TL;DR: I totally agree with the general spirit of this post, we need people to solve alignment, and weâre not on track. Go and work on alignment but before you do, try to engage with the existing research, there are reasons why it exists. There are a lot of things not getting worked on within AI alignment research, and I can almost guarantee you that within six months to a year, you can find things that people havenât worked on.
So go and find these underexplored areas in a way where you engage with what people have done before you!
Thereâs no secret elite SEAL team coming to save the day. This is it. Weâre not on track.
If timelines are short and we donât get our act together, weâre in a lot of trouble. Scalable alignmentâaligning superhuman AGI systemsâis a real, unsolved problem. Itâs quite simple: current alignment techniques rely on human supervision, but as models become superhuman, humans wonât be able to reliably supervise them.
But my pessimism on the current state of alignment research very much doesnât mean Iâm an Eliezer-style doomer. Quite the opposite, Iâm optimistic. I think scalable alignment is a solvable problemâand itâs an ML problem, one we can do real science on as our models get more advanced. But we gotta stop fucking around. We need an effort that matches the gravity of the challenge.[1]
I also agree in that Eliezerâs style of doom seems uncalled for and that this is a solvable but difficult problem. My personal p(doom) is something around 20%, and I think this seems quite reasonable.
Barely anyone is going for the throat of solving the core difficulties of scalable alignment. Many of the people who are working on alignment are doing blue-sky theory, pretty disconnected from actual ML models. Most of the rest are doing work thatâs vaguely related, hoping it will somehow be useful, or working on techniques that might work now but predictably fail to work for superhuman systems.
Now I do want to give pushback on this claim as I see a lot of people who havenât fully engaged with the more theoretical alignment landscape making this claim. There are only 300 people working on alignment, but those people are actually doing things, and most of them arenât doing blue in the sky theory.
A note on the ARC claim:
But his research now (âheuristic argumentsâ) is roughly âtrying to solve alignment via galaxy-brained math proofs.â As much as I respect and appreciate Paul, Iâm really skeptical of this: basically all deep learning progress has been empirical, often via dumb hacks[3] and intuitions, rather than sophisticated theory. My baseline expectation is that aligning deep learning systems will be achieved similarly.[4]
This is essentially a claim about the methodology of science in that working on existing systems gives more information and breakthroughs compared to working on a blue-sky theory. The current hypothesis for this is that it is just a lot more information-rich to do real-world research. This is, however, not the only way to get real-world feedback loops. Christiano is not working on blue sky theory; heâs using real-world feedback loops in a different way; he looks at the real world and looks for information thatâs already there.
A discovery of this type is, for example, the tragedy of the commons; whilst we could have created computer simulations to see the process in action, itâs 10x easier to look at the world and see the real-time failures. He tells stories and sees where they fail in the future as his research methodology. This gives bits of information on where to do future experiments, like how we would be able to tell that humans would fail to stop overfishing without actually running an experiment on it.
This is also what John Wentworth does with his research; he looks at the real world as a reference frame which is quite rich in information. Now a good question is why we havenât seen that many empirical predictions from Agent Foundations. I believe it is because alignment is quite hard, and specifically, it is hard to define agency in a satisfactory way due to some really fuzzy problems (boundaries, among others) and, therefore, hard to make predictions.
We donât want to mathematize things too early either, as doing so would put us into a predefined reference frame that it might be hard to escape from. We want to find the right ballpark for agents since if we fail we might base evaluations on something that turns out to be false.
In general, thereâs a difference in the types of problems in alignment and empirical ML; the reference class of a âsharp-left turnâ is different from something empirically verifiable as it is unclearly defined, so a good question is how we should turn one into the other. This question of how we take recursive self-improvement, inner misalignment and agent foundations into empirically verifiable ML experiments is actually something that most of the people I know in AI Alignment are currently actively working on.
I could go on with examples, but what I really want to say here is that alignment researchers are doing things; itâs just hard to realise why theyâre doing things when youâre not doing alignment research yourself. (If you want to start, book my calendly and I might be able to help you.)
So what does this mean for an average person? You can make a huge difference by going in and engaging with arguments and coming up with counter-examples, experiments and theories of what is actually going on.
I just want to say that itâs most likely paramount to engage with the existing alignment research landscape before as itâs free information and easy to fall into traps if you donât. (a good resource for avoiding some traps is Johnâs Why Not Just sequence)
Thereâs a couple of years worth of research there; it is not worth rediscovering from the ground up. Still, this shouldnât stop you, go and do it; you donât need a hero licence.
TL;DR: I totally agree with the general spirit of this post, we need people to solve alignment, and weâre not on track. Go and work on alignment but before you do, try to engage with the existing research, there are reasons why it exists. There are a lot of things not getting worked on within AI alignment research, and I can almost guarantee you that within six months to a year, you can find things that people havenât worked on.
So go and find these underexplored areas in a way where you engage with what people have done before you!
I also agree in that Eliezerâs style of doom seems uncalled for and that this is a solvable but difficult problem. My personal p(doom) is something around 20%, and I think this seems quite reasonable.
Now I do want to give pushback on this claim as I see a lot of people who havenât fully engaged with the more theoretical alignment landscape making this claim. There are only 300 people working on alignment, but those people are actually doing things, and most of them arenât doing blue in the sky theory.
A note on the ARC claim:
This is essentially a claim about the methodology of science in that working on existing systems gives more information and breakthroughs compared to working on a blue-sky theory. The current hypothesis for this is that it is just a lot more information-rich to do real-world research. This is, however, not the only way to get real-world feedback loops. Christiano is not working on blue sky theory; heâs using real-world feedback loops in a different way; he looks at the real world and looks for information thatâs already there.
A discovery of this type is, for example, the tragedy of the commons; whilst we could have created computer simulations to see the process in action, itâs 10x easier to look at the world and see the real-time failures. He tells stories and sees where they fail in the future as his research methodology. This gives bits of information on where to do future experiments, like how we would be able to tell that humans would fail to stop overfishing without actually running an experiment on it.
This is also what John Wentworth does with his research; he looks at the real world as a reference frame which is quite rich in information. Now a good question is why we havenât seen that many empirical predictions from Agent Foundations. I believe it is because alignment is quite hard, and specifically, it is hard to define agency in a satisfactory way due to some really fuzzy problems (boundaries, among others) and, therefore, hard to make predictions.
We donât want to mathematize things too early either, as doing so would put us into a predefined reference frame that it might be hard to escape from. We want to find the right ballpark for agents since if we fail we might base evaluations on something that turns out to be false.
In general, thereâs a difference in the types of problems in alignment and empirical ML; the reference class of a âsharp-left turnâ is different from something empirically verifiable as it is unclearly defined, so a good question is how we should turn one into the other. This question of how we take recursive self-improvement, inner misalignment and agent foundations into empirically verifiable ML experiments is actually something that most of the people I know in AI Alignment are currently actively working on.
This post from Alexander Turner is a great example of doing this as they try âjust retargeting the searchâ
Other people are trying other things, such as bounding the maximisation in RL into quantilisers. This would, in turn, make AI more âcontentâ with not maximising. (fun parallel to how utilitarianism shouldnât be unbounded)
I could go on with examples, but what I really want to say here is that alignment researchers are doing things; itâs just hard to realise why theyâre doing things when youâre not doing alignment research yourself. (If you want to start, book my calendly and I might be able to help you.)
So what does this mean for an average person? You can make a huge difference by going in and engaging with arguments and coming up with counter-examples, experiments and theories of what is actually going on.
I just want to say that itâs most likely paramount to engage with the existing alignment research landscape before as itâs free information and easy to fall into traps if you donât. (a good resource for avoiding some traps is Johnâs Why Not Just sequence)
Thereâs a couple of years worth of research there; it is not worth rediscovering from the ground up. Still, this shouldnât stop you, go and do it; you donât need a hero licence.