Concrete Advice for Forming Inside Views on AI Safety

By Neel Nanda @ 2022-08-17T23:26 (+58)

This is a linkpost to https://www.alignmentforum.org/posts/LbSWMfbAHQnFpApQ7/concrete-advice-for-forming-inside-views-on-ai-safety

A lot of people want to form inside views on AI Safety, and this is a post about concrete advice on how to get started. I have a lot of hot takes on inside views and ways I think people misframe them, so I'll begin with some thoughts on this - if you just want to concrete advice, I recommend skipping ahead. This post is aimed at people who already have some context on AI Safety + Inside Views and want help making progress, it's not pitched as totally introductory resource.

Meta Thoughts on Inside Views

This is mostly a compressed version of a previous post of mine: How I Formed My Own Views About AI Safety (and ways that trying to do this was pretty stressful and counterproductive) - go read that post if you want something more in-depth! 

Inside Views are Overrated

First point - I think people often wildly overrate inside views. I think they're important, and worth trying to cultivate (else I wouldn't write this post), but I think less so than many people (especially in the Bay Area) often think. Why?

Thoughts on how to orient to inside views

With all that said, I do think there's a bunch of reasons to care about inside views! You probably can track truth slightly better, it can be very motivating, and it's hard to do good research without a gears-level model in your head about why what you're doing matters. Some general advice for orienting to it healthily:

Inside Views = Zooming In

Now for concrete advice! Firstly, I want to present a concrete model for how to think about inside views, and how to make progress on them - inside views are about zooming in. Concretely, in this framework, inside views look like starting with some high-level confusing claim, and then breaking it down into sub-claims, breaking those down into sub-claims, etc.

Example high-level claim: It is valuable to work on reducing AGI x-risk

Breakdown:

Key features: 

Exercise 1: Practice Zooming In!

Set a 5 minute timer. Then pick a high-level question that feels important to you, and practice expanding it into sub-claims, expanding those, etc. See how far you can get!

(Note: 5 minutes is just a minimum time to ensure you get momentum - feel free to keep going!)

Example Qs:

Separate Understanding & Evaluation

The second key part of this framework is to separate understanding and evaluating. Separate the act of understanding a belief/model, and the act of deeply engaging with it and figuring out what you believe about it. Both are important, but it's extremely hard to do both at the same time - it's easy to nod along to something that feels obvious and intuitive without really getting it or noticing subtle flaws, and easy to reject a weird idea out of hand without engaging enough to see its merits. But if you first understand, holding aside your personal views, your evaluation will be far more grounded

Understanding

The goal of understanding is to have a coherent, gears-level model in your head of a perspective or set of beliefs. This can mean a bunch of things! Modelling what a specific research believes, trying to model the theory of change for an agenda, steelmanning an argument, understanding a friend's beliefs, etc. The output of understanding will normally look like a tree of arguments, as in my model of inside views, but where the goal is to have the claims in the tree follow from sub-claims and prior claims, not to be fully grounded or be airtight arguments.

You want to be as concrete and gears-level as possible - probing into flaws, noticing errors, or bits that don't quite work, but ideally holding off on properly evaluating it. This is a somewhat subtle distinction - you can't notice and try to fix errors without somewhat evaluating it. But this tends to involve mental moves like:

Ideally you want to be validating this opinion as much as you can - getting feedback from the world on how good your model is. 

Concrete ways of doing this

Exercise 2

Practice reading and summarising! Take a piece you're interested in, read it while taking rough notes as you go, and then at the end try to distill these down to a coherent narrative. I am a big fan of doing this as a bullet point structure, and trying to distill the post down to a series of key points. Set a 5 minute timer after reading, and spend at least this long summarising (it's fine to take much longer!)

Tips:

Reading recommendations:

Evaluation

The second stage is try to actually integrate these models into your inside view, or to figure out that you disagree with them (and importantly, why and how!). It's totally fine if you try evaluating and just feel really confused - these are genuinely hard and confusing questions! But it's important to note that confusion to yourself, and explicitly have it as a black box in your model, rather than lying to yourself and pretending that you really understand it.

Some techniques:

Exercise 3

Set a 5 minute timer. Pick an interesting sub-claim within your inside view tree from exercise 1 or model from exercise 2, and practice zooming in on it.

Exercise 4

Set a 5 minute timer. Pick an interesting sub-claim within your inside view tree from exercise 1 or model from exercise 2, and practice generating counter-arguments for it.

Rohin Shah @ 2022-08-18T09:54 (+36)

Lots of thoughts on this post:

Value of inside views

Inside Views are Overrated [...]

The obvious reason to form inside views is to form truer beliefs

No? The reason to form inside views is that it enables better research, and I'm surprised this mostly doesn't feature in your post. Quoting past-you:

  • Research quality - Doing good research involves having good intuitions and research taste, sometimes called an inside view, about why the research matters and what’s really going on. This conceptual framework guides the many small decisions and trade-offs you make on a daily basis as a researcher
    • I think this is really important, but it’s worth distinguishing this from ‘is this research agenda ultimately useful’. This is still important in eg pure maths research just for doing good research, and there are areas of AI Safety where you can do ‘good research’ without actually reducing the probability of x-risk.

Quoting myself:

There’s a longstanding debate about whether one should defer to some aggregation of experts (an “outside view”), or try to understand the arguments and come to your own conclusion (an “inside view”). This debate mostly focuses on which method tends to arrive at correct conclusions. I am not taking a stance on this debate; I think it’s mostly irrelevant to the problem of doing good research. Research is typically meant to advance the frontiers of human knowledge; this is not the same goal as arriving at correct conclusions. If you want to advance human knowledge, you’re going to need a detailed inside view.

Let’s say that Alice is an expert in AI alignment, and Bob wants to get into the field, and trusts Alice’s judgment. Bob asks Alice what she thinks is most valuable to work on, and she replies, “probably robustness of neural networks”. What might have happened in Alice’s head?

Alice (hopefully) has a detailed internal model of risks from failures of AI alignment, and a sketch of potential solutions that could help avert those risks. Perhaps one cluster of solutions seems particularly valuable to work on. Then, when Bob asks her what work would be valuable, she has to condense all of the information about her solution sketch into a single word or phrase. While “robustness” might be the closest match, it’s certainly not going to convey all of Alice’s information.

What happens if Bob dives straight into a concrete project to improve robustness? I’d expect the project will improve robustness along some axis that is different from what Alice meant, ultimately rendering the improvement useless for alignment. There are just too many constraints and considerations that Alice is using in rendering her final judgment, that Bob is not aware of.

I think Bob should instead spend some time thinking about how a solution to robustness would mean that AI risk has been meaningfully reduced. Once he has a satisfying answer to that, it makes more sense to start a concrete project on improving robustness. In other words, when doing research, use senior researchers as a tool for deciding what to think about, rather than what to believe.

It’s possible that after all this reflection, Bob concludes that impact regularization is more valuable than robustness. The outside view suggests that Alice is more likely to be correct than Bob, given that she has more experience. If Bob had to bet which of them was correct, he should probably bet on Alice. But that’s not the decision he faces: he has to decide what to work on. His options probably look like:

  1. Work on a concrete project in robustness, which has perhaps a 1% chance of making valuable progress on robustness. The probability of valuable work is low since he does not share Alice’s models about how robustness can help with AI alignment.
  2. Work on a concrete project in impact regularization, which has perhaps a 50% chance of making valuable progress on impact regularization.

It’s probably not the case that progress in robustness is 50x more valuable than progress in impact regularization, and so Bob should go with (2). Hence the advice: build a gearsy, inside-view model of AI risk, and think about that model to find solutions.

(Though I should probably edit that section to also mention that Bob could execute on Alice's research agenda, if Alice is around to mentor him; and that would probably be more directly impactful than either of the other two options.)

Other meta thoughts on inside views

  • Relatedly, it's much more important to understand other people's views than to evaluate them - if I can repeat a full, gears-level model of someone's view back to them in a way that they endorse , that's a lot more valuable than figuring out how much I agree or disagree with their various beliefs and conclusions.
    • [...] having several models lets you compare and contrast them, figure out novel predictions, better engage with technical questions, do much better research, etc


 

I'm having trouble actually visualizing a scenario where Alice understands Bob's views (well enough to make novel predictions that Bob endorses, and say how Bob would update upon seeing various bits of evidence), but Alice is unable to evaluate Bob's view. Do you think this actually happens? Any concrete examples that I can try to visualize?

(Based on later parts of the post maybe you are mostly saying "don't reject an expert's view before you've tried really hard to understand it and make it something that does work", which I roughly agree with.)

Forming a "true" inside view - one where you fully understand something from first principles with zero deferring - is wildly impractical.

Yes, clearly true. I don't think anyone is advocating for this. I would say I have an inside view on bio anchors as a way to predict timelines, but I haven't looked into the data for Moore's Law myself and am deferring to others on that.

People often orient to inside views pretty unhealthily.

:'(

What fraction of people who are trying to build inside views do you think have these problems? (Relevant since I often encourage people to do it)

I know some people who do great safety relevant work, despite not having an inside view.

Hmm, I kind of agree in that there are people without inside views who are working on projects that other people with inside views are mentoring them on. I'm not immediately thinking of examples of people without inside views doing independent research that I would call "great safety relevant work".

(Unless perhaps you're counting e.g. people who do work on forecasting AGI, without having an inside view on how AGI leads to x-risk? I would say they have a domain-specific inside view on forecasting AGI.)

Forming inside views will happen naturally, and will happen much better alongside actually trying to do things and contribute to safety - you don't form them by locking yourself in your room for months and meditating on safety!

Idk, I feel like I formed my inside views by locking myself in my room for months and meditating on safety. This did involve reading things other people wrote, and talking with other junior grad students at CHAI who were also orienting to the problem. But I think it did not involve trying to do things and contributing to safety (I did do some of that but I think that was mostly irrelevant to me developing an inside view).

I do agree that if you work on topic X, you will naturally form an inside view on topic X as you get more experience with it. But in AI safety that would look more like "developing a domain-specific inside view on (say) learning from human feedback and its challenges" rather than an overall view on AI x-risk and how to address it. (In fact it seems like the way to get experience with an overall view on AI x-risk and how to address it is to meditate on it, because you can't just run experiments on AGI.)

Inside views lie on a spectrum. You will never form a "true" inside view, but conversely, not having a true inside view doesn't mean you're failing, or that you shouldn't even try. You want to aim to get closer to having an inside view! And making progress here is great and worthy

Strong +1

Aim for domain specific inside views. As an interpretability researcher, it's much more important to me to have an inside view re how to make interpretability progress and how this might interact with AI X-risk, than it is for me to have an inside view on timelines, the worth of conceptual alignment work, etc.

Yes, once you've decided that you're going to be an interpretability researcher, then you should focus on an interpretability-specific inside view. But "what should I work on" is also an important decision, and benefits from a broader inside view on a variety of topics. (I do agree though that it is a pretty reasonable strategy to just pick a domain based on deference and then only build a domain-specific inside view.)

Concrete advice

inside views are about zooming in. Concretely, in this framework, inside views look like starting with some high-level confusing claim, and then breaking it down into sub-claims, breaking those down into sub-claims, etc.

I agree that this is a decent way to measure your inside view -- like, "how big can you make this zooming-in tree before you hit a claim where you have to defer" is a good metric for "how detailed your inside view is".

I'm less clear on whether this is a good way to build an inside view, because a major source of difficulty for this strategy is in coming up with the right decomposition into sub-claims. Especially in the earlier stages of building an inside view, even your first and second levels of decomposition are going to be bad and will change over time. (For example, even for something like "why work on AI safety", Buck and I have different decompositions.) It does seem more useful once you've got a relatively fleshed out inside view, as a way to extend it further -- at this point I can in fact write out a tree of claims and expect that they will stay mostly the same (at the higher levels) after a few years, and so the leaves that I get to probably are good things to investigate.

Exercises

These seem great and I'd strongly recommend people try them out :)

Owen Cotton-Barratt @ 2022-08-18T07:04 (+13)

I loved this post; I imagine myself sending it to people in the future, and I'd be interested in seeing more in this genre from you or others. [Edit: it's even better now it has Rohin's comment.]

(Mostly I'm pulled to comment because I'm confused by the apparent disconnect between value-I-perceive and karma-it's-receiving.)

Yonatan Cale @ 2022-08-23T14:04 (+4)

Such a useful post omg! Me and Soof started summarizing safety approaches following your suggestions here

David Johnston @ 2022-08-18T00:44 (+1)

Have you actually tried these exercises? If so, I'd be curious to hear what your experiences were when doing them.