Discord: LemonUniverse (.lemonuniverse). Reddit: u/Smack-works. About my situation: here.
I wrote some bad posts before 2024 because I was very uncertain how the events may develop.
I do philosophical/conceptual research, have no mathematical or programming skills. But I do know a bunch of mathematical and computer science concepts.
How strong is that evidence, for you personally? Is it constructive (gives you vague ideas of how problems 1-3 could be solved without ignoring computational cost) or not (just gives you abstract hope)? Would be very interested to read your thoughts. Your post doesn’t emphasize your opinions.
My personal impression about mathematical theories for alignment is very pessimistic. Someone proposes a mathematical framework too general to make progress on. The framework struggles to prove even the most general theorems without relying on questionable assumptions. We don’t know if the proved theorems even mean what we think they mean (because the questionable assumptions narrow down the scope of the theorems in a strange way, making them very easy to misinterpret). Finally, the math has nothing to say about something super important (such as computational cost). Disclaimer: this impression is very indirect (I’m bad at math) and it might be unreasonable to ask for anything better.
How unique/unexpected is Sam’s theory, compared to other mathematical theories for alignment (from your subjective perspective)?
If I understand correctly, the standard worries about interpretability are
different types of AIs don’t have universal data structures
you can’t prove anything useful about those structures without AGI (the problem is too complicated/advanced for humanity)
if you induce interpretability in your AI, it won’t be competitive.
Why does condensation give hope for solving those problems? Where could the magic happen? For example, John Wentworth attacks 1-3 by making assumptions about our universe.
I don’t buy 4.1. I believe the laws of the universe should be simple, but simplicity of specifying the initial state is not an argument for me (unless there are additional factors). Don’t buy 4.2 either, because I don’t trust anthropic reasoning that much. However, I’m still interested in ideas like this, even if I don’t believe in them.
4.3. I don’t know enough math, but I don’t believe that a) we have the math to prove that an agent can’t survive in an environment with incompressible low-level dynamics or b) it means anything non-trivial enough to prove the kind of points you need to prove. It would be a magically good result.
In a chaotic environment there could be some high-level patterns which the agent might exploit, but those high-level patterns could be barely similar to humanlike abstractions. The Gooder Regulator theorem could be false, maybe, if we don’t need to be prepared for “any possible game”.
In other words: we have a probability distribution over sequences of transformations of the data we should attempt to try, and we keep sampling from it until we spot a familiar abstraction; until we find a way of looking at the data under which it resolves into a known, simple pattern.
Maybe you’re focused on the wrong thing. Maybe the explicit search for the right transformation happens only when things go wrong (and even then a lot of the search is not explicit).
You could have a multi-step detection method which deduces the right transformation instead of searching for it explicitly. Contrast two algorithms:
I slide and scale the dog template across an image until I find a good fit.
I detect the most salient parts of the image, combine those parts into lines, connect the lines, check if the lines have the right proportions. If they do, it’s a dog. If they don’t, it’s unlikely there are any dogs in the image.
But if it’s not an abstraction over the cells, then over… what? There aren’t really any other elements in the picture! Maybe it’s an abstraction over sets of cells, with cells being “subvariables” over which the glider is synergistic? But how exactly do we define those sets?
I think you should filter out sets which are naturally “salient” and then do something with them. For example, imagine I have a single glider in an empty world. I model the world for 1000 timesteps. I pick the set of cells which changed values during this time (that would be an extremely natural thing to do). This way I find at least the “trail” of the glider. By doing other natural things I can narrow it down farther or make it work in a world with many objects.
...
All of the above shouldn’t be groundbreaking, I’m just surprised you didn’t mention it at all and focused your examples on explicit search. Maybe you’re unsure if the trick with naturally salient subsets generalizes?
Writing a relatively high-effort comment because I’m thinking about similar things and intend to turn what I wrote about here into a top-level post.
First I’ll write some criticism, then mention some “leads” you might be interested in. For context, I started writing this comment before this exchange.
The High-Level Outline confused me. Feels like the argument you’re making is a bit tricky, yet you don’t spell it out explicitly enough. I’m not sure what assumptions are made.
Here’s my train of thought:
The universe has some structure (S) which makes human abstractions useful. But human abstractions != S, so human abstractions can still be hard to specify. You seem to be 100% aware of that.
...But then you connect human abstractions to the shortest description of the universe. Which doesn’t follow. …Or maybe you merely say that any short description of the universe is isomorphic to S, therefore human abstractions should be useful for understanding it. …You also introduce some amount of tautology here, by defining complexity in terms of (a subset of) human abstractions.
My best interpretation is that you’re making the argument that we can abstract the way in which human abstractions are interpretable (human abstractions are interpretable by being “symbolic” / “chunky”) and then search for any symbolic/chunky abstractions. You assume that the best ways to compress S gonna be symbolic/chunky. Or that we can enforce looking for symbolic/chunky representations of S. …But then the section about ontology shifts doesn’t make sense. Why are there multiple kinds of symbolicness/chunkiness?
There might be a general disagreement or misunderstanding. I would classify programs into 3 types: big and chaotic (A), small and chaotic (B), not chaotic (C). For example, neural networks are big & chaotic and Busy Beavers or pseudorandom number generators or stuff like Van Eck’s sequence are small & chaotic. The existence of B-type programs is the biggest problem in capturing the intuitive notion of “complexity” (description length doesn’t cut it = some short programs behave very weirdly). If the universe can be compressed by a B-type program, then the compression won’t be “chunky” or truly “symbolic”. So the core difficulty of formalizing chunkiness/symbolicness is excluding B-type programs. You mention this concern, but dismiss it as not a big deal.
...
What are the most substantial differences between your agenda and NAH?
Human world-models rely on chunking. To understand a complex phenomenon, we break it down into parts, understand the parts individually, then understand the whole in terms of the parts. (The human biology in terms of cells/tissues/organs, the economy in terms of various actors and forces, a complex codebase in terms of individual functions and modules.)
If you split a thing enough times, you’ll end up with simple parts. But that doesn’t mean that the whole thing is simple, even if it’s small. Small Turing Machines can have very complex behavior. I believe it’s the core problem of defining what a “chunk” is. Or what “well-abstracting” means. And I think the solution should have the following form:
“Parts of a program X are non-trivial (= can be considered ‘chunks’ or ‘symbols’) if they are substantially simpler than X and allow to prove important things about X by using a predetermined method (= a method not requiring a stroke of creativity or meticulous search through proofs) substantially faster than running X.”[1]
Examples:
Consider an algorithm which solves chess positions through exhaustive brute-force search. The operation of expanding the game tree is a non-trivial part of the algorithm, because it allows to easily prove important things about the algorithm (e.g. its runtime) faster than running it. Just by applying induction. However, if I give you the rules of a random Turing Machine, you might be unable to prove anything important by applying generic proof methods to those rules. Which would mean that those rules don’t split the program into “digestible” parts — your predetermined proof machinery can’t digest them.
By splitting the world into big areas which are assumed to be affecting each other slowly and locally, we can easily prove important things about the world.
Once you know what “organs” are, it allows you to prove extremely non-trivial things about animal bodies. E.g. “deep stab has a high chance to stop an animal forever (because it has a high chance to hit an organ)”.
You can take a variable X (e.g. “a particular human”) and measure how useful all other variables are for easily proving important things about X. You’ll get a “semantic field” of X. The field will include mesavariables (e.g. “organs”, “cells”) and metavariables (e.g. “democracy”, “social norms”). The field will fade away at overly fine-grained variables (e.g. individual atoms) and overly coarse-grained variables.
Maybe we can deduce something about the structure of abstractions from first principles, in a way completely agnostic to the fundamental building blocks of the universe. Consider this argument:
Assume we model the world as programs modifying themselves and each other.
We have a finite set (S) of abstractions which capture most important properties of the programs’ states. Those abstractions are useful for predicting the future. Since the world is predictable, not all abstractions are equally likely (≈ not all future states of programs are equally likely).
Any output of a program can serve as input for another program. Programs can modify each other in the middle of execution. This has two consequences. First: most combinations of abstractions should map to S (i.e. simplify into base abstractions). Second: any predictable program should be modified by many programs in the same way.
The latter means that the semantic field of a predictable variable (X) should have self-similar / redundant structure. A couple of silly examples in the spoiler.
Big forest stone. It doesn’t move itself; animals don’t try to move it; the landscape around it doesn’t move. So, you could say the stone maximizes “avoidance of movement” in different ways.
Lake. When animals dive into the lake, they displace water (create “holes” in the body of water), but said water comes back; the water evaporates, gets drank by animals, seeps through the ground, but comes back with rain; the water of the lake fills a hole in the landscape. So, you could say the lake maximizes “refillment” in different ways.
A question about natural latents.
Imagine you have 100 blue boxes. Each time you roll the dice, their shape changes. But all 100 boxes always share the same shape. If I understand correctly, in this situation the shape is the natural latent. While color is just static background information.
Imagine you have 100 boxes. Each time you roll the dice, their color changes. But all 100 boxes always share the same color. In this situation, color is the natural latent.
Imagine you have 100 boxes. Each time you roll the dice, their color changes. If at least one box is blue, all boxes are blue. Otherwise their color is independent. Is “all boxes are blue or all boxes have independent colors” a natural latent (it’s something you learn about all boxes by examining a single box)?
Does the latter (3) type of natural latents have any special properties, is it some sort of “meta-level” natural latent (compared to 2)? I’m asking because I think this type of latents might be relevant to how human abstractions work. Here’s where I wrote about it in more detail.
Yes, it could be that “special, inherently more alignable cognition” doesn’t exist or can’t be discovered by mere mortal humans. It could be that humanlike reasoning isn’t inherently more alignable. Finally, it could be that we can’t afford to study it because the dominating paradigm is different. Also, I realize that glass box AI is a pipe dream.
Wrt sociopaths/psychopaths. I’m approaching it from a more theoretical standpoint. If I knew a method of building a psychopath AI (caring about something selfish, e.g. gaining money or fame or social power or new knowledge or even paperclips) and knew the core reasons of why it works, I would consider it a major progress. Because it would solve many alignment subproblems, such as ontology identification and subsystems alignment.
I’m approaching it from a “theoretical” perspective[1], so I want to know how “humanlike reasoning” could be defined (beyond “here’s some trusted model which somehow imitates human judgement”) or why human-approved capability gain preserves alignment (like, what’s the core reason, what makes human judgement good?). So my biggest worry is not that the assumption might be false, but that the assumption is barely understood on the theoretical level.
What are your research interests? Are you interested in defining what “explanation” means (or at least in defining some properties/principles of explanations)? Typical LLM stuff is highly empirical, but I’m kinda following the pipe dream of glass box AI.
I’m contrasting theoretical and empirical approaches.
Empirical—“this is likely to work, based on evidence”. Theoretical—“this should work, based on math / logic / philosophy”.
Empirical—“if we can operationalize X for making experiments, we don’t need to look for a deeper definition”. Theoretical—“we need to look for a deeper definition anyway”.
Alignment plans can be split into two types:
Usual plans. AI gains capabilities . We figure out how to point to (alignment target). There’s no deep connection between and . One thing is mounted onto the other.
HRLM plans. We give AI special , with a deep connection to .
HRLM is the idea that there’s some special reasoning/learning method which is crucial for alignment or makes it fundamentally easier. HRLM means “humanlike reasoning/learning method” or “special, human-endorsed reasoning/learning method”. There’s no hard line separating the two types of plans. It’s a matter of degree.
I believe HRLM is ~never discussed in full generality and ~never discussed from a theoretical POV. This is a small post where I want to highlight the idea and facilitate discussion, not make a strong case for it.
(My description of other people’s work is not endorsed by them.)
Corrigibility. “Corrigible cognition” is a hypothetical, special type of self-reflection () which is extremely well-suited for learning human values/desires ().
In “Don’t align agents to evaluations of plans” Alex Turner argues “there’s a correct way to reason () about goals () and consequentialist maximization of an ‘ideal’ function is not it”, “‘direct cognition’ () about goals () is fundamentally better than ‘indirect cognition’”. Shard Theory, in general, proposes a very special method for learning and thinking about values.
A post about “follow-the-trying game” by Steve Byrnes basically says “AI will become aligned or misaligned at the stage of generating thoughts, so we need to figure out the ‘correct’ way of generating thoughts (), instead of overfocusing on judging what thoughts are aligned ()”. Steve’s entire agenda is about HRLM.
Large Language Models. I’m not familiar with the debate, but I would guess it boils down to two possibilities: “understanding human language is a core enough capability () for a LLM, which makes it inherently more alignable to human goals ()” and “LLMs ‘understand’ human language through some alien tricks which don’t make them inherently more alignable”. If the former is true, LLMs are an example of HRLM.
Policy Alignment (Abram Demski) is tangentially related, but it’s more in the camp of “usual plans”.
Notice how, despite multiple agendas falling under HRLM (Shard Theory, brain-like AGI, LLM-focused proposals), there’s almost no discussion of HRLM from a theoretical POV. What is, abstractly speaking, “humanlike reasoning”? What are the general principles of it? What are the general arguments for safety guarantees it’s supposed to bring about? What are the True Names here? With Shard Theory, there’s ~zero explanation of how simpler shards aggregate into more complex shards and how it preserves goals. With brain-like AGI, there’s ~zero idea of how to prevent thought generation from bullshitting thought assessment. But those are the very core questions of the agendas. So they barely move us from square one.[1]
There are many possibilities. It could be that any HRLM handicaps AI’s capabilities (a superintelligence is supposed to be unimaginably better at reasoning than humans, so why wouldn’t it have an alien reasoning method). It also could be that HRLM is necessary for general intelligence. But maybe general intelligence is overrated...
Here’s what I personally believe right now:
What we value is inherently tied to how we think about it. In general, what we think about is often inherently tied to how we think about it.
General intelligence is based on a special principle. It has a relatively crisp “core”.
Some special computational principle is needed for solving subsystems alignment.
If 1-3 is true, 1-3 is most likely the same thing. Therefore, HRLM is needed for general intelligence, outer and inner alignment (including subsystems alignment). Separately, I think general intelligence boosts capabilities below peak human level.
I consider 1-3 to be plausible enough postulates. I have no further arguments for 4.
I have a couple of very unfinished ideas. Will try to write about them this or the next month.
I believe there could be a special type of cognition which helps to avoid specification gaming and goal misgeneralization. AI should create simple models which describe “costs/benefits” of actions (e.g. “actions” can be body movements, “cost” can be the amount and complexity of movements, “benefit” can be distance covered), this way AI can notice if certain actions produce anomalously high benefit (e.g. maybe certain body movements exploit a glitch in the physics simulation, making the body cover kilometers per second).
“By default, manipulating easier to optimize/comprehend variables is better than manipulating harder to optimize/comprehend variables” — this is the idea from one of my posts. The problem with it is that I only defined “optimization” and “comprehension” for world-models, not for any modelling (= cognition) in general.
A formal algorithm can have parts and it will critically depend on those parts (for example, an algorithm for solving equations might have an absolutely necessary addition sub-algorithm). An informal algorithm can have parts without critically depending on those parts (for example, the algorithm answering “is this a picture of a dog?” might have a sub-algorithm answering “is this patch of pixels the focal point of the image / does it contrast enough with other patches / is it as detailed as the other patches?”—the sub-algorithm is not very necessary, but it lowers pareidolia, by preventing the algorithm from overanalyzing random parts of the image). I think we can say something about the latter type of algorithms, about how they work.
IMO that’s downstream of inner alignment being extremely hard. It’s almost impossible to come up with at least mildly promising solution which explains, at least in some detail, how the hardest part of the problem might get solved. I’m not trying to throw shade. Also, I might just be ignorant about some ideas in those agendas.
(Here’s an observation about adjectives, verbs, and language in general. It might be important even if I’m misinterpreting the definition of natural latents.)
For many adjectives, we can define the concept “salience of <insert an adjective>”. Salience of color / texture / shape / size / etc.
For example, what’s “salience of a texture”? It’s a function of how much of the texture is present (in your field of view) and how strongly the texture contrasts with other present textures.
We can learn an empirical rule: “if a texture is salient enough, then it’s probably caused by a single object or a single kind of objects”.[1] Yet this object or kind is random. Would this make “being caused by an object/kind X” a natural latent over “pixels with a salient texture Y” and vice-versa? An object’s texture tends to be similarly salient in many different situations, so a particular value of “salience of a texture” can itself be a natural latent. “Salience of a texture” is not the same thing as “a texture”, but it’s one of the reasons why textures are important.
landscape example
Similarly, we can consider “salience of an action”. With an empirical rule like “salient actions (e.g. salient movements) are usually caused by a single object / a single kind of objects / a single causal process”.[2] Such rule makes fine-grained classification of actions important, making action-related words important.
Beyond “salience of a texture”, we can consider concepts like “danger of an object” (is it a good idea to touch or step on?) and “traversability of an object” (can you walk/climb on it?). For many types of objects, those concepts will be independent from most facts about the objects but their texture. Let’s call concepts like this (salience, danger, traversability, etc.) “auxiliary”.
Maybe this sort of reasoning can explain analogies, connotation. Because auxiliary concepts are a type of connotation.
Imagine looking at a nature landscape. You notice a bunch of angular texture (mountains), a bunch of puffy texture (clouds), a bunch fluffy texture (trees), a bunch of smooth texture (fields). The rule says each type of texture most likely belongs to a single object or kind. Note that it’s not trivial—we could live in a world where we often see radically different things with similar, equally salient textures at the same time.
For example, imagine you see a burst of flame and rubble flying around—most likely it’s a single causal process, an explosion.
I have a couple of questions/points. Might stem from not understanding the math.
1) The very first example shows that absolutely arbitrary things (e.g. arbitrary green lines) can be “natural latents”. Does it mean that “natural latents” don’t capture the intuitive idea of “natural abstractions”? That all natural abstractions are natural latents, but not all natural latents are natural abstractions. You seem to be confirming this interpretation, but I just want to be sure:
So we view natural latents as a foundational tool. The plan is to construct more expressive structures out of them, rich enough to capture the type signatures of the kinds of concepts humans use day-to-day, and then use the guarantees provided by natural latents to make similar isomorphism claims about those more-complicated structures. That would give a potential foundation for crossing the gap between the internal ontologies of two quite different minds.
Is there any writing about what those “more expressive structures” could be?
2) Natural latents can describe both things which propagate through very universal, local physical laws (e.g. heat) and any commonalities in made up categories (e.g. “cakes”). Natural latents seem very interesting in the former case, but I’m not sure about the latter. Not sure the analogy between the two gives any insight. I’m still not seeing any substantial similarity between cakes and heat or Ising models. I.e. I see that an analogy can be made, but I don’t feel that this analogy is “grounded” in important properties of reality (locality, universality, low-levelness, stability, etc). Does this make sense?
3) I don’t understand what “those properties can in-principle be well estimated by intensive study of just one or a few mai tais” (from here) means. To me a natural latent is something like ~”all words present in all of 100 books”, it’s impossible to know unless you read every single book.
If I haven’t missed anything major, I’d say core insights about abstractions are still missing.
EDIT 17/07: I did miss at least one major thing. I haven’t understood the independence condition. If you take all words present in all 100 books, it doesn’t guarantee that those words make the books or their properties independent.
Have an idea about interpretability and defining search/optimization.
Finite algorithms can solve infinite (classes of) problems. For example, the algorithm for adding two numbers has a finite description, yet can solve an infinity of examples.
This is a basic truth of computability theory.
Intuitively, it means that algorithms can exploit regularities in problems. But “regularity” here can only be defined tautologically (any smaller thing which solves/defines a bigger thing).
Many algorithms have the following property:
The algorithm computes function by computing another function many times. Computing is useful for computing . is computed significantly more times than .
Computing is significantly easier than computing .
gives important information about . This information is easy to compute and (maybe) trivial to prove.
Intuitively, it means that algorithms can exploit regularities in problems. However, here “regularity” has a stricter definition than in the trivial property (TP). A “regularity” is an easily computable thing which gives you important, easily computable information about a hard-to-compute thing.
Now, the question is: what classes of algorithms have the less trivial property (LTP)?
LTP includes a bunch of undefined terms:
“Significantly simpler”, “easily computable”, and “useful for computing”. Those could be defined in terms of computational complexity.
“Significantly more times”. Could be defined asymptotically.
“Important information”. I have little idea how it should be defined. “Important information” may mean upper and lower bounds of a function, for example.
Probably giving fully general definitions right away is not important. We could try starting with overly restrictive definitions and see if we can prove anything about those.
If neural networks implement algorithms with LTP, we could try finding those algorithms by looking for (which is much easier than looking for ).
Furthermore, LTP seems very relevant to defining search / optimization.
Examples of algorithms with LTP:
Any greedy algorithm is an example. The whole solution to the problem is , the greedy choice is .
Many dynamic programming algorithms are an example. Solution to the whole problem is , solution to a smaller problem (which we need to use many times) is .
Consider a strong chess engine. : “given a position, find how to e.g. win material given an ~N moves horizon”. : “given a position, find how to e.g. win material given a ~K moves horizon”. Where K << N. Simply put, the N-moves-deep engine will often punish K-moves-deep mistakes. That’s the reason why you can understand that you’re losing long before you get checkmated. This wouldn’t be true in a game much more chaotic than chess.
Consider simulated annealing. : “given a function, find the global optimum”. AFAICT, simulated annealing is greedy search + randomness. The greedy part is .
Got around to interrogating Gemini for a bit.
Seems like KSF talks about programs generating sets. It doesn’t say anything about the internal structure of the programs (but that’s where the objects such as “real diamonds” live). So let’s say is a very long video about dogs doing various things. If I apply KSF, I get programs (aka “codes”) generating sets of videos. But it doesn’t help me identify “the most dog-like thing” inside each program. For example, one of the programs might be an atomic model of physics, where “the most dog-like things” are stable clouds of atoms. But KSF doesn’t help me find those clouds. A similarity metric between videos doesn’t help either.
My conceptual solution to the above problem, proposed in the post: if you have a simple program with special internal structure describing simple statistical properties of “dog-shaped pixels” (such program is guaranteed to exist), there also exists a program with very similar internal structure describing “valuable physical objects causing dog-shaped pixels” (if such program doesn’t exist, then “valuable physical objects causing dog-shaped pixels” don’t exist either).[1] Finding “the most dog-like things” in such program is trivial. Therefore, we should be able to solve ontology identification by heavily restricting the internal structure of programs (to structures which look similar to simple statistical patterns in sensory data).
So, to formalize my “conceptual solution” we need models which are visually/structurally/spatially/dynamically similar to the sensory data they model. I asked Gemini about it, multiple times, with Deep Research. The only interesting reference Gemini found is Agent-based models (AFAIU, “agents” just means “any objects governed by rules”).
This is not obvious, requires analyzing basic properties of human values.
Then you can use the three dot points in my comment to construct source code for a new agent that does the same thing, but is not nicely separated.
This is the step I don’t get (how we make the construction), because I don’t understand SGD well. What does “sample N world models” mean?
My attempt to understand: We have a space of world models () and a space of plans (). We pick points from (using SGD) and evaluate them on the best points of (we got those best points by trying to predict the world and applying SGD).
My thoughts/questions: To find the best points of , we still need to do modelling independently from planning? While the world model is not stored in memory, some pointer to the best points of is stored? We at least have “the best current plan” stored independently from the world models?
First things first, defining a special language which creates a safe but useful AGI absolutely is just a restatement of the problem, more or less. But the post doesn’t just restate the problem, it describes the core principle of the language (the comprehension/optimization metric) and makes arguments for why the language should be provably sufficient for solving a big part of alignment.
You’re saying that the simple AI can tell if the more complex AI’s plans are good, bad, or unnecessary—but also the latter “can know stuff humans don’t know”. How?
This section deduces the above from claims A and B. What part of the deduction do you disagree with/confused about? Here’s how the deduction would apply to the task “protect a diamond from destruction”:
cares about an ontologically fundamental diamond. models the world as clouds of atoms.
According to the principle, we can automatically find what object in corresponds to the “ontologically fundamental diamond”.
Therefore, we can know what plans would preserve the diamond. We also can know if applying any weird optimization to the diamond is necessary for preserving it. Checking for necessity is probably hard, might require another novel insight. But “necessity” is a simple object-level property.
The automatic finding of the correspondence (step 2) between an important comprehensible concept and an important incomprehensible concept resolves the apparent contradiction.[1]
Now, without context, step 2 is just a restatement of the ontology identification problem. The first two sections of the post (mostly the first one) explain why the comprehension/optimization metric should solve it. I believe my solution is along the lines of the research avenues Eliezer outlined.
If my principle is hard to agree with, please try to assume that it’s true and see if you can follow how it solves some alignment problems.
So, somehow you’re able to know when an AI is exerting optimization power in “a way that flows through” some specific concepts?
Yes, we’re able to tell if AI optimizes through a specific class of concepts. In most/all sections of the post I’m assuming the AI generates concepts in a special language (i.e. it’s not just a trained neural network), a language which allows to measure the complexity of concepts. The claim is that if you’re optimizing through concepts of certain complexity, then you can’t fulfill a task in a “weird” way. If the claim is true and AI doesn’t think in arbitrary languages, then it’s supposed to be impossible to create a harmful Doppelganger.
But I don’t get if, or why, you think that adds up to anything like the above.
Clarification: only the interpretability section deals with inner alignment. The claims of the previous sections are not supposed to follow from the interpretability section.
Anyway, is the following basically what you’re proposing?
Yes. The special language is supposed to have the property that can automatically learn if plans good, bad, or unnecessary actions. can’t be arbitrarily smarter than humans, but it’s a general intelligence which doesn’t imitate humans and can know stuff humans don’t know.
Could you ELI15 the difference between Kolmogorov complexity (KC) and Kolmogorov structure function (KSF)?
Here are some of the things needed to formalize the proposal in the post:
A complexity metric defined for different model classes.
A natural way to “connect” models. So we can identify the same object (e.g. “diamond”) in two different models. Related: multi-level maps.
I feel something like KSF could tackle 1, but what about 2?
Thanks for clarifying! Even if I still don’t fully understand your position, I now see where you’re coming from.
No, I think it’s what humans actually pursue today when given the options. I’m not convinced that these values are static, or coherent, much less that we would in fact converge.
Then those values/motivations should be limited by the complexity of human cognition, since they’re produced by it. Isn’t that trivially true? I agree that values can be incoherent, fluid, and not converging to anything. But building Task AGI doesn’t require building an AGI which learns coherent human values. It “merely” requires an AGI which doesn’t affect human values in large and unintended ways.
No, because we don’t comprehend them, we just evaluate what we want locally using the machinery directly, and make choices based on that.
This feels like arguing over definitions. If you have an oracle for solving certain problems, this oracle can be defined as a part of your problem-solving ability. Even if it’s not transparent compared to your other problem-solving abilities. Similarly, the machinery which calculates a complicated function from sensory inputs to judgements (e.g. from Mona Lisa to “this is beautiful”) can be defined as a part of our comprehension ability. Yes, humans don’t know (1) the internals of the machinery or (2) some properties of the function it calculates — but I think you haven’t given an example of how human values depend on knowledge of 1 or 2. You gave an example of how human values depend on the maxima of the function (e.g. the desire to find the most delicious food), but that function having maxima is not an unknown property, it’s a trivial property (some foods are worse than others, therefore some foods have the best taste).
That’s a very big “if”! And simplicity priors are made questionable, if not refuted, by the fact that we haven’t gotten any convergence about human values despite millennia of philosophy trying to build such an explanation.
I agree that ambitious value learning is a big “if”. But Task AGI doesn’t require it.
I think he already came to some conclusions and you already gave some good references (which support some of the conclusions).
Do those methods have names or address problems which have names (like the Byzantine generals problem)?