspect I'm much more pessimistic than you about daemons taking over the control mechanism that course-corrects our AI, especially if it's doing something like 1,000,000 years' worth of human cognition, unless we can continuously zero-shot reason that this control mechanism will remain intact. (Equivalently, I feel very pessimistic about the process of executing and reevaluating plans millions/billions+ times over, unless the evaluation process is extraordinarily robust.) What's your take on this?
This proposal judges explanations by plausibility and articulateness. Truthfulness is only incidentally relevant and will be Goodharted away.
Keep in mind that the overseer (two steps forward) is always far more powerful than the agent we're distilling (one step back), is trained to not Goodhart, is training the new agent to not Goodhart (this is largely my interpretation of what corrigibility gets you), and is explicitly searching for ways in which the new agent may want to Goodhart.
I see. Given this, I think "zero-shot learning" makes sense but "zero-shot reasoning" still doesn't, since in the former "zero" refers to "zero demonstrations" and you're learning something without doing a learning process targeted at that specific thing, whereas in the latter "zero" isn't referring to anything and you're trying to get the reasoning correct in one attempt so "one-shot" is a more sensible description.
I was imagining something like "zero failed attempts", where each failed attempt approximately corresponds to a demonstration.
Are you saying that in the slow-takeoff world, we will be able to coordinate to stop AI progress after reaching AGI and then solve the full alignment problem at leisure? If so, what's your conditional probability P(successful coordination to stop AI progress | slow takeoff)?
More like, conditioning on getting international coordination after our first AGI, P(safe intelligence explosion | slow takeoff) is a lot higher, like 80%. I don't think slow takeoff does very much to help international coordination.
1. If at the time of implementing ALBA, our conceptual understanding of corrigibility is the same as it is today, how doomed would you feel?
2. How are you imagining imposing an extra constraint that our model behave corrigibly on all inputs?
3. My current best guess is that your model of how to achieve corrigibility is to train the AI on a bunch of carefully labeled examples of corrigible behavior. To what extent is this accurate?
This is all assuming an ontology where there exists a utility function that an AI is optimizing, and changes to the AI seem especially likely to change the utility function in a random direction. In such a scenario, yes, you probably should be worried.
I'm mostly concerned with daemons, not utility functions changing in random directions. If I knew that corrigibility were robust and that a corrigible AI would never encounter daemons, I'd feel pretty good about it recursively self-improving without formal zero-shot reasoning.
You could worry about daemons exploiting these bugs under this view. I think this is a reasonable worry, but don't expect formalizing zero-shot reasoning to help with it. It seems to me that daemons occur by falling into a local optimum when you are trying to optimize for doing some task -- the daemon does that task well in order to gain influence, and then backstabs you. This can arise both in ideal zero-shot reasoning, and when introducing approximations to it (as we will have to do when building any practical system).
I'm imagining the AI zero-shot reasoning about the correctness and security of its source code (including how well it's performing zero-shot reasoning), making itself nigh-impossible for daemons to exploit.
In particular, the one context where we're most confident that daemons arise is Solomonoff induction, which is one of the best instances of formalizing zero-shot reasoning that we have. Solomonoff gives you strong guarantees, of the sort you can use in proofs -- and yet, daemons arise.
I think of Solomonoff induction less as a formalization of zero-shot reasoning, and more as a formalization of some unattainable ideal of rationality that will eventually lead to better conceptual understandings of bounded rational agents, which will in turn lead to progress on formalizing zero-shot reasoning.
I would be very surprised if we were able to handle daemons without some sort of daemon-specific research.
In my mind, there's no clear difference between preventing daemons and securing complex systems. For example, I think there's a fundamental similarity between the following questions:
- How can we build an organization that we trust to optimize for its founders' original goals for 10,000 years?
- How can ensure a society of humans flourishes for 1,000,000,000 years without falling apart?
- How can we build an AGI which, when run for 1,000,000,000 years, still optimizes for its original goals with > 99% probability? (If it critically malfunctions, e.g. if it "goes insane", it will not be optimizing for its original goals.)
- How can we build an AGI which, after undergoing an intelligence explosion, still optimizes for its original goals with > 99% probability?
I think of AGIs as implementing miniature societies teeming with subagents that interact in extraordinarily sophisticated ways (for example they might play politics or Goodhart like crazy). On this view, ensuring the robustness of an AGI entails ensuring the robustness of a society at least as complex as human society, which seems to me like it requires zero-shot reasoning.
It seems like a simpler task would be building a spacecraft that can explore distant galaxies for 1,000,000,000 years without critically malfunctioning (perhaps with the help of self-correction mechanisms). Maybe it's just a failure of my imagination, but I can't think of any way to accomplish even this task without delegating it to a skilled zero-shot reasoner.
Why "zero-shot"? You're talking about getting something right in one try, so wouldn't "one-shot" make more sense?
I've flip-flopped between "one-shot" and "zero-shot". I'm calling it "zero-shot" in analogy with zero-shot learning, which refers to the ability to perform a task after zero demonstrations. "One-shot reasoning" probably makes more sense to folks outside of ML.
I think this paragraph gives an overly optimistic impression of how much progress has been made. We are still very confused about what probabilities really are, we haven't made any progress on the problem of Apparent Unformalizability of “Actual” Induction, and decision theory seems to have mostly stalled since about 8 years ago (the MIRI paper you cite does not seem to represent a substantial amount of progress over UDT 1.1).
I used "substantial progress" to mean "real and useful progress", rather than "substantial fraction of the necessary progress". Most of my examples happened in the eary to mid-1900s, suggesting that if we continue at that rate we might need at least another century.
This isn't obvious to me. Can you explain why you think this?
I'd feel much better about delegating the problem to a post-AGI society, because I'd expect such a society to be far more stable if takeoff is slow, and far more capable of taking its merry time to solve the full problem in earnest. (I think it will be more stable because I think it would be much harder for a single actor to attain a decisive strategic advantage over the rest of the world.)
To clarify: your position is that 100,000 scientists thinking for a week each, one after another, could not replicate the performance of one scientist thinking for 1 year?
Actually I would be surprised if that's the case, and I think it's plausible that large teams of scientists thinking for one week each could safely replicate arbitrary human intellectual progress.
But if you replaced 100,000 scientists thinking for a week each with 1,000,000,000,000 scientists thinking for 10 minutes each, I'd feel more skeptical. In particular I think 10,000,000 10-minute scientists can't replicate the performance of one 1-week scientist, unless the 10-minute scientists become human transistors. In my mind there isn't a qualitative difference between this scenario and the low-bandwidth oversight scenario. It's specifically dealing with human transistors that I worry about.
I also haven't thought too carefully about the 10-minute-thought threshold in particular and wouldn't be too surprised if I revised my view here. But if we replaced "10,000,000 10-minute scientists" with "arbitrarily many 2-minute scientists" I would even more think we couldn't assemble the scientists safely.
I'm assuming in all of this that the scientists have the same starting knowledge.
There's an old SlateStarCodex post that's a reasonable intuition pump for my perspective. It seems to me that the HCH-scientists' epistemic processis fundamentally similar to that of the alchemists. And the alchemists' thoughts were constrained by their lifespan, which they partially overcame by distilling past insights to future generations of alchemists. But there still remained massive constraints on their thoughts, and I imagine qualitatively similar constraints present for HCH's.
I also imagine them to be far more constraining if "thought-lifespans" shrank from ~30 years to ~30 minutes. But "thought-lifespans" on the order of ~1 week might be long enough that the overhead from learning distilled knowledge (knowledge = intellectual progress from other parts of the HCH, representing maybe decades or centuries of human reasoning) is small enough (on the order of a day or two?) that individual scientists can hold in their heads all the intellectual progress made thus far and make useful progress on top of that, without any knowledge having to be distributed across human transistors.
I don't understand at all how that could be true for brain uploading at the scale of a week vs. year.
Solving this problem considering multiple possible approaches. Those can't be decomposed with 100% efficiency, but it sure seems like they can be split up across people.
Evaluating an approach requires considering a bunch of different possible constraints, considering a bunch of separate steps, building models of relevant phenomena, etc.
Building models requires considering several hypotheses and modeling strategies. Evaluating how well a hypothesis fits the data involves considering lots of different observations. And so on.
I agree with all this.
EDIT: In summary, my view is that:
- if all the necessary intellectual progress can be distilled into individual scientists' heads, I feel good about HCH making a lot of intellectual progress
- if the agents are thinking long enough (1 week seems long enough to me, 30 minutes doesn't), this distillation can happen.
- if this distillation doesn't happen, we'd have to end up doing a lot of cognition on "virtual machines", and cognition on virtual machines is unsafe.
You're right -- I edited my comment accordingly. But my confusion still stands. Say the problem is "figure out how to upload a human and run him at 10,000x". On my current view:
(1) However you decompose this problem, you'd need something equivalent to at least 1 year's worth of a competent scientist doing general reasoning to solve this problem.
(2) In particular, this general reasoning would require the ability to accumulate new knowledge and synthesize it to make novel inferences.
(3) This sort of reasoning would end up happening on a "virtual machine AGI" built out of "human transistors".
(4) Unless we know how to ensure cognition is safe (e.g. daemon-free) we wouldn't know how to make safe "virtual machine AGI's".
(5) So either we aren't able to perform this reasoning (because it's unsafe and recognized as such), or we perform it anyway unsafely, which may lead to catastrophic outcomes.
Which of these points would you say you agree with? (Alternatively, if my picture of the situation seems totally off, could you help show me where?)
D-imitations agglomerate to sufficient cognitive power to perform a pivotal act in a way that causes the alignment of the components to be effective upon aligning the whole; and imperfect DD-imitation preserves this property.
This is the crux I currently feel most skeptical of. I don't understand how we could safely decompose the task of emulating 1 year's worth of von Neumann-caliber general reasoning on some scientific problem. (I'm assuming something like this is necessary for a pivotal act; maybe it's possible to build nanotech or whole-brain emulations without such reasoning being automated, in which case my picture for the world becomes rosier.) (EDIT: Rather than "decomposing the task of emulating a year's worth of von Neumann-caliber general reasoning", I meant to say "decomposing any problem whose solution seems to require 1 year's worth of von Neumann-caliber general reasoning".)
In particular, I'm still picturing Paul's agenda as implementing some form of HCH, and I don't understand how anything that looks like an HCH can accumulate new knowledge, synthesize it, and make new discoveries on top of it, without the HCH-humans effectively becoming "human transistors" that implement an AGI. (An analogy: the HCH-humans would be like ants; the AGI would be like a very complicated ant colony.) And unless we know how to build a safe AGI (for example we'd need to ensure it has no daemons), I don't see how the HCH-humans would know how to configure themselves into a safe AGI, so they just wouldn't (if they're benign).
Oops, I think I was conflating "corrigible agent" with "benign act-based agent". You're right that they're separate ideas. I edited my original comment accordingly.
X-and-only-X is what I call the issue where the property that's easy to verify and train is X, but the property you want is "this was optimized for X and only X and doesn't contain a whole bunch of possible subtle bad Ys that could be hard to detect formulaically from the final output of the system".
If X is "be a competent, catastrophe-free, corrigible act-based assistant", it's plausible to me that an AGI trained to do X is sufficient to lead humanity to a good outcome, even if X doesn't capture human values. For example, the operator might have the AGI develop the technology for whole brain emulations, enabling human uploads that can solve the safety problem in earnest, after which the original AGI is shut down.
Being an act-based (and thus approval-directed) agent is doing a ton of heavy lifting in this picture. Humans obviously wouldn't approve of daemons, so your AI would just try really hard to not do that. Humans obviously wouldn't approve of a Rubik's cube solution that modulates RAM to send GSM cellphone signals, so your AI would just try really hard to not do that.
I think most of the difficulty here is shoved into training an agent to actually have property X, instead of just some approximation of X. It's plausible to me that this is actually straightforward, but it also feels plausible that X is a really hard property to impart (though still much easier to impart than "have human values").
A crux for me whether property X is sufficient is whether the operator could avoid getting accidentally manipulated. (A corrigible assistant would never intentionally manipulate, but if it satisfies property X while more directly optimizing Y, it might accidentally manipulate the humans into doing some Y distinct from human values.) I feel very uncertain about this, but it currently seems plausible to me that some operators could successfully just use the assistant to solve the safety problem in earnest, and then shut down the original AGI.
D-imitations and DD-imitations robustly preserve the goodness of the people being imitated, despite the imperfection of the imitation;
My model of Paul thinks it's sufficient to train the AI's to be corrigible act-based assistants that are competent enough to help us significantly, while also able to avoid catastrophes. If possible, this would allow significant wiggle room for imperfect imitation.
Paul and I disagreed about the ease of training such assistants, and we hashed out a specific thought experiment: if we humans were trying our hardest to be competent, catastrophe-free, corrigible act-based assistants to some aliens, is there some reasonable training procedure they could give us that would enable us to significantly and non-catastrophically assist the aliens perform a pivotal act? Paul thought yes (IIRC), while I felt iffy about it. After all, we might need to understand tons and tons of alien minutiae to avoid any catastrophes, and given how different our cultures (and brains) are from theirs, it seems unlikely we'd be able to capture all the relevant minutiae.
I've since warmed up to the feasibility of this. It seems like there aren't too many ways to cause existential catastrophes, it's pretty easy to determine what things constitute existential catastrophes, and it's pretty easy to spot them in advance (at least as well as the aliens would). Yes we might still make some catastrophic mistakes, but they're likely to be benign, and it's not clear that the risk of catastrophe we'd incur is much worse than the risk the aliens would incur if a large team of them tried to execute a pivotal act. Perhaps there's still room for things like accidental mass manipulation, but this feels much less worrisome than existential catastrophe (and also seems plausibly preventable with a sufficiently competent operator).
I suspect another major crux on this point is whether there is a broad basin of corrigibility (link). If so, it shouldn't be too hard for D-imitations to be corrigible, nor for IDA to preserve corrigibility for DD-imitations. If not, it seems likely that corrigibility would be lost through distillation. I think this is also a crux for Vaniver in his post about his confusions with Paul's agenda.
I think the world will end up in a catastrophic epistemic pit. For example, if any religious leader got massively amplified, I think it's pretty likely (>50%) the whole world will just stay religious forever.
Us making progress on metaphilosophy isn't an improvement over the empowered person making progress on metaphilosophy, conditioning on the empowered person making enough progress on metaphilosophy. But in general I wouldn't trust someone to make enough progress on metaphilosophy unless they had a strong enough metaphilosophical base to begin with.
Yeah, what you described indeed matches my notion of "values-on-reflection" pretty well. So for example, I think a religious person's values-on-reflection should include valuing logical consistency and coherent logical arguments (because they do implicitly care about those in their everyday lives, even if they explicitly deny it). This means their values-on-reflection should include having true beliefs, and thus be atheistic. But I also wouldn't generally trust religious people to update away from religion if they reflected a bunch.
I wish I were clearer in my title that I'm not trying to reframe all misaligned AGI's, just a particular class of them. I agree that an AGI that fully understood your values would not optimize for them (and would not be "well-intentioned") if it had a bad goal.
That problem is basically ignorance, and so by making the PA smarter or more aware, we can solve the problem.
I think if we've correctly specified the values in an AGI, then I agree that when the AGI is smart enough it'll correctly optimize for our values. But it's not necessarily robust to scaling down, and I think it's likely to hit a weird place where it's trying and failing to optimize for our values. This post is about my intuitions for what that might look like.
Oh, I actually think that giving the 100th best person a bunch of power is probably better.than the status quo, assuming there are ~100 people who pass the bar (I also feel pessimistic about the status quo). The only reason why I think the status quo might be better is that more metaphilosophy would develop, and then whoever gets amplified would have more metaphilosophical competence to begin with, which seems safer.
Thanks a lot Ben! =D
I am somewhat hesitant to share simple intuition pumps about important topics, in case those intuition pumps are misleading.
On that note, Paul has recently written a blog post clarifying that his notion of "misaligned AI" does not coincide with what I wrote about here.
I'd say that it wouldn't appear catastrophic to the amplified human, but might be catastrophic for that human anyway (e.g. if their values-on-reflection actually look a lot like humanity's values-on-reflection, but they fail to achieve their values-on-reflection).
Those examples may be good evidence that humans have a lot of implicit knowledge, but I don't think they suggest that an AI needs to learn human representations in order to be safe.
I think my present view is something like a conjunction of:
1. An AI needs to learn human representations in order to generalize like a human does.
2. For sufficiently general and open-ended tasks, the AI will need to generalize like a human does in order to be safe. Otherwise, the default is to expect a (possibly existential) catastrophe from a benign failure.
3. For a very broad range of narrow tasks, the AI does not need to generalize like a human does in order to be safe (or, it's easy for it to generalize like a human). Go is in this category, ZFC theorem-provers are probably in this category, and I can imagine a large swath of engineering automation also falls into this category.
4. To the extent that "general and open-ended tasks" can be broken down into narrow tasks that don't require human generalization, they don't require human generalization to learn safely.
My current understanding is that we agree on (3) and (4), and that you either think that (2) is false, or that it's true but the bar for "sufficiently general and open-ended" is really high, and tasks like achieving global stability can be safely broken down into safe narrow tasks. Does this sound right to you?
I'm confused about your thoughts on (1).
(I'm currently rereading your blog posts to get a better sense of your models of how you think broad and general tasks can get broken down into narrow ones.)
I replied about (2) and black swans in a comment way down.
I'm curious to hear more about your thoughts about (4).
To flesh out my intuitions around (4) and (5): I think there are many tasks where a high-dimensional and difficult to articulate piece of knowledge is critical for completing the task. For example:
- if you're Larry or Sergey trying to hire a new CEO, you need your new CEO to be a culture fit. Which in this case means something like "being technical, brilliant, and also a hippie at heart". It's really, really hard to communicate this to a slick MBA. Especially the "be a hippie at heart" part. Maybe if you sent them to Burning Man and had them take a few drugs, they'd grok it?
- if you're Bill Gates hiring a new CEO, you should make sure your new CEO is also a developer at heart, not a salesman. Otherwise, you might hire Steve Ballmer, who drove Microsoft's revenues through the roof for a few years, but also had little understanding of developers (for example he produced an event where he celebrated developers in a way developers don't tend to like being celebrated). This led to an overall trend of the company losing its technical edge, and thus its competitive edge... this was all while Ballmer had worked with Gates at Microsoft for two decades. If Ballmer was a developer, he may have been able avoid this, but he very much wasn't.
- if you're a self-driving car engineer delegating image classification to a modern-day neural net, you'd really want its understanding of what the classifications mean to match yours, lest they be susceptible to clever adversarial attacks. Humans understand the images to represent projections of crisp three-dimensional objects that exist in a physical world; image classifiers don't, which is why they can get fooled so easily by overlays of random patterns. Maybe it's possible to replicate this understanding without being an embodied agent, but it seems you'd need something beyond training a big neural net on a large collection of images, and making incremental fixes.
- if you're a startup trying to build a product, it's very hard to do so correctly if you don't have a detailed implicit model of your users' workflows and pain points. It helps a lot to talk to them, but even then, you may only be getting 10% of the picture if you don't know what it's like to be them. Most startups die by not having this picture, flying blind, and failing to acquire any users.
- if you're trying to help your extremely awkward and non-neurotypical friend find a romantic partner, you might find it difficult to convey what exactly is so bad about carrying around slips of paper with clever replies, and pulling them out and reading from them when your date says something you don't have a reply to. (It's not that hard to convey why doing this particuar thing is bad. It's hard to convey what exactly about it is bad, that would have him properly generalize and avoid all classes of mistakes like this going forward, rather than just going like "Oh, pulling out slips of paper is jarring and might make her feel bad, so I'll stop doing this particular thing".) (No, I did not make up this example.)
In these sorts of situations, I wouldn't trust an AI to capture my knowledge/understanding. It's often tacit and perceptual, it's often acquired through being a human making direct contact with reality, and it might require a human cognitive architecture to even comprehend in the first place. (Hence my claims that proper generalization requires having the same ontologies as the overseer, which they obtained from their particular methods of solving a problem.)
In general, I feel really sketched about amplifying oversight, if the mechanism involves filtering your judgment through a bunch of well-intentioned non-neurotypical assistants, since I'd expect the tacit understandings that go into your judgment to get significantly distorted. (Hence my curiosity about whether you think we can avoid the judgment getting significantly distorted, and/or whether you think we can do fine even with significantly distorted judgment.)
It's also pretty plausible that I'm talking completely past you here; please let me know if this is the case.
I really like that list of points! Not that I'm Rob, but I'd mentally classified each of those as alignment failures, and the concern I was trying to articulate was that, by default, I'd expect an AI trying to do the right thing will make something like one of these mistakes. Those are good examples of the sorts of things I'd be scared of if I had a well-intentioned non-neurotypical assistant. Those are also what I was referring to when I talked about "black swans" popping up. And when I said:
2. Corrigibility depends critically on high-impact calibration (when your AI is considering doing a high-impact thing, it's critical that it knows to check that action with you).
I meant that, if an AI trying to do the right thing was considering one of these actions, for it to be safe it should consult you before going ahead with any one of these. (I didn't mean "the AI is incorrigible if it's not high-impact calibrated", I meant "the AI, even if corrigible, would be unsafe it's not high-impact calibrated".)
If these kinds of errors are included in "alignment," then I'd want some different term that referred to the particular problem of building AI that was trying to do the right thing, without including all of the difficulty of figuring out what is right (except insofar as "figure out more about what is right" is one way to try to build an AI that is trying to do the right thing.)
I think I understand your position much better now. The way I've been describing "ability to figure out what is right" is "metaphilosophical competence", and I currently take the stance that an AI trying to do the right thing will by default be catastrophic if it's not good enough at figuring out what is right, even if it's corrigible.