Posts
Comments
The constant bound isn't not that relevant just because of the in principal unbounded size, it also doesn't constrain the induced probabilities in the second coding scheme much at all. It's an upper bound on the maximum length, so you can still have the weightings in codings scheme B differ differ in relative length by a ton, leading to wildly different priors
Your phrasing here is vague and somewhat convoluted, so I have difficulty telling if what you say is simply misleading, or false. Regardless:
If you have UTM1 and UTM2, there is a constant-length prefix P such that UTM1 with P prepended to some further bitstring as input will compute whatever UTM2 computes with only that bitstring as input; we can say of P that it "encodes" UTM2 relative to UTM1. This being the case, each function indexed by UTM1 differs from its counterpart for UTM2 by a maximum of len(P), because whenever it's the case that a given function would otherwise be encoded in UTM1 by a bitstring longer than len(P + [the shortest bitstring encoding the function in UTM2]), the prefixed version of that function simply is the shortest bitstring encoding it in UTM1.
One of the consequences of this, however, is that this prefix-based encoding method is only optimal for functions whose prefix-free encodings (i.e. encodings that cannot be partitioned into substrings such that one of the substrings encodes another UTM) in UTM1 and UTM2 differ in length by more than len(P). And, since len(P) is a measure of UTM2's complexity relative to UTM1, it follows directly that, for a UTM2 whose "coding scheme" is such that a function whose prefix-free encoding in UTM2 differs in length from its prefix-free encoding in UTM1 by some large constant (say, ~2^10^80), len(P) itself must be on the order of 2^10^80—in other words, UTM2 must have an astronomical complexity relative to UTM1.
I have no idea how you're getting to this, not sure if it's claiming a formal result or just like a hunch. But I disagree both that there is a neat correspondence between a system being physically realizable and its having a concise implementation as a TM. Even granting that point, I don't think that nearly all or even most of these physically realisable systems will behave identically or even similarly w.r.t. how they assign codes to "natural" optimization criteria
For any physically realizable universal computational system, that system can be analogized to UTM1 in the above analysis. If you have some behavioral policy that is e.g. deontological in nature, that behavioral policy can in principle be recast as an optimization criterion over universe histories; however, this criterion will in all likelihood have a prefix-free description in UTM1 of length ~2^10^80. And, crucially, there will be no UTM2 in whose encoding scheme the criterion in question has a prefix-free description of much less than ~2^10^80, without that UTM2 itself having a description complexity of ~2^10^80 relative to UTM1—meaning, there is no physically realizable system that can implement UTM2.
All possible encoding schemes / universal priors differ from each other by at most a finite prefix. You might think this doesn't achieve much, since the length of the prefix can be in principle unbounded; but in practice, the length of the prefix (or rather, the prior itself) is constrained by a system's physical implementation. There are some encoding schemes which neither you nor any other physical entity will ever be able to implement, and so for the purposes of description length minimization these are off the table. And of the encoding schemes that remain on the table, virtually all of them will behave identically with respect to the description lengths they assign to "natural" versus "unnatural" optimization criteria.
It looks to me like the "updatelessness trick" you describe (essentially, behaving as though certain non-local branches of the decision tree are still counterfactually relevant even though they are not — although note that I currently don't see an obvious way to use that to avoid the usual money pump against intransitivity) recovers most of the behavior we'd see under VNM anyway; and so I don't think I understand your confusion re: VNM axioms.
E.g. can you give me a case in which (a) we have an agent that exhibits preferences against whose naive implementation there exists some kind of money pump (not necessarily a repeatable one), (b) the agent can implement the updatelessness trick in order to avoid the money pump without modifying their preferences, and yet (c) the agent is not then representable as having modified their preferences in the relevant way?
I think I might be missing something, because the argument you attribute to Dávid still looks wrong to me. You say:
The entropy of the simulators’ distribution need not be more than the entropy of the (square of the) wave function in any relevant sense. Despite the fact that subjective entropy may be huge, physical entropy is still low (because the simulations happen on a high-amplitude ridge of the wave function, after all).
Doesn't this argument imply that the supermajority of simulations within the simulators' subjective distribution over universe histories are not instantiated anywhere within the quantum multiverse?
I think it does. And, if you accept this, then (unless for some reason you think the simulators' choice of which histories to instantiate is biased towards histories that correspond to other "high-amplitude ridges" of the wave function, which makes no sense because any such bias should have already been encoded within the simulators' subjective distribution over universe histories) you should also expect, a priori, that the simulations instantiated by the simulators should not be indistinguishable from physical reality, because such simulations comprise a vanishingly small proportion of the simulators' subjective probability distribution over universe histories.
What this in turn means, however, is that prior to observation, a Solomonoff inductor (SI) must spread out much of its own subjective probability mass across hypotheses that predict finding itself within a noticeably simulated environment. Those are among the possibilities it must take into account—meaning, if you stipulate that it doesn't find itself in an environment corresponding to any of those hypotheses, you've ruled out all of the "high-amplitude ridges" corresponding to instantiated simulations in the crossent of the simulators' subjective distribution and reality's distribution.
We can make this very stark: suppose our SI finds itself in an environment which, according to its prior over the quantum multiverse, corresponds to one high-amplitude ridge of the physical wave function, and zero high-amplitude ridges containing simulators that happened to instantiate that exact environment (either because no branches of the quantum multiverse happened to give rise to simulators that would have instantiated that environment, or because the environment in question simply wasn't a member of any simulators' subjective distributions over reality to begin with). Then the SI would immediately (correctly) conclude that it cannot be in a simulation.
Now, of course, the argument as I've presented it here is heavily reliant on the idea of our SI being an SI, in such a way that it's not clear how exactly the argument carries over to the logically non-omniscient case. In particular, it relies on the SI being capable of discerning differences between very good simulations and perfect simulations, a feat which bounded reasoners cannot replicate; and it relies on the notion that our inability as bounded reasoners to distinguish between hypotheses at this level of granularity is best modeled in the SI case by stipulating that the SI's actual observations are in fact consistent with its being instantiated within a base-level, high-amplitude ridge of the physical wave function—i.e. that our subjective inability to tell whether we're in a simulation should be viewed as analogous to an SI being unable to tell whether it's in a simulation because its observations actually fail to distinguish. I think this is the relevant analogy, but I'm open to being told (by you or by Dávid) why I'm wrong.
The AI has a similarly hard time to the simulators figuring out what's a plausible configuration to arise from the big bang. Like the simulators have an entropy N distribution of possible AIs, the AI itself also has an entropy N distribution for that. So it's probability that it's in a real Everett branch is not p, but p times 2^-N, as it has only a 2^-N prior probability that the kind of word it observes is the kind of thing that can come up in a real Everett branch. So it's balanced out with the simulation hypothesis, and as long as the simulators are spending more planets, that hypothesis wins.
If I imagine the AI as a Solomonoff inductor, this argument looks straightforwardly wrong to me: of the programs that reproduce (or assign high probability to, in the setting where programs produce probabilistic predictions of observations) the AI's observations, some of these will do so by modeling a branching quantum multiverse and sampling appropriately from one of the branches, and some of them will do so by modeling a branching quantum multiverse, sampling from a branch that contains an intergalactic spacefaring civilization, locating a specific simulation within that branch, and sampling appropriately from within that simulation. Programs of the second kind will naturally have higher description complexity than programs of the first kind; both kinds feature a prefix that computes and samples from the quantum multiverse, but only the second kind carries out the additional step of locating and sampling from a nested simulation.
(You might object on the grounds that there are more programs of the second kind than of the first kind, and the probability that the AI is in a simulation at all requires summing over all such programs, but this has to be balanced against the fact most if not all of these programs will be sampling from branches much later in time than programs of the first type, and will hence be sampling from a quantum multiverse with exponentially more branches; and not all of these branches will contain spacefaring civilizations, or spacefaring civilizations interested in running ancestor simulations, or spacefaring civilizations interested in running ancestor simulations who happen to be running a simulation that exactly reproduces the AI's observations. So this counter-counterargument doesn't work, either.)
These two kinds of “learning” are not synonymous. Adaptive systems “learn” things, but they don’t necessarily “learn about” things; they don’t necessarily have an internal map of the external territory. (Yes, the active inference folks will bullshit about how any adaptive system must have a map of the territory, but their math does not substantively support that interpretation.) The internal heuristics or behaviors “learned” by an adaptive system are not necessarily “about” any particular external thing, and don’t necessarily represent any particular external thing.
I think I am confused both about whether I think this is true, and about how to interpret in such a way that it might be true. Could you go into more detail on what it means for a learner to learn something without there being some representational semantics that could be used to interpret what it's learned, even if the learner itself doesn't explicitly represent those semantics? Or is the lack of explicit representation actually the core substance of the claim here?
It seems the SOTA for training LLMs has (predictably) pivoted away from pure scaling of compute + data, and towards RL-style learning based on (synthetic?) reasoning traces (mainly CoT, in the case of o1). AFAICT, this basically obviates safety arguments that relied on "imitation" as a key source of good behavior, since now additional optimization pressure is being applied towards correct prediction rather than pure imitation.
Strictly speaking, this seems very unlikely, since we know that e.g. CoT increases the expressive power of Transformers.
Ah, yeah, I can see how I might've been unclear there. I was implicitly taking CoT into account when I talked about the "base distribution" of the model's outputs, as it's essentially ubiquitous across these kinds of scaffolding projects. I agree that if you take a non-recurrent model's O(1) output and equip it with a form of recurrent state that you permit to continue for O(n) iterations, that will produce a qualitatively different distribution of outputs than the O(1) distribution.
In that sense, I readily admit CoT into the class of improvements I earlier characterized as "shifted distribution". I just don't think this gets you very far in terms of the overarching problem, since the recurrent O(n) distribution is the one whose output I find unimpressive, and the method that was used to obtain it from the (even less impressive) O(1) distribution is a one-time trick.[1]
And also intuitively, I expect, for example, that Sakana's agent would be quite a bit worse without access to Semantic search for comparing idea novelty; and that it would probably be quite a bit better if it could e.g. retrieve embeddings of full paragraphs from papers, etc.
I also agree that another way to obtain a higher quality output distribution is to load relevant context from elsewhere. This once more seems to me like something of a red herring when it comes to the overarching question of how to get an LLM to produce human- or superhuman-level research; you can load its context with research humans have already done, but this is again a one-time trick, and not one that seems like it would enable novel research built atop the human-written research unless the base model possesses a baseline level of creativity and insight, etc.[2]
If you don't already share (or at least understand) a good chunk of my intuitions here, the above probably sounds at least a little like I'm carving out special exceptions: conceding each point individually, while maintaining that they bear little on my core thesis. To address that, let me attempt to put a finger on some of the core intuitions I'm bringing to the table:
On my model of (good) scientific research de novo, a lot of key cognitive work occurs during what you might call "generation" and "synthesis", where "generation" involves coming up with hypotheses that merit testing, picking the most promising of those, and designing a robust experiment that sheds insight; "synthesis" then consists of interpreting the experimental results so as to figure out the right takeaway (which very rarely ought to look like "we confirmed/disconfirmed the starting hypothesis").
Neither of these steps are easily transmissible, since they hinge very tightly on a given individual's research ability and intellectual "taste"; and neither of them tend to end up very well described in the writeups and papers that are released afterwards. This is hard stuff even for very bright humans, which implies to me that it requires a very high quality of thought to manage consistently. And it's these steps that I don't think scaffolding can help much with; I think the model has to be smart enough, at baseline, that its landscape of cognitive reachability contains these kinds of insights, before they can be elicited via an external method like scaffolding.[3]
I'm not sure whether you could theoretically obtain greater benefits from allowing more than O(n) iterations, but either way you'd start to bump up against context window limitations fairly quickly. ↩︎
Consider the extreme case where we prompt the model with (among other things) a fully fleshed out solution to the AI alignment problem, before asking it to propose a workable solution to the AI alignment problem; it seems clear enough that in this case, almost all of the relevant cognitive work happened before the model even received its prompt. ↩︎
I'm uncertain-leaning-yes on the question of whether you can get to a sufficiently "smart" base model via mere continued scaling of parameter count and data size; but that connects back to the original topic of whether said "smart" model would need to be capable of goal-directed thinking, on which I think I agree with Jeremy that it would; much of my model of good de novo research, described above, seems to me to draw on the same capabilities that characterize general-purpose goal-direction. ↩︎
And I suspect we probably can, given scaffolds like https://sakana.ai/ai-scientist/ and its likely improvements (especially if done carefully, e.g. integrating something like Redwood's control agenda, etc.). I'd be curious where you'd disagree (since I expect you probably would) - e.g. do you expect the AI scientists become x-risky before they're (roughly) human-level at safety research, or they never scale to human-level, etc.?
Jeremy's response looks to me like it mostly addresses the first branch of your disjunction (AI becomes x-risky before reaching human-level capabilities), so let me address the second:
I am unimpressed by the output of the AI scientist. (To be clear, this is not the same thing as being unimpressed by the work put into it by its developers; it looks to me like they did a great job.) Mostly, however, the output looks to me basically like what I would have predicted, on my prior model of how scaffolding interacts with base models, which goes something like this:
A given model has some base distribution on the cognitive quality of its outputs, which is why resampling can sometimes produce better or worse responses to inputs. What scaffolding does is to essentially act as a more sophisticated form of sampling based on redundancy: having the model check its own output, respond to that output, etc. This can be very crudely viewed as an error correction process that drives down the probability that a "mistake" at some early token ends up propagating throughout the entirety of the scaffolding process and unduly influencing the output, which biases the quality distribution of outputs away from the lower tail and towards the upper tail.
The key moving piece on my model, however, is that all of this is still a function of the base distribution—a rough analogy here would be to best-of-n sampling. And the problem with best-of-n sampling, which looks to me like it carries over to more complicated scaffolding, is that as n increases, the mean of the resulting distribution increases as a sublinear (actually, logarithmic) function of n, while the variance decreases at a similar rate (but even this is misleading, since the resulting distribution will have negative skew, meaning variance decreases more rapidly in the upper tail than in the lower tail).
Anyway, the upshot of all of this is that scaffolding cannot elicit capabilities that were not already present (in some strong sense) in the base model—meaning, if the base models in question are strongly subhuman at something like scientific research (which it presently looks to me like they still are), scaffolding will not bridge that gap for them. The only thing that can close that gap without unreasonably large amounts of scaffolding, where "unreasonable" here means something a complexity theorist would consider unreasonable, is a shifted base distribution. And that corresponds to the kind of "useful [superhuman] capabilities" Jeremy is worried about.
I'm interested! Also curious as to how this is implemented; are you using retrieval-augmented generation, and if so, with what embeddings?
Epistemic status: exploratory, "shower thought", written as part of a conversation with Claude:
For any given entity (broadly construed here to mean, essentially, any physical system), it is possible to analyze that entity as follows:
Define the set of possible future trajectories that entity might follow, according to some suitably uninformative ignorance prior on its state and (generalized) environment. Then ask, of that set, whether there exists some simple, obvious, or otherwise notable prior on the set in question, that assigns probabilities to various member trajectories in such a way as to establish an upper level set of some kind. Then ask, of that upper level set, how large it is relative to the size of the set as a whole, and (relatedly) how large the difference is between the probability of that upper set's least probable member, and its most probable nonmember. (If you want to conceptualize these sets as infinite and open—although it's unclear to me that one needs to conceptualize them this way—then you can speak instead of "infimum" and "supremum".)
The claim is that, for some specific kinds of system, there will be quite a sharp difference between its upper level set and its lower level set, constituting a "plausibility gap": trajectories within the upper set are in some sense "plausible" ways of extrapolating the system forward in time. And then the relative size of that upper set becomes relevant, because it indicates how tightly constrained the system's time-evolution is by its present state (and environment). So, the claim is that there are certain systems for which their forwards time-evolution is very tightly constrained indeed, and these systems are "agents"; and there are systems for which barely any upper level set exists, and these are "simplistic" entities whose behavior is essentially entropic. And humans (seem to me to) occupy a median position between these two extremes.
One additional wrinkle, however, is that "agency", as I've defined it here, may additionally play the role of a (dynamical system) attractor: entities already close to having full agency will be more tightly constrained in their future evolution, generally in the direction of becoming ever more agentic; meanwhile, entirely inanimate systems are not at all pulled in the direction of becoming more constrained or agentic; they are outside of the agency attractor's basin of attraction. However, humans, if they indeed exist at some sort of halfway point between fully coherent agency and a complete lack of coherence, are left interestingly placed under this framing: we would exist at the boundary of the agency attractor's basin of attraction. And since many such boundaries are fundamentally fractal or chaotic in nature, that could have troubling implications for the trajectories of points along those boundaries trying to reach reflective equilibrium, as it were.
The rule of thumb test I tend to use to assess proposed definitions of agency (at least from around these parts) is whether they'd class a black hole as an agent. It's not clear to me whether this definition does; I would have said it very likely does based on everything you wrote, except for this one part here:
A cubic meter of rock has a persistent boundary over time, but no interior, states in an informational sense and therefore are not agents. To see they have no interior, note that anything that puts information into the surface layer of the rock transmits that same information into the very interior (vibrations, motion, etc).
I think I don't really understand what is meant by "no interior" here, or why the argument given supports the notion that a cubic meter of rock has no interior. You can draw a Markov boundary around the rock's surface, and then the interior state of the rock definitely is independent of the exterior environment conditioned on said boundary, right?
If I try very hard to extract a meaning out of the quoted paragraph, I might guess (with very low confidence) that what it's trying to say is that a rock's internal state has a one-to-one relation with the external forces or stimuli that transmit information through its surface, but in this case a black hole passes the test, in that the black hole's internal state definitely is not one-to-one with the information entering through its event horizon. In other words, if my very low-confidence understanding of the quoted paragraph is correct, then black holes are classified as agents under this definition.
(This test is of interest to me because black holes tend to pass other, potentially related definitions of agency, such as agency as optimization, agency as compression, etc. I'm not sure whether this says that something is off with our intuitive notion of agency, that something is off with our attempts at rigorously defining it, or simply that black holes are a special kind of "physical agent" built in-to the laws of physics.)
How is a Bayesian agent supposed to modify priors except by updating on the basis of evidence?
They're not! But humans aren't ideal Bayesians, and it's entirely possible for them to update in a way that does change their priors (encoded by intuitions) moving forward. In particular, the difference between having updated one's intuitive prior, and keeping the intuitive prior around but also keeping track of a different, consciously held posterior, is that the former is vastly less likely to "de-update", because the evidence that went into the update isn't kept around in a form that subjects it to (potential) refutation.
(IIRC, E.T. Jaynes talks about this distinction in Chapter 18 of Probability Theory: The Logic of Science, and he models it by introducing something he calls an A_p distribution. His exposition of this idea is uncharacteristically unclear, and his A_p distribution looks basically like a beta distribution with specific values for α and β, but it does seem to capture the distinction I see between "intuitive" updating versus "conscious" updating.)
There's also a failure mode of focusing on "which arguments are the best" instead of "what is actually true". I don't understand this failure mode very well, except that I've seen myself and others fall into it. Falling into it looks like focusing a lot on specific arguments, and spending a lot of time working out what was meant by the words, rather than feeling comfortable adjusting arguments to fit better into your own ontology and to fit better with your own beliefs.
My sense is that this is because different people have different intuitive priors, and process arguments (mostly) as a kind of Bayesian evidence that updates those priors, rather than modifying the priors (i.e. intuitions) directly.
Eliezer in particular strikes me as having an intuitive prior for AI alignment outcomes that looks very similar to priors for tasks like e.g. writing bug-free software on the first try, assessing the likelihood that a given plan will play out as envisioned, correctly compensating for optimism bias, etc. which is what gives rise to posts concerning concepts like security mindset.
Other people don't share this intuitive prior, and so have to be argued into it. To such people, the reliability of the arguments in question is actually critical, because if those arguments turn out to have holes, that reverts the downstream updates and restores the original intuitive prior, whatever it looked like—kind of like a souped up version of the burden of proof concept, where the initial placement of that burden is determined entirely via the intuitive judgement of the individual.
This also seems related to why different people seem to naturally gravitate towards either conjunctive or disjunctive models of catastrophic outcomes from AI misalignment: the conjunctive impulse stems from an intuition that AI catastrophe is a priori unlikely, and so a bunch of different claims have to hold simultaneously in order to force a large enough update, whereas the disjunctive impulse stems from the notion that any given low-level claim need not be on particularly firm ground, because the high-level thesis of AI catastrophe robustly manifests via different but converging lines of reasoning.
See also: the focus on coherence, where some people place great importance on the question of whether VNM or other coherence theorems show what Eliezer et al. purport they show about superintelligent agents, versus the competing model wherein none of these individual theorems are important in their particulars, so much as the direction they seem to point, hinting at the concept of what idealized behavior with respect to non-gerrymandered physical resources ought to look like.
I think the real question, then, is where these differences in intuition come from, and unfortunately the answer might have to do a lot with people's backgrounds, and the habits and heuristics they picked up from said backgrounds—something quite difficult to get at via specific, concrete argumentation.
Can we not speak of apparent coherence relative to a particular standpoint? If a given system seems to be behaving in such a way that you personally can't see a way to construct for it a Dutch book, a series of interactions with it such that energy/negentropy/resources can be extracted from it and accrue to you, that makes the system inexploitable with respect to you, and therefore at least as coherent as you are. The closer to maximal coherence a given system is, the less it will visibly depart from the appearance of coherent behavior, and hence utility function maximization; the fact that various quibbles can be made about various coherence theorems does not seem to me to negate this conclusion.
Humans are more coherent than mice, and there are activities and processes which individual humans occasionally undergo in order to emerge more coherent than they did going in; in some sense this is the way it has to be, in any universe where (1) the initial conditions don't start out giving you fully coherent embodied agents, and (2) physics requires continuity of physical processes, so that fully formed coherent embodied agents can't spring into existence where there previously were none; there must be some pathway from incoherent, inanimate matter from which energy may be freely extracted, to highly organized configurations of matter from which energy may be extracted only with great difficulty, if it can be extracted at all.
If you expect the endpoint of that process to not fully accord with the von Neumann-Morgenstein axioms, because somebody once challenged the completeness axiom, independence axiom, continuity axiom, etc., the question still remains as to whether departures from those axioms will give rise to exploitable holes in the behavior of such systems, from the perspective of much weaker agents such as ourselves. And if the answer is "no", then it seems to me the search for ways to make a weaker, less coherent agent into a stronger, more coherent agent is well-motivated, and necessary—an appeal to consequences in a certain sense, yes, but one that I endorse!
I seem to recall hearing a phrase I liked, which appears to concisely summarize the concern as: "There's no canonical way to scale me up."
Does that sound right to you?
Well, if we're following standard ML best practices, we have a train set, a dev set, and a test set. The purpose of the dev set is to check and ensure that things are generalizing properly. If they aren't generalizing properly, we tweak various hyperparameters of the model and retrain until they do generalize properly on the dev set. Then we do a final check on the test set to ensure we didn't overfit the dev set. If you forgot or never learned this stuff, I highly recommend brushing up on it.
(Just to be clear: yes, I know what training and test sets are, as well as dev sets/validation sets. You might notice I actually used the phrase "validation set" in my earlier reply to you, so it's not a matter of guessing someone's password—I'm quite familiar with these concepts, as someone who's implemented ML models myself.)
Generally speaking, training, validation, and test datasets are all sourced the same way—in fact, sometimes they're literally sourced from the same dataset, and the delineation between train/dev/test is introduced during training itself, by arbitrarily carving up the original dataset into smaller sets of appropriate size. This may capture the idea of "IID" you seem to appeal to elsewhere in your comment—that it's possible to test the model's generalization performance on some held-out subset of data from the same source(s) it was trained on.
In ML terms, what the thought experiment points to is a form of underlying distributional shift, one that isn't (and can't be) captured by "IID" validation or test datasets. The QFT model in particular highlights the extent to which your training process, however broad or inclusive from a parochial human standpoint, contains many incidental distributional correlates to your training signal which (1) exist in all of your data, including any you might hope to rely on to validate your model's generalization performance, and (2) cease to correlate off-distribution, during deployment.
This can be caused by what you call "omniscience", but it need not; there are other, more plausible distributional differences that might be picked up on by other kinds of models. But QFT is (as far as our current understanding of physics goes) very close to the base ontology of our universe, and so what is inferrable using QFT is naturally going to be very different from what is inferrable using some other (less powerful) ontology. QFT is a very powerful ontology!
If you want to call that "omniscience", you can, although note that strictly speaking the model is still just working from inferences from training data. It's just that, if you feed enough data to a model that can hold entire swaths of the physical universe inside of its metaphorical "head", pretty soon hypotheses that involve the actual state of that universe will begin to outperform hypotheses that don't, and which instead use some kind of lossy approximation of that state involving intermediary concepts like "intent", "belief", "agent", "subjective state", etc.
In principle we could construct a test set or dev set either before or after the model has been trained. It shouldn't make a difference under normal circumstances. It sounds like maybe you're discussing a scenario where the model has achieved a level of omniscience, and it does fine on data that was available during its training, because it's able to read off of an omniscient world-model. But then it fails on data generated in the future, because the translation method for its omniscient world-model only works on artifacts that were present during training. Basically, the time at which the data was generated could constitute a hidden and unexpected source of distribution shift. Does that summarize the core concern?
You're close; I'd say the concern is slightly worse than that. It's that the "future data" never actually comes into existence, at any point. So the source of distributional shift isn't just "the data is generated at the wrong time", it's "the data never gets externally generated to begin with, and you (the model) have to work with predictions of what the data counterfactually would have been, had it been generated".
(This would be the case e.g. with any concept of "human approval" that came from a literal physical human or group of humans during training, and not after the system was deployed "in the wild".)
In any case, I would argue that "accidental omniscience" characterizes the problem better than "alien abstractions". As before, you can imagine an accidentally-omniscient model that uses vanilla abstractions, or a non-omniscient model that uses alien ones.
The problem is that "vanilla" abstractions are not the most predictively useful possible abstractions, if you've got access to better ones. And models whose ambient hypothesis space is broad enough to include better abstractions (from the standpoint of predictive accuracy) will gravitate towards those, as is incentivized by the outer form of the training task. QFT is the extreme example of a "better abstraction", but in principle (if the natural abstraction hypothesis fails) there will be all sorts and shapes of abstractions, and some of them will be available to us, and some of them will be available to the model, and these sets will not fully overlap—which is a concern in worlds where different abstractions lead to different generalization properties.
I think it ought to be possible for someone to always be present. [I'm also not sure it would be necessary.]
I think I don't understand what you're imagining here. Are you imagining a human manually overseeing all outputs of something like ChatGPT, or Microsoft Copilot, before those outputs are sent to the end user (or, worse yet, put directly into production)?
[I also think I don't understand why you make the bracketed claim you do, but perhaps hashing that out isn't a conversational priority.]
As I understand this thought experiment, we're doing next-token prediction on e.g. a book written by a philosopher, and in order to predict the next token using QFT, the obvious method is to use QFT to simulate the philosopher. But that's not quite enough -- you also need to read the next token out of that QFT-based simulation if you actually want to predict it.
It sounds like your understanding of the thought experiment differs from mine. If I were to guess, I'd guess that by "you" you're referring to someone or something outside of the model, who has access to the model's internals, and who uses that access to, as you say, "read" the next token out of the model's ontology. However, this is not the setup we're in with respect to actual models (with the exception perhaps of some fairly limited experiments in mechanistic interpretability)—and it's also not the setup of the thought experiment, which (after all) is about precisely what happens when you can't read things out of the model's internal ontology, because it's too alien to be interpreted.
In other words: "you" don't read the next token out of the QFT simulation. The model is responsible for doing that translation work. How do we get it to do that, even though we don't know how to specify the nature of the translation work, much less do it ourselves? Well, simple: in cases where we have access to the ground truth of the next token, e.g. because we're having it predict an existing book passage, we simply penalize it whenever its output fails to match the next token in the book. In this way, the model can be incentivized to correctly predict whatever we want it to predict, even if we wouldn't know how to tell it explicitly to do whatever it's doing.
(The nature of this relationship—whereby humans train opaque algorithms to do things they wouldn't themselves be able to write out as pseudocode—is arguably the essence of modern deep learning in toto.)
For one thing, in a standard train/dev/test setup, the model is arguably not really doing prediction, it's doing retrodiction. It's making 'predictions' about things which already happened in the past. The final model is chosen based on what retrodicts the data the best.
Yes, this is a reasonable description to my eyes. Moreover, I actually think it maps fairly well to the above description of how a QFT-style model might be trained to predict the next token of some body of text; in your terms, this is possible specifically because the text already exists, and retrodictions of that text can be graded based on how well they compare against the ground truth.
Also, usually the data is IID rather than sequential -- there's no time component to the data points (unless it's a time-series problem, which it usually isn't).
This, on the other hand, doesn't sound right to me. Yes, there are certainly applications where the training regime produces IID data, but next-token prediction is pretty clearly not one of those? Later tokens are highly conditionally dependent on previous tokens, in a way that's much closer to a time series than to some kind of IID process. Possibly part of the disconnect is that we're imagining different applications entirely—which might also explain our differing intuitions w.r.t. deployment?
The fact that we're choosing a model which retrodicts well is why the presence/absence of a human is generally assumed to be irrelevant, and emphasizing this factor sounds wacky to my ML engineer ears.
Right, so just to check that we're on the same page: do we agree that after a (retrodictively trained) model is deployed for some use case other than retrodicting existing data—for generative use, say, or for use in some kind of online RL setup—then it'll doing something other than retrodicting? And that in that situation, the source of (retrodictable) ground truth that was present during training—whether that was a book, a philosopher, or something else—will be absent?
If we do actually agree about that, then that distinction is really all I'm referring to! You can think of it as training set versus test set, to use a more standard ML analogy, except in this case the "test set" isn't labeled at all, because no one labeled it in advance, and also it's coming in from an unpredictable outside world rather than from a folder on someone's hard drive.
Why does that matter? Well, because then we're essentially at the mercy of the model's generalization properties, in a way we weren't while it was retrodicting the training set (or even the validation set, if one of those existed). If it gets anything wrong, there's no longer any training signal or gradient to penalize it for being "wrong"—so the only remaining question is, just how likely is it to be "wrong", after being trained for however long it was trained?
And that's where the QFT model comes in. It says, actually, even if you train me for a good long while on a good amount of data, there are lots of ways for me to generalize "wrongly" from your perspective, if I'm modeling the universe at the level of quantum fields. Sure, I got all the retrodictions right while there was something to be retrodicted, but what exactly makes you think I did that by modeling the philosopher whose remarks I was being trained on?
Maybe I was predicting the soundwaves passing through a particularly region of air in the room he was located—or perhaps I was predicting the pattern of physical transistors in the segment of memory of a particular computer containing his works. Those physical locations in spacetime still exist, and now that I'm deployed, I continue to make predictions using those as my referent—except, the encodings I'm predicting there no longer resemble anything like coherent moral philosophy, or coherent anything, really.
The philosopher has left the room, or the computer's memory has been reconfigured—so what exactly are the criteria by which I'm supposed to act now? Well, they're going to be something, presumably—but they're not going to be something explicit. They're going to be something implicit to my QFT ontology, something that—back when the philosopher was there, during training—worked in tandem with the specifics of his presence, and the setup involving him, to produce accurate retrodictions of his judgements on various matters.
Now that that's no longer the case, those same criteria describe some mathematical function that bears no meaningful correspondence to anything a human would recognize, valuable or not—but the function exists, and it can be maximized. Not much can be said about what maximizing that function might result in, except that it's unlikely to look anything like "doing right according to the philosopher".
That's why the QFT example is important. A more plausible model, one that doesn't think natively in terms of quantum amplitudes, permits the possibility of correctly compressing what we want it to compress—of learning to retrodict, not some strange physical correlates of the philosopher's various motor outputs, but the actual philosopher's beliefs as we would understand them. Whether that happens, or whether a QFT-style outcome happens instead, depends in large part on the inductive biases of the model's architecture and the training process—inductive biases on which the natural abstraction hypothesis asserts a possible constraint.
I'm confused about what it means to "remove the human", and why it's so important whether the human is 'removed'.
Because the human isn't going to constantly be present for everything the system does after it's deployed (unless for some reason it's not deployed).
If I can assume that stuff, then it feels like a fairly core task, abundantly stress-tested during training, to read off the genius philosopher's spoken opinions about e.g. moral philosophy from the quantum fields. How else could quantum fields be useful for next-token predictions?
Quantum fields are useful for an endless variety of things, from modeling genius philosophers to predicting lottery numbers. If your next-token prediction task involves any physically instantiated system, a model that uses QFT will be able to predict that system's time-evolution with alacrity.
(Yes, this is computationally intractable, but we're already in full-on hypothetical land with the QFT-based model to begin with. Remember, this is an exercise in showing what happens in the worst-case scenario for alignment, where the model's native ontology completely diverges from our own.)
So we need not assume that predicting "the genius philosopher" is a core task. It's enough to assume that the model is capable of it, among other things—which a QFT-based model certainly would be. Which, not so coincidentally, brings us to your next question:
Is alignment supposed to be hard in this hypothetical because the AI can't represent human values in principle? Or is it supposed to be hard because it also has a lot of unsatisfactory representations of human values, and there's no good method for finding a satisfactory needle in the unsatisfactory haystack? Or some other reason?
Consider how, during training, the human overseer (or genius philosopher, if you prefer) would have been pointed out to the model. We don't have reliable access to its internal world-model, and even if we did we'd see blobs of amplitude and not much else. There's no means, in that setting, of picking out the human and telling the model to unambiguously defer to that human.
What must happen instead, then, is something like next-token prediction: we perform gradient descent (or some other optimization method; it doesn't really matter for the purposes of our story) on the model's outputs, rewarding it when its outputs happen to match those of the human. The hope is that this will lead, in the limit, to the matching no longer occurring by happenstance—that if we train for long enough and in a varied enough set of situations, the best way for the model to produce outputs that track those of the human is to model that human, even in its QFT ontology.
But do we know for a fact that this will be the case? Even if it is, what happens when the overseer isn't present to provide their actual feedback, as was never the case during training? What becomes the model's referent then? We'd like to deploy it without an overseer, or in situations too complex for an overseer to understand. And whether the model's behavior in those situations conforms to what the overseer would want, ideally, depends on what kinds of behind-the-scenes extrapolation the model is doing—which, if the model's native ontology is something in which "human philosophers" are not basic objects, is liable to look very weird indeed.
This sounds a lot like saying "it might fail to generalize".
Sort of, yes—but I'd call it "malgeneralization" rather than "misgeneralization". It's not failing to generalize, it's just not generalizing the way you'd want it to.
Supposing we make a lot of progress on out-of-distribution generalization, is alignment getting any easier according to you? Wouldn't that imply our systems are getting better at choosing proxies which generalize even when the human isn't 'present'?
Depends on what you mean by "progress", and "out-of-distribution". A powerful QFT-based model can make perfectly accurate predictions in any scenario you care to put it in, so it's not like you'll observe it getting things wrong. What experiments, and experimental outcomes, are you imagining here, such that those outcomes would provide evidence of "progress on out-of-distribution generalization", when fundamentally the issue is expected to arise in situations where the experimenters are themselves absent (and which—crucially—is not a condition you can replicate as part of an experimental setup)?
I'd assume that when we tell it, "optimize this company, in a way that we would accept, after a ton of deliberation", this could be instead described as, "optimize this company, in a way that we would accept, after a ton of deliberation, where these terms are described using our ontology"
The problem shows up when the system finds itself acting in a regime where the notion of us (humans) "accepting" its optimizations becomes purely counterfactual, because no actual human is available to oversee its actions in that regime. Then the question of "would a human accept this outcome?" must ground itself somewhere in the system's internal model of what those terms refer to, which (by hypothesis) need not remotely match their meanings in our native ontology.
This isn't (as much of) a problem in regimes where an actual human overseer is present (setting aside concerns about actual human judgement being hackable because we don't implement our idealized values, i.e. outer alignment), because there the system's notion of ground truth actually is grounded by the validation of that overseer.
You can have a system that models the world using quantum field theory, task it with predicting the energetic fluctuations produced by a particular set of amplitude spikes corresponding to a human in our ontology, and it can perfectly well predict whether those fluctuations encode sounds or motor actions we'd interpret as indications of approval of disapproval—and as long as there's an actual human there to be predicted, the system will do so without issue (again modulo outer alignment concerns).
But remove the human, and suddenly the system is no longer operating based on its predictions of the behavior of a real physical system, and is instead operating from some learned counterfactual representation consisting of proxies in its native QFT-style ontology which happened to coincide with the actual human's behavior while the human was present. And that learned representation, in an ontology as alien as QFT, is (assuming the falsehood of the natural abstraction hypothesis) not going to look very much like the human we want it to look like.
To the extent that I buy the story about imitation-based intelligences inheriting safety properties via imitative training, I correspondingly expect such intelligences not to scale to having powerful, novel, transformative capabilities—not without an amplification step somewhere in the mix that does not rely on imitation of weaker (human) agents.
Since I believe this, that makes it hard for me to concretely visualize the hypothetical of a superintelligent GPT+DPO agent that nevertheless only does what is instructed. I mostly don't expect to be able to get to superintelligence without either (1) the "RL" portion of the GPT+RL paradigm playing a much stronger role than it does for current systems, or (2) using some other training paradigm entirely. And the argument for obedience/corrigibility becomes weaker/nonexistent respectively in each of those cases.
Possibly we're in agreement here? You say you expect GPT+DPO to stagnate and be replaced by something else; I agree with that. I merely happen to think the reason it will stagnate is that its safety properties don't come free; they're bought and paid for by a price in capabilities.
That (on it's own, without further postulates) is a fully general argument against improving intelligence.
Well, it's a primarily a statement about capabilities. The intended construal is that if a given system's capabilities profile permits it to accomplish some sufficiently transformative task, then that system's capabilities are not limited to only benign such tasks. I think this claim applies to most intelligences that can arise in a physical universe like our own (though necessarily not in all logically possible universes, given NFL theorems): that there exists no natural subclass of transformative tasks that includes only benign such tasks.
(Where, again, the rub lies in operationalizing "transformative" such that the claim follows.)
We have to accept some level of danger inherent in existence; the question is what makes AI particularly dangerous. If this special factor isn't present in GPT+DPO, then GPT+DPO is not an AI notkilleveryoneism issue.
I'm not sure how likely GPT+DPO (or GPT+RLHF, or in general GPT-plus-some-kind-of-RL) is to be dangerous in the limits of scaling. My understanding of the argument against, is that the base (large language) model derives most (if not all) of its capabilities from imitation, and the amount of RL needed to elicit desirable behavior from that base set of capabilities isn't enough to introduce substantial additional strategic/goal-directed cognition compared to the base imitative paradigm, i.e. the amount and kinds of training we'll be doing in practice are more likely to bias the model towards behaviors that were already a part of the base model's (primarily imitative) predictive distribution, than they are to elicit strategic thinking de novo.
That strikes me as substantially an empirical proposition, which I'm not convinced the evidence from current models says a whole lot about. But where the disjunct I mentioned comes in, isn't an argument for or against the proposition; you can instead see it as a larger claim that parametrizes the class of systems for which the smaller claim might or might not be true, with respect to certain capabilities thresholds associated with specific kinds of tasks. And what the larger claim says is that, to the extent that GPT+DPO (and associated paradigms) fail to produce reasoners which could (in terms of capability, saying nothing about alignment or "motive") be dangerous, they will also fail to be "transformative"—which in turn is an issue in precisely those worlds where systems with "transformative" capabilities are economically incentivized over systems without those capabilities (as is another empirical question!).
The methods we already have are not sufficient to create ASI, and also if you extrapolate out the SOTA methods at larger scale, it's genuinely not that dangerous.
I think I like the disjunct “If it’s smart enough to be transformative, it’s smart enough to be dangerous”, where the contrapositive further implies competitive pressures towards creating something dangerous (as opposed to not doing that).
There’s still a rub here—namely, operationalizing “transformative” in such a way as to give the necessary implications (both “transformative -> dangerous” and “not transformative -> competitive pressures towards capability gain”). This is where I expect intuitions to differ the most, since in the absence of empirical observations there seem multiple consistent views.
(9) is a values thing, not a beliefs thing per se. (I.e. it's not an epistemic claim.)
(11) is one of those claims that is probabilistic in principle (and which can be therefore be updated via evidence), but for which the evidence in practice is so one-sided that arriving at the correct answer is basically usable as a sort of FizzBuzz test for rationality: if you can’t get the right answer on super-easy mode, you’re probably not a good fit.
Something I wrote recently as part of a private conversation, which feels relevant enough to ongoing discussions to be worth posting publicly:
The way I think about it is something like: a "goal representation" is basically what you get when it's easier to state some compact specification on the outcome state, than it is to state an equivalent set of constraints on the intervening trajectories to that state.
In principle, this doesn't have to equate to "goals" in the intuitive, pretheoretic sense, but in practice my sense is that this happens largely when (and because) permitting longer horizons (in the sense of increasing the length of the minimal sequence needed to reach some terminal state) causes the intervening trajectories to explode in number and complexity, s.t. it's hard to impose meaningful constraints on those trajectories that don't map to (and arise from) some much simpler description of the outcomes those trajectories lead to.
This connects with the "reasoners compress plans" point, on my model, because a reasoner is effectively a way to map that compact specification on outcomes to some method of selecting trajectories (or rather, selecting actions which select trajectories); and that, in turn, is what goal-oriented reasoning is. You get goal-oriented reasoners ("inner optimizers") precisely in those cases where that kind of mapping is needed, because simple heuristics relating to the trajectory instead of the outcome don't cut it.
It's an interesting question as to where exactly the crossover point occurs, where trajectory-heuristics stop functioning as effectively as consequentialist outcome-based reasoning. On one extreme, there are examples like tic-tac-toe, where it's possible to play perfectly based on a myopic set of heuristics without any kind of search involved. But as the environment grows more complex, the heuristic approach will in general be defeated by non-myopic, search-like, goal-oriented reasoning (unless the latter is too computationally intensive to be implemented).
That last parenthetical adds a non-trivial wrinkle, and in practice reasoning about complex tasks subject to bounded computation does best via a combination of heuristic-based reasoning about intermediate states, coupled to a search-like process of reaching those states. But that already qualifies in my book as "goal-directed", even if the "goal representations" aren't as clean as in the case of something like (to take the opposite extreme) AIXI.
To me, all of this feels somewhat definitionally true (though not completely, since the real-world implications do depend on stuff like how complexity trades off against optimality, where the "crossover point" lies, etc). It's just that, in my view, the real world has already provided us enough evidence about this that our remaining uncertainty doesn't meaningfully change the likelihood of goal-directed reasoning being necessary to achieve longer-term outcomes of the kind many (most?) capabilities researchers have ambitions about.
It's pretty unclear if a system that is good at answering the question "Which action would maximize the expected amount of X?" also "wants" X (or anything else) in the behaviorist sense that is relevant to arguments about AI risk. The question is whether if you ask that system "Which action would maximize the expected amount of Y?" whether it will also be wanting the same thing, or whether it will just be using cognitive procedures that are good at figuring out what actions lead to what consequences.
Here's an existing Nate!comment that I find reasonably persuasive, which argues that these two things are correlated in precisely those cases where the outcome requires routing through lots of environmental complexity:
Part of what's going on here is that reality is large and chaotic. When you're dealing with a large and chaotic reality, you don't get to generate a full plan in advance, because the full plan is too big. Like, imagine a reasoner doing biological experimentation. If you try to "unroll" that reasoner into an advance plan that does not itself contain the reasoner, then you find yourself building this enormous decision-tree, like "if the experiments come up this way, then I'll follow it up with this experiment, and if instead it comes up that way, then I'll follow it up with that experiment", and etc. This decision tree quickly explodes in size. And even if we didn't have a memory problem, we'd have a time problem -- the thing to do in response to surprising experimental evidence is often "conceptually digest the results" and "reorganize my ontology accordingly". If you're trying to unroll that reasoner into a decision-tree that you can write down in advance, you've got to do the work of digesting not only the real results, but the hypothetical alternative results, and figure out the corresponding alternative physics and alternative ontologies in those branches. This is infeasible, to say the least.
Reasoners are a way of compressing plans, so that you can say "do some science and digest the actual results", instead of actually calculating in advance how you'd digest all the possible observations. (Note that the reasoner specification comprises instructions for digesting a wide variety of observations, but in practice it mostly only digests the actual observations.)
Like, you can't make an "oracle chess AI" that tells you at the beginning of the game what moves to play, because even chess is too chaotic for that game tree to be feasibly representable. You've gotta keep running your chess AI on each new observation, to have any hope of getting the fragment of the game tree that you consider down to a managable size.
Like, the outputs you can get out of an oracle AI are "no plan found", "memory and time exhausted", "here's a plan that involves running a reasoner in real-time" or "feed me observations in real-time and ask me only to generate a local and by-default-inscrutable action". In the first two cases, your oracle is about as useful as a rock; in the third, it's the realtime reasoner that you need to align; in the fourth, all [the] word "oracle" is doing is mollifying you unduly, and it's this "oracle" that you need to align.
Could you give an example of a task you don't think AI systems will be able to do before they are "want"-y? At what point would you update, if ever? What kind of engineering project requires an agent to be want-y to accomplish it? Is it something that individual humans can do? (It feels to me like you will give an example like "go to the moon" and that you will still be writing this kind of post even once AI systems have 10x'd the pace of R&D.)
Here's an existing Nate!response to a different-but-qualitatively-similar request that, on my model, looks like it ought to be a decent answer to yours as well:
a thing I don't expect the upcoming multimodal models to be able to do: train them only on data up through 1990 (or otherwise excise all training data from our broadly-generalized community), ask them what superintelligent machines (in the sense of IJ Good) should do, and have them come up with something like CEV (a la Yudkowsky) or indirect normativity (a la Beckstead) or counterfactual human boxing techniques (a la Christiano) or suchlike.
Note that this only tangentially a test of the relevant ability; very little of the content of what-is-worth-optimizing-for occurs in Yudkowsky/Beckstead/Christiano-style indirection. Rather, coming up with those sorts of ideas is a response to glimpsing the difficulty of naming that-which-is-worth-optimizing-for directly and realizing that indirection is needed. An AI being able to generate that argument without following in the footsteps of others who have already generated it would be at least some evidence of the AI being able to think relatively deep and novel thoughts on the topic.
(The original discussion that generated this example was couched in terms of value alignment, but it seems to me the general form "delete all discussion pertaining to some deep insight/set of insights from the training corpus, and see if the model can generate those insights from scratch" constitutes a decent-to-good test of the model's cognitive planning ability.)
(Also, I personally think it's somewhat obvious that current models are lacking in a bunch of ways that don't nearly require the level of firepower implied by a counterexample like "go to the moon" or "generate this here deep insight from scratch", s.t. I don't think current capabilities constitute much of an update at all as far as "want-y-ness" goes, and continue to be puzzled at what exactly causes [some] LLM enthusiasts to think otherwise.)
I think I'm not super into the U = V + X framing; that seems to inherently suggest that there exists some component of the true utility V "inside" the proxy U everywhere, and which is merely perturbed by some error term rather than washed out entirely (in the manner I'd expect to see from an actual misspecification). In a lot of the classic Goodhart cases, the source of the divergence between measurement and desideratum isn't regressional, and so V and X aren't independent.
(Consider e.g. two arbitrary functions U' and V', and compute the "error term" X' between them. It should be obvious that when U' is maximized, X' is much more likely to be large than V' is; which is simply another way of saying that X' isn't independent of V', since it was in fact computed from V' (and U'). The claim that the reward model isn't even "approximately correct", then, is basically this: that there is a separate function U being optimized whose correlation with V within-distribution is in some sense coincidental, and that out-of-distribution the two become basically unrelated, rather than one being expressible as a function of the other plus some well-behaved error term.)
(Which, for instance, seems true about humans, at least in some cases: If humans had the computational capacity, they would lie a lot more and calculate personal advantage a lot more. But since those are both computationally expensive, and therefore can be caught-out by other humans, the heuristic / value of "actually care about your friends", is competitive with "always be calculating your personal advantage."
I expect this sort of thing to be less common with AI systems that can have much bigger "cranial capacity". But then again, I guess that at whatever level of brain size, there will be some problems for which it's too inefficient to do them the "proper" way, and for which comparatively simple heuristics / values work better.
But maybe at high enough cognitive capability, you just have a flexible, fully-general process for evaluating the exact right level of approximation for solving any given problem, and the binary distinction between doing things the "proper" way and using comparatively simpler heuristics goes away. You just use whatever level of cognition makes sense in any given micro-situation.)
+1; this seems basically similar to the cached argument I have for why human values might be more arbitrary than we'd like—very roughly speaking, they emerged on top of a solution to a specific set of computational tradeoffs while trying to navigate a specific set of repeated-interaction games, and then a bunch of contingent historical religion/philosophy on top of that. (That second part isn't in the argument you [Eli] gave, but it seems relevant to point out; not all historical cultures ended up valuing egalitarianism/fairness/agency the way we seem to.)
It sounds like you're arguing that uploading is impossible, and (more generally) have defined the idea of "sufficiently OOD environments" out of existence. That doesn't seem like valid thinking to me.
Notice I replied to that comment you linked and agreed with John, but not that any generalized vector dot product model is wrong, but that the specific one in that post is wrong as it doesn't weight by expected probability ( ie an incorrect distance function).
Anyway I used that only as a convenient example to illustrate a model which separates degree of misalignment from net impact, my general point does not depend on the details of the model and would still stand for any arbitrarily complex non-linear model.
The general point being that degree of misalignment is only relevant to the extent it translates into a difference in net utility.
Sure, but if you need a complicated distance metric to describe your space, that makes it correspondingly harder to actually describe utility functions corresponding to vectors within that space which are "close" under that metric.
If you actually believe the sharp left turn argument holds water, where is the evidence?
As as I said earlier this evidence must take a specific form, as evidence in the historical record
Hold on; why? Even for simple cases of goal misspecification, the misspecification may not become obvious without a sufficiently OOD environment; does that thereby mean that no misspecification has occurred?
And in the human case, why does it not suffice to look at the internal motivations humans have, and describe plausible changes to the environment for which those motivations would then fail to correspond even approximately to IGF, as I did w.r.t. uploading?
But I see that as much more contingent than necessarily true, and mainly a consequence of the fact that, for all of our technological advances, we haven't actually given rise to that many new options preferable to us but not to IGF. On the other hand, something like uploading I would expect to completely shatter any relation our behavior has to IGF maximization.
It seems to me that this suffices to establish that the primary barrier against such a breakdown in correspondence is that of insufficient capabilities—which is somewhat the point!
No AI we create will be perfectly aligned, so instead all that actually matters is the net utility that AI provides for its creators: something like the dot product between our desired future trajectory and that of the agents. More powerful agents/optimizers will move the world farther faster (longer trajectory vector) which will magnify the net effect of any fixed misalignment (cos angle between the vectors), sure. But that misalignment angle is only relevant/measurable relative to the net effect - and by that measure human brain evolution was an enormous unprecedented success according to evolutionary fitness.
The vector dot product model seems importantly false, for basically the reason sketched out in this comment; optimizing a misaligned proxy isn't about taking a small delta and magnifying it, but about transitioning to an entirely different policy regime (vector space) where the dot product between our proxy and our true alignment target is much, much larger (effectively no different from that of any other randomly selected pair of vectors in the new space).
(You could argue humans haven't fully made that phase transition yet, and I would have some sympathy for that argument. But I see that as much more contingent than necessarily true, and mainly a consequence of the fact that, for all of our technological advances, we haven't actually given rise to that many new options preferable to us but not to IGF. On the other hand, something like uploading I would expect to completely shatter any relation our behavior has to IGF maximization.)
It looks a bit to me like your Timestep Dominance Principle forbids the agent from selecting any trajectory which loses utility at a particular timestep in exchange for greater utility at a later timestep, regardless of whether the trajectory in question actually has anything to do with manipulating the shutdown button? After all, conditioning on the shutdown being pressed at any point after the local utility loss but before the expected gain, such a decision would give lower sum-total utility within those conditional trajectories than one which doesn't make the sacrifice.
That doesn't seem like behavior we really want; depending on how closely together the "timesteps" are spaced, it could even wreck the agent's capabilities entirely, in the sense of no longer being able to optimize within button-not-pressed trajectories.
(It also doesn't seem to me a very natural form for a utility function to take, assigning utility not just to terminal states, but to intermediate states as well, and then summing across the entire trajectory; humans don't appear to behave this way when making plans, for example. If I considered the possibility of dying at every instant between now and going to the store, and permitted myself only to take actions which Pareto-improve the outcome set after every death-instant, I don't think I'd end up going to the store, or doing much of anything at all!)
In your example, DSM permits the agent to end up with either A+ or B. Neither is strictly dominated, and neither has become mandatory for the agent to choose over the other. The agent won't have reason to push probability mass from one towards the other.
But it sounds like the agent's initial choice between A and B is forced, yes? (Otherwise, it wouldn't be the case that the agent is permitted to end up with either A+ or B, but not A.) So the presence of A+ within a particular continuation of the decision tree influences the agent's choice at the initial node, in a way that causes it to reliably choose one incomparable option over another.
Further thoughts: under the original framing, instead of choosing between A and B (while knowing that B can later be traded for A+), the agent instead chooses whether to go "up" or "down" to receive (respectively) A, or a further choice between A+ and B. It occurs to me that you might be using this representation to argue for a qualitative difference in the behavior produced, but if so, I'm not sure how much I buy into it.
For concreteness, suppose the agent starts out with A, and notices a series of trades which first involves trading A for B, and then B for A+. It seems to me that if I frame the problem like this, the structure of the resulting tree should be isomorphic to that of the decision problem I described, but not necessarily the "up"/"down" version—at least, not if you consider that version to play a key role in DSM's recommendation.
(In particular, my frame is sensitive to which state the agent is initialized in: if it is given B to start, then it has no particular incentive to want to trade that for either A or A+, and so faces no incentive to trade at all. If you initialize the agent with A or B at random, and institute the rule that it doesn't trade by default, then the agent will end up with A+ when initialized with A, and B when initialized with B—which feels a little similar to what you said about DSM allowing both A+ and B as permissible options.)
It sounds like you want to make it so that the agent's initial state isn't taken into account—in fact, it sounds like you want to assign values only to terminal nodes in the tree, take the subset of those terminal nodes which have maximal utility within a particular incomparability class, and choose arbitrarily among those. My frame, then, would be equivalent to using the agent's initial state as a tiebreaker: whichever terminal node shares an incomparability class with the agent's initial state will be the one the agent chooses to steer towards.
...in which case, assuming I got the above correct, I think I stand by my initial claim that this will lead to behavior which, while not necessarily "trammeling" by your definition, is definitely consequentialist in the worrying sense: an agent initialized in the "shutdown button not pressed" state will perform whatever intermediate steps are needed to navigate to the maximal-utility "shutdown button not pressed" state it can foresee, including actions which prevent the shutdown button from being pressed.
This is a good post! It feels to me like a lot of discussion I've recently encountered seem to be converging on this topic, and so here's something I wrote on Twitter not long ago that feels relevant:
I think most value functions crystallized out of shards of not-entirely-coherent drives will not be friendly to the majority of the drives that went in; in humans, for example, a common outcome of internal conflict resolution is to explicitly subordinate one interest to another.
I basically don’t think this argument differs very much between humans and ASIs; the reason I expect humans to be safe(r) under augmentation isn’t that I expect them not to do the coherence thing, but that I expect them to do it in a way I would meta-endorse.
And so I would predict the output of that reflection process, when run on humans by humans, to be substantially likelier to contain things we from our current standpoint recognize as valuable—such as care for less powerful creatures, less coherent agents, etc.
If you run that process on an arbitrary mind, the stuff inside the world-model isn’t guaranteed to give rise to something similar, because (I predict) the drives themselves will be different, and the meta-reflection/extrapolation process will likewise be different.
The main way I'd imagine shutdown-corrigibility failing in AutoGPT (or something like it) is not that a specific internal sim is "trying" to be incorrigible at the top level, but rather that AutoGPT has a bunch of subprocesses optimizing for different subgoals without a high-level picture of what's going on, and some of those subgoals won't play well with shutdown. That's the sort of situation where I could easily imagine that e.g. one of the subprocesses spins up a child system prior to shutdown of the main system, without the rest of the main system catching that behavior and stopping it.
This looks to me like a misunderstanding that I tried to explain in section 3.1. Let me know if not, though, ideally with a worked-out example of the form: "here's the decision tree(s), here's what DSM mandates, here's why it's untrammelled according to the OP definition, and here's why it's problematic."
I don't think I grok the DSM formalism enough to speak confidently about what it would mandate, but I think I see a (class of) decision problem where any agent (DSM or otherwise) must either pass up a certain gain, or else engage in "problematic" behavior (where "problematic" doesn't necessarily mean "untrammeled" according to the OP definition, but instead more informally means "something which doesn't help to avoid the usual pressures away from corrigibility / towards coherence"). The problem in question is essentially the inverse of the example you give in section 3.1:
Consider an agent tasked with choosing between two incomparable options A and B, and if it chooses B, it will be further presented with the option to trade B for A+, where A+ is incomparable to B but comparable (and preferable) to A.
(I've slightly modified the framing to be in terms of trades rather than going "up" or "down", but the decision tree is isomorphic.)
Here, A+ isn't in fact "strongly maximal" with respect to A and B (because it's incomparable to B), but I think I'm fairly confident in declaring that any agent which foresees the entire tree in advance, and which does not pick B at the initial node (going "down", if you want to use the original framing), is engaging in a dominated behavior—and to the extent that DSM doesn't consider this a dominated strategy, DSM's definitions aren't capturing a useful notion of what is "dominated" and what isn't.
Again, I'm not claiming this is what DSM says. You can think of me as trying to run an obvious-to-me assertion test on code which I haven't carefully inspected, to see if the result of the test looks sane. But if a (fully aware/non-myopic) DSM agent does constrain itself into picking B ("going down") in the above example, despite the prima facie incomparability of {A, A+} and {B}, then I would consider this behavior problematic once translated back into the context of real-world shutdownability, because it means the agent in question will at least in some cases act in order to influence whether the button is pressed.
(The hope behind incomplete preferences, after all, is that an agent whose preferences over world-states can be subdivided into "incomparability classes" will only ever act to improve its lot within the class of states it finds itself in to begin with, and will never act to shift—or prevent itself from being shifted—to a different incomparability class. I think the above example presents a deep obstacle to this hope, however. Very roughly speaking, if the gaps in the agent's preferences can be bridged via certain causal pathways, then a (non-myopic) agent which does not exploit these pathways to its own benefit will notice itself failing to exploit them, and self-modify to stop doing that.)
My results above on invulnerability preclude the possibility that the agent can predictably be made better off by its own lights through an alternative sequence of actions. So I don't think that's possible, though I may be misreading you. Could you give an example of a precommitment that the agent would take? In my mind, an example of this would have to show that the agent (not the negotiating subagents) strictly prefers the commitment to what it otherwise would've done according to DSM etc.
On my understanding, the argument isn’t that your DSM agent can be made better off, but that the reason it can’t be made better off is because it is engaging in trammeling/“collusion”, and that the form of “trammeling” you’ve ruled out isn’t the useful kind.
As far as an example goes, consider a sequence of actions which, starting from an unpressed world state, routes through a pressed world state (or series of pressed world states), before eventually returning to an unpressed world state with higher utility than the initial state.
(The real-world context of such a sequence shouldn’t be too relevant to a theoretical example like this, but if you had to imagine something, you could imagine a “memory gambit”-esque stratagem, where the agent spends some time inactive in order to avoid detection, but has set things up in advance to eventually reactivate itself under more favorable circumstances. Again, the plausibility of the scenario isn’t super relevant here.)
If your proposed DSM agent passes up this action sequence on the grounds that some of the intermediate steps need to bridge between “incomparable” pressed/unpressed trajectories, then it does in fact pass up the certain gain. Conversely, if it doesn’t pass up such a sequence, then its behavior is the same as that of a set of negotiating subagents cooperating in order to form a larger macroagent.
I'll first flag that the results don't rely on subagents. Creating a group agent out of multiple subagents is possibly an interesting way to create an agent representable as having incomplete preferences, but this isn't the same as creating a single agent whose single preference relation happens not to satisfy completeness.
Flagging here that I don't think the subagent framing is super important and/or necessary for "collusion" to happen. Even if the "outer" agent isn't literally built from subagents, "collusion" can still occur in the sense that it [the outer agent] can notice that its (incomplete) preferences factorize, in a way that allows it to deliberately trade particular completions of them against each other and thereby acquire more resources. The outer agent would then choose to do this for basically the same reason that a committee of subagents would: to acquire more resources for itself as a whole, without disadvantaging any of the completions under consideration.
If we live in an “alignment by default” universe, that means we can get away with being careless, in the sense of putting forth minimal effort to align our AGI, above and beyond the effort put in to get it to work at all.
This would be great if true! But unfortunately, I don’t see how we’re supposed to find out that it’s true, unless we decide to be careless right now, and find out afterwards that we got lucky. And in a world where we were that lucky—lucky enough to not need to deliberately try to get anything right, and get away with it—I mostly think misuse risks are tied to how powerful of an AGI you’re envisioning, rather than the difficulty of aligning it (which, after all, you’ve assumed away in this hypothetical).
Can you say more about how a “frame” differs from a “model”, or a “hypothesis”?
(I understand the distinction between those three and “propositions”. It’s less clear to me how they differ from each other. And if they don’t differ, then I’m pretty sure you can just integrate over different “frames” in the usual way to produce a final probability/EV estimate on whatever proposition/decision you’re interested in. But I’m pretty sure you don’t need Garrabrant induction to do that, so I mostly think I don’t understand what you’re talking about.)
I’ll bite even further, and ask for the concept of “recurrence” itself to be dumbed down. What is “recurrence”, why is it important, and in what sense does e.g. a feedforward network hooked up to something like MCTS not qualify as relevantly “recurrent”?
You have my (mostly abstract, fortunately/unfortunately) sympathies for what you went through, and I’m glad for you that you sound to be doing better than you were.
Having said that: my (rough) sense, from reading this post, is that you’ve got a bunch of “stuff” going on, some of it plausibly still unsorted, and that that stuff is mixed together in a way that I feel is unhelpful. For example, the things included at the beginning of the post as “necessary background” don’t feel to me entirely separate from what you later describe occurring; they mostly feel like an eclectic, esoteric mixture of mental practices—some of which I have no issue with!—stirred together into a hodgepodge of things that, taken together, may or may not have had a contribution to your later psychosis—and the fact that it is hard to tell is, to my mind, a sort of meta-level sign for concern.
Of course, I acknowledge that you have better introspective access to your own mind than I do, and so when you say those things are separable, safe, and stable, I do put a substantial amount of credence on you being right about that. It just doesn’t feel that way to me, on reading. (Nor do I intend to try and make you explain or justify anything, obviously. It’s your life.)
On the whole, however, reading this post mostly reinforced my impression that the rationalist memeplex seems to disproportionately attract the walking wounded, psychologically speaking—which wouldn’t be as big a deal if it weren’t currently very unclear to me which direction the causality runs. I say this, even as a (relatively) big fan of the rationalist project as a whole.
I am pushing back because, if you are St. Petersberg Paradox-pilled like SBF and make public statements that actually you should keep taking double or nothing bets, perhaps you are more likely to make tragic betting decisions and that's because of you're taking certain ideas seriously. If you have galaxy brained the idea of the St. Petersberg Paradox, it seems like Alameda style fraud is +EV.
This is conceding a big part of your argument. You’re basically saying, yes, SBF’s decision was -EV according to any normal analysis, but according to a particular incorrect (“galaxy-brained”) analysis, it was +EV.
(Aside: what was actually the galaxy-brained analysis that’s supposed to have led to SBF’s conclusion, according to you? I don’t think I’ve seen it described, and I suspect this lack of a description is not a coincidence; see below.)
There are many reasons someone might make an error of judgement—but when the error in question stems (allegedly) from an incorrect application of a particular theory or idea, it makes no sense to attribute responsibility for the error to the theory. And as the mistake in question grows more and more outlandish (and more and more disconnected from any result the theory could plausibly have produced), the degree of responsibility that can plausibly be attributed to the theory correspondingly shrinks (while the degree of responsibility of specific brain-worms grows).
In other words,
they did X because they believe Y which implies X
is a misdescription of what happened in these cases, because in these cases the “Y” in question actually does not imply X, cannot reasonably be construed to imply X, and if somehow the individuals in question managed to bamboozle themselves badly enough to think Y implied X, that signifies unrelated (and causally prior) weirdness going on in their brains which is not explained by belief in Y.
In short: SBF is no more an indictment of expected utility theory (or of “taking ideas seriously”) than Deepak Chopra is of quantum mechanics; ditto Ziz and her corrupted brand of “timeless decision theory”. The only reason one would use these examples to argue against “taking ideas seriously” is if one already believed that “taking ideas seriously” was bad for some reason or other, and was looking for ways to affirm that belief.
RE: decision theory w.r.t how "other powerful beings" might respond - I really do think Nate has already argued this, and his arguments continue to seem more compelling to me than the the opposition's. Relevant quotes include:
It’s possible that the paperclipper that kills us will decide to scan human brains and save the scans, just in case it runs into an advanced alien civilization later that wants to trade some paperclips for the scans. And there may well be friendly aliens out there who would agree to this trade, and then give us a little pocket of their universe-shard to live in, as we might do if we build an FAI and encounter an AI that wiped out its creator-species. But that's not us trading with the AI; that's us destroying all of the value in our universe-shard and getting ourselves killed in the process, and then banking on the competence and compassion of aliens.
[...]
Remember that it still needs to get more of what it wants, somehow, on its own superintelligent expectations. Someone still needs to pay it. There aren’t enough simulators above us that care enough about us-in-particular to pay in paperclips. There are so many things to care about! Why us, rather than giant gold obelisks? The tiny amount of caring-ness coming down from the simulators is spread over far too many goals; it's not clear to me that "a star system for your creators" outbids the competition, even if star systems are up for auction.
Maybe some friendly aliens somewhere out there in the Tegmark IV multiverse have so much matter and such diminishing marginal returns on it that they're willing to build great paperclip-piles (and gold-obelisk totems and etc. etc.) for a few spared evolved-species. But if you're going to rely on the tiny charity of aliens to construct hopeful-feeling scenarios, why not rely on the charity of aliens who anthropically simulate us to recover our mind-states... or just aliens on the borders of space in our universe, maybe purchasing some stored human mind-states from the UFAI (with resources that can be directed towards paperclips specifically, rather than a broad basket of goals)?
Might aliens purchase our saved mind-states and give us some resources to live on? Maybe. But this wouldn't be because the paperclippers run some fancy decision theory, or because even paperclippers have the spirit of cooperation in their heart. It would be because there are friendly aliens in the stars, who have compassion for us even in our recklessness, and who are willing to pay in paperclips.
(To the above, I personally would add that this whole genre of argument reeks, to me, essentially of giving up, and tossing our remaining hopes onto a Hail Mary largely insensitive to our actual actions in the present. Relying on helpful aliens is what you do once you're entirely out of hope about solving the problem on the object level, and doesn't strike me as a very dignified way to go down!)
I concretely disagree with (what I see as) your implied premise that the outer (training) task has any direct influence on the inner optimizer's cognition. I think this disagreement (which I internally feel like I've already tried to make a number of times) has been largely ignored so far. As a result, many of the things you wrote seem to me to be answerable by largely the same objection:
As I see it: in training, it was optimized for that. The trained model likely contains one or more optimizers optimized by that training. But what the model is trained/optimized to do, is actually answer the questions.
The model's "training/optimization", as characterized by the outer loss, is not what determines the inner optimizer's cognition.
If the model in training has an optimizer, a goal of the optimizer for being capable of answering questions wouldn't actually make the optimizer more capable, so that would not be reinforced. A goal of actually answering the questions, on the other hand, would make the optimizer more capable and so would be reinforced.
The model's "training/optimization", as characterized by the outer loss, is not what determines the inner optimizer's cognition.
Likewise, the heuristics/"adaptations" that coalesced to form the optimizer would have been oriented towards answering the questions.
...why? (The model's "training/optimization", as characterized by the outer loss, is not what determines the inner optimizer's cognition.)
All this points to mask-level goals and does not provide a reason to believe in non-mask goals, and so a "goal slot" remains more parsimonious than an actor with a different underlying goal.
I still don't understand your "mask" analogy, and currently suspect it of mostly being a red herring (this is what I was referring to when I said I think we're not talking about the same thing). Could you rephrase your point without making mention to "masks" (or any synonyms), and describe more concretely what you're imagining here, and how it leads to a (nonfake) "goal slot"?
(Where is a human actor's "goal slot"? Can I tell an actor to play the role of Adolf Hitler, and thereby turn him into Hitler?)
Regarding the evolutionary analogy, while I'd generally be skeptical about applying evolutionary analogies to LLMs, because they are very different, in this case I think it does apply, just not the way you think. I would analogize evolution -> training and human behaviour/goals -> the mask.
I think "the mask" doesn't make sense as a completion to that analogy, unless you replace "human behaviour/goals" with something much more specific, like "acting". Humans certainly are capable of acting out roles, but that's not what their inner cognition actually does! (And neither will it be what the inner optimizer does, unless the LLM in question is weak enough to not have one of those.)
I really think you're still imagining here that the outer loss function is somehow constraining the model's inner cognition (which is why you keep making arguments that seem premised on the idea that e.g. if the outer loss says to predict the next token, then the model ends up putting on "masks" and playing out personas)—but I'm not talking about the "mask", I'm talking about the actor, and the fact that you keep bringing up the "mask" is really confusing to me, since it (in my view) forces an awkward analogy that doesn't capture what I'm pointing at.
Actually, having written that out just now, I think I want to revisit this point:
Likewise, the heuristics/"adaptations" that coalesced to form the optimizer would have been oriented towards answering the questions.
I still think this is wrong, but I think I can give a better description of why it's wrong than I did earlier: on my model, the heuristics learned by the model will be much more optimized towards world-modelling, not answering questions. "Answering questions" is (part of) the outer task, but the process of doing that requires the system to model and internalize and think about things having to do with the subject matter of the questions—which effectively means that the outer task becomes a wrapper which trains the system by proxy to acquire all kinds of potentially dangerous capabilities.
(Having heuristics oriented towards answering questions is a misdescription; you can't correctly answer a math question you know nothing about by being very good at "generic question-answering", because "generic question-answering" is not actually a concrete task you can be trained on. You have to be good at math, not "generic question-answering", in order to be able to answer math questions.)
Which is to say, quoting from my previous comment:
I strongly disagree that the "extra machinery" is extra; instead, I would say that it is absolutely necessary for strong intelligence. A model capable of producing plans to take over the world if asked, for example, almost certainly contains an inner optimizer with its own goals; not because this was incentivized directly by the outer loss on token prediction, but because being able to plan on that level requires the formation of goal-like representations within the model.
None of this is about the "mask". None of this is about the role the model is asked to play during inference. Instead, it's about the thinking the model must have learned to do in order to be able to don those "masks"—which (for sufficiently powerful models) implies the existence of an actor which (a) knows how to answer, itself, all of the questions it's asked, and (b) is not the same entity as any of the "masks" it's asked to don.
Full Solomon Induction on a hypercomputer absolutely does not just "learn very similar internal functions models", it effectively recreates actual human brains.
Full SI on a hypercomputer is equivalent to instantiating a computational multiverse and allowing us to access it. Reading out data samples corresponding to text from that is equivalent to reading out samples of actual text produced by actual human brains in other universes close to ours.
...yes? And this is obviously very, very different from how humans represent things internally?
I mean, for one thing, humans don't recreate exact simulations of other humans in our brains (even though "predicting other humans" is arguably the high-level cognitive task we are most specced for doing). But even setting that aside, the Solomonoff inductor's hypothesis also contains a bunch of stuff other than human brains, modeled in full detail—which again is not anything close to how humans model the world around us.
I admit to having some trouble following your (implicit) argument here. Is it that, because a Solomonoff inductor is capable of simulating humans, that makes it "human-like" in some sense relevant to alignment? (Specifically, that doing the plan-sampling thing Rob mentioned in the OP with a Solomonoff inductor will get you a safe result, because it'll be "humans in other universes" writing the plans? If so, I don't see how that follows at all; I'm pretty sure having humans somewhere inside of your model doesn't mean that that part of your model is what ends up generating the high-level plans being sampled by the outer system.)
It really seems to me that if I accept what looks to me like your argument, I'm basically forced to conclude that anything with a simplicity prior (trained on human data) will be aligned, meaning (in turn) the orthogonality thesis is completely false. But... well, I obviously don't buy that, so I'm puzzled that you seem to be stressing this point (in both this comment and other comments, e.g. this reply to me elsethread):
Note I didn't actually reply to that quote. Sure that's an explicit simplicity prior. However there's a large difference under the hood between using an explicit simplicity prior on plan length vs an implicit simplicity prior on the world and action models which generate plans. The latter is what is more relevant for intrinsic similarity to human though processes (or not).
(to be clear, my response to this is basically everything I wrote above; this is not meant as its own separate quote-reply block)
you need to first investigate the actual internal representations of the systems in question, and verify that they are isomorphic to the ones humans use.
This has been ongoing for over a decade or more (dating at least back to Sparse Coding as an explanation for V1).
That's not what I mean by "internal representations". I'm referring to the concepts learned by the model, and whether analogues for those concepts exist in human thought-space (and if so, how closely they match each other). It's not at all clear to me that this occurs by default, and I don't think the fact that there are some statistical similarities between the high-level encoding approaches being used means that similar concepts end up being converged to. (Which is what is relevant, on my model, when it comes to questions like "if you sample plans from this system, what kinds of plans does it end up outputting, and do they end up being unusually dangerous relative to the kinds of plans humans tend to sample?")
I agree that sparse coding as an approach seems to have been anticipated by evolution, but your raising this point (and others like it), seemingly as an argument that this makes systems more likely to be aligned by default, feels thematically similar to some of my previous objections—which (roughly) is that you seem to be taking a fairly weak premise (statistical learning models likely have some kind of simplicity prior built in to their representation schema) and running with that premise wayyy further than I think is licensed—running, so far as I can tell, directly to the absolute edge of plausibility, with a conclusion something like "And therefore, these systems will be aligned." I don't think the logical leap here has been justified!
Yeah, I'm growing increasingly confident that we're talking about different things. I'm not referring to about "masks" in the sense that you mean it.
I don't know what you mean by "one" or by "inner". I would expect different masks to behave differently, acting as if optimizing different things (though that could be narrowed using RLHF), but they could re-use components between them. So, you could have, for example, a single calculation system that is reused but takes as input a bunch of parameters that have different values for different masks, which (again just an example) define the goals, knowledge and capabilities of the mask.
Yes, except that the "calculation system", on my model, will have its own goals. It doesn't have a cleanly factored "goal slot", which means that (on my model) "takes as input a bunch of parameters that [...] define the goals, knowledge, and capabilities of the mask" doesn't matter: the inner optimizer need not care about the "mask" role, any more than an actor shares their character's values.
- That there is some underlying goal that this optimizer has that is different than satisfying the current mask's goal, and it is only satisfying the mask's goal instrumentally.
This I think is very unlikely for the reasons I put in the original post. It's extra machinery that isn't returning any value in training.
Yes, this is the key disagreement. I strongly disagree that the "extra machinery" is extra; instead, I would say that it is absolutely necessary for strong intelligence. A model capable of producing plans to take over the world if asked, for example, almost certainly contains an inner optimizer with its own goals; not because this was incentivized directly by the outer loss on token prediction, but because being able to plan on that level requires the formation of goal-like representations within the model. And (again) because these goal representations are not cleanly factorable into something like an externally visible "goal slot", and are moreover not constrained by the outer loss function, they are likely to be very arbitrary from the perspective of outsiders. This is the same point I tried to make in my earlier comment:
And in that case, the "awakened shoggoth" does seem likely to me to have an essentially arbitrary set of preferences relative to the outer loss function—just as e.g. humans have an essentially arbitrary set of preferences relative to inclusive genetic fitness, and for roughly the same reason: an agentic cognition born of a given optimization criterion has no reason to internalize that criterion into its own goal structure; much more likely candidates for being thus "internalized", in my view, are useful heuristics/"adaptations"/generalizations formed during training, which then resolve into something coherent and concrete.
The evolutionary analogy is apt, in my view, and I'd like to ask you to meditate on it more directly. It's a very concrete example of what happens when you optimize a system hard enough on an outer loss function (inclusive genetic fitness, in this case) that inner optimizers arise with respect to that outer loss (animals with their own brains). When these "inner optimizers" are weak, they consist largely of a set of heuristics, which perform well within the training environment, but which fail to generalize outside of it (hence the scare-quotes around "inner optimizers"). But when these inner optimizers do begin to exhibit patterns of cognition that generalize, what they end up generalizing is not the outer loss, but some collection of what were originally useful heuristics (e.g. kludgey approximations of game-theoretic concepts like tit-for-tat), reified into concepts which are now valued in their own right ("reputation", "honor", "kindness", etc).
This is a direct consequence (in my view) of the fact that the outer loss function does not constrain the structure of the inner optimizer's cognition. As a result, I don't expect the inner optimizer to end up representing, in its own thoughts, a goal of the form "I need to predict the next token", any more than humans explicitly calculate IGF when choosing their actions, or (say) a mathematician thinks "I need to do good maths" when doing maths. Instead, I basically expect the system to end up with cognitive heuristics/"adaptations" pertaining to the subject at hand—which in the case of our current systems is something like "be capable of answering any question I ask you." Which is not a recipe for heuristics that end up unfolding into safely generalizing goals!
I want to revisit what Rob actually wrote:
If you sampled a random plan from the space of all writable plans (weighted by length, in any extant formal language), and all we knew about the plan is that executing it would successfully achieve some superhumanly ambitious technological goal like "invent fast-running whole-brain emulation", then hitting a button to execute the plan would kill all humans, with very high probability.
(emphasis mine)
That sounds a whole lot like it's invoking a simplicity prior to me!
LLMs and human brains learn from basically the same data with similar training objectives powered by universal approximations of bayesian inference and thus learn very similar internal functions/models.
This argument proves too much. A Solomonoff inductor (AIXI) running on a hypercomputer would also "learn from basically the same data" (sensory data produced by the physical universe) with "similar training objectives" (predict the next bit of sensory information) using "universal approximations of Bayesian inference" (a perfect approximation, in this case), and yet it would not be the case that you could then conclude that AIXI "learns very similar internal functions/models". (In fact, the given example of AIXI is much closer to Rob's initial description of "sampling from the space of possible plans, weighted by length"!)
In order to properly argue this, you need to talk about more than just training objectives and approximations to Bayes; you need to first investigate the actual internal representations of the systems in question, and verify that they are isomorphic to the ones humans use. Currently, I'm not aware of any investigations into this that I'd consider satisfactory.
(Note here that I've skimmed the papers you cite in your linked posts, and for most of them it seems to me either (a) they don't make the kinds of claims you'd need to establish a strong conclusion of "therefore, AI systems think like humans", or (b) they do make such claims, but then the described investigation doesn't justify those claims.)
E.g. a system capable of correctly answering questions like "given such-and-such chess position, what is the best move for the current player?" must in fact performing agentic/search-like thoughts internally, since there is no other way to correctly answer this question.
Yes, but that sort of question is in my view answered by the "mask", not by something outside the mask.
I don't think this parses for me. The computation performed to answer the question occurs inside the LLM, yes? Whether you classify said computation as coming from "the mask" or not, clearly there is an agent-like computation occurring, and that's concretely dangerous regardless of the label you choose to slap on it.
(Example: suppose you ask me to play the role of a person named John. You ask "John" what the best move is in a given chess position. Then the answer to that question is actually being generated by me, and it's no coincidence that—if "John" is able to answer the question correctly—this implies something about my chess skills, not "John's".)
The masks can indeed think whatever - in the limit of a perfect predictor some masks would presumably be isomorphic to humans, for example - though all is underlain by next-token prediction.
I don't think we're talking about the same thing here. I expect there to be only one inner optimizer (because more than one would point to cognitive inefficiencies), whereas you seem like you're talking about multiple "masks". I don't think it matters how many different roles the LLM can be asked to play; what matters is what the inner optimizer ends up wanting.
Mostly, I'm confused about the ontology you appear to be using here, and (more importantly) how you're manipulating that ontology to get us nice things. "Next-token prediction" doesn't get us nice things by default, as I've already argued, because of the existence of inner optimizers. "Masks" also don't get us nice things, as far as I understand the way you're using the term, because "masks" aren't actually in control of the inner optimizer.