AGI safety from first principles: Introduction

I haven't had time to reread this sequence in depth, but I wanted to at least touch on how I'd evaluate it. It seems to be aiming to be both a good introductory sequence, while being a "complete and compelling case I can for why the development of AGI might pose an existential threat".

The question is who is this sequence for,  what is it's goal, and how does it compare to other writing targeting similar demographics. 

Some writing that comes to mind to compare/contrast it with includes:

• Scott Alexander's Superintelligence FAQ. This is the post I've found most helpful for convincing people (including myself), that yes, AI is just actually a big deal and an extinction risk. It's 8000 words. It's written fairly entertainingly. What I find particularly compelling here are a bunch of factual statements about recent AI advances that I hadn't known about at the time.
• Tim Urban's Road To Superintelligence series. This is even more optimized for entertainingness. I recall it being a bit more handwavy and making some claims that were either objectionable, or at least felt more objectionable. It's 22,000 words.
• Alex Flint's AI Risk for Epistemic Minimalists. This goes in a pretty different direction – not entertaining, and not really comprehensive either . It came to mind because it's doing a sort-of-similar thing of "remove as many prerequisites or assumptions as possible". (I'm not actually sure it's that helpful, the specific assumptions it's avoiding making don't feel like issues I expect to come up for most people, and then it doesn't make a very strong claim about what to do)

(I recall Scott Alexander once trying to run a pseudo-study where he had people read a randomized intro post on AI alignment, I think including his own Superintelligence FAQ and Tim Urban's posts among others, and see how it changed people's minds. I vaguely recall it didn't find that big a difference between them. I'd be curious how this compared)

At a glance, AGI Safety From First Principles seems to be more complete than Alex Flint's piece, and more serious/a-bit-academic than Scott or Tim's writing. I assume it's aiming for a somewhat skeptical researcher, and is meant to not only convince them the problem exists, but give them some technical hooks of how to start thinking about it. I'm curious how well it actually succeeds at that.

Early work tends to be less relevant in the context of modern machine learning

I'm curious why you think the orthogonality thesis, instrumental convergence, the treacherous turn or Goodhart's law arguments are less relevant in the context of modern machine learning. (We can use here Facebook's feed-creation-algorithm as an example of modern machine learning, for the sake of concreteness.)

I'm excited about this sequence!

Just a question: what audience do you have in mind? Is it a sequence for newcomers to AI Safety, or more a reframing of AI Safety arguments for researchers?

Promoted to curated: I really enjoyed reading through this sequence. I have some disagreements with it, but overall it's one of the best plain language introductions to AI safety that I've seen, and I expect I will link to this as a good introduction many times in the future. I was also particularly happy with how the sequence bridged and synthesized a number of different perspectives that usually feel in conflict with each other.

Planned summary of this sequence for the Alignment Newsletter:

This sequence presents the author’s personal view on the current best arguments for AI risk, explained from first principles (that is, without taking any previous claims for granted). The argument is a specific instantiation of the _second species argument_ that sufficiently intelligent AI systems could become the most intelligent species, in which case humans could lose the ability to create a valuable and worthwhile future.

We should clarify what we mean by superintelligence, and how it might arise. The author considers intelligence as quantifying simply whether a system “could” perform a wide range of tasks, separately from whether it is motivated to actually perform those tasks. In this case, we could imagine two rough types of intelligence. The first type, epitomized by most current AI systems, trains an AI system to perform many different tasks, so that it is then able to perform all of those tasks; however, it cannot perform tasks it has not been trained on. The second type, epitomized by human intelligence and <@GPT-3@>(@Language Models are Few-Shot Learners@), trains AI systems in a task-agnostic way, such that they develop general cognitive skills that allow them to solve new tasks quickly, perhaps with a small amount of training data. This second type seems particularly necessary for tasks where data is scarce, such as the task of being a CEO of a company. Note that these two types should be thought of as defining a spectrum, not a binary distinction, since the type of a particular system depends on how you define your space of “tasks”.

How might we get AI systems that are more intelligent than humans? Assuming we can get a model to human-level intelligence, there are then three key advantages of an AI system that would allow it to go further. First, they can be easily replicated, suggesting that we could get a _collective_ superintelligence via a collection of replicated AI systems working together and learning from each other. Second, there are no limits imposed by biology, and so we can e.g. make the models arbitrarily large, unlike with human brains. Finally, the process of creation of AI systems will be far better understood than that of human evolution, and AI systems will be easier to directly modify, allowing for AI systems to recursively improve their own training process (complementing human researchers) much more effectively than humans can improve themselves or their children.

The second species argument relies on the argument that superintelligent AI systems will gain power over humans, which is usually justified by arguing that the AI system will be goal-directed. Making this argument more formal is challenging: the EU maximizer framework <@doesn’t work for this purpose@>(@Coherent behaviour in the real world is an incoherent concept@) and applying the intentional stance only helps when you have some prior information about what goals the AI system might have, which begs the question.

The author decides to instead consider a more conceptual, less formal notion of agency, in which a system is more goal-directed the more its cognition has the following properties: (1) self-awareness, (2) planning, (3) judging actions or plans by their consequences, (4) being sensitive to consequences over large distances and long time horizons, (5) internal coherence, and (6) flexibility and adaptability. (Note that this can apply to a single unified model or a collective AI system.) It’s pretty hard to say whether current training regimes will lead to the development of these capabilities, but one argument for it is that many of these capabilities may end up being necessary prerequisites to training AI agents to do intellectual work.

Another potential framework is to identify a goal as some concept learned by the AI system, that then generalizes in such a way that the AI system pursues it over longer time horizons. In this case, we need to predict what concepts an AI system will learn and how likely it is that they generalize in this way. Unfortunately, we don’t yet know how to do this.

What does alignment look like? The author uses <@intent alignment@>(@Clarifying "AI Alignment"@), that is, the AI system should be “trying to do what the human wants it to do”, in order to rule out the cases where the AI system causes bad outcomes through incompetence where it didn’t know what it was supposed to do. Rather than focusing on the outer and inner alignment decomposition, the author prefers to take a holistic view in which the choice of reward function is just one (albeit quite important) tool in the overall project of choosing a training process that shapes the AI system towards safety (either by making it not agentic, or by shaping its motivations so that the agent is intent aligned).

Given that we’ll be trying to build aligned systems, why might we still get an existential catastrophe? First, a failure of alignment is still reasonably likely, since (1) good behavior is hard to identify, (2) human values are complex, (3) influence-seeking may be a useful subgoal during training, and thus incentivized, (4) it is hard to generate training data to disambiguate between different possible goals, (5) while interpretability could help it seems quite challenging. Then, given a failure of alignment, the AI systems could seize control via the mechanisms suggested in <@What failure looks like@> and Superintelligence. How likely this is depends on factors like (1) takeoff speed, (2) how easily we can understand what AI systems are doing, (3) how constrained AI systems are at deployment, and (4) how well humanity can coordinate.

Planned opinion:

I like this sequence: I think it’s a good “updated case” for AI risk that focuses on the situation in which intelligent AI systems arise through training of ML models. The points it makes are somewhat different from the ones I would make if I were writing such a case, but I think they are still sufficient to make the case that humanity has work to do if we are to ensure that AI systems we build are aligned.

Note: There is currently a lot of stuff I want to cover in the newsletter, so this will probably go out in the 10/21 newsletter.

Translation into Russian by me: 1, 2, 3, 4, 5, 6