I wrote an AI Impacts page summary of the situation as I understand it. If anyone feels like looking, I'm interested in corrections/suggestions (either here or in the AI Impacts feedback box).  

A few quick thoughts on reasons for confusion:

I think maybe one thing going on is that I already took the coherence arguments to apply only in getting you from weakly having goals to strongly having goals, so since you were arguing against their applicability, I thought you were talking about the step from weaker to stronger goal direction. (I’m not sure what arguments people use to get from 1 to 2 though, so maybe you are right that it is also something to do with coherence, at least implicitly.)

It also seems natural to think of ‘weakly has goals’ as something other than ‘goal directed’, and ‘goal directed’ as referring only to ‘strongly has goals’, so that ‘coherence arguments do not imply goal directed behavior’ (in combination with expecting coherence arguments to be in the weak->strong part of the argument) sounds like ‘coherence arguments do not get you from ‘weakly has goals’ to ‘strongly has goals’.

I also think separating out the step from no goal direction to weak, and weak to strong might be helpful in clarity. It sounded to me like you were considering an argument from 'any kind of agent' to 'strong goal directed' and finding it lacking, and I was like 'but any kind of agent includes a mix of those that this force will work on, and those it won't, so shouldn't it be a partial/probabilistic move toward goal direction?' Whereas you were just meaning to talk about what fraction of existing things are weakly goal directed.

Thanks. Let me check if I understand you correctly:

You think I take the original argument to be arguing from ‘has goals' to ‘has goals’, essentially, and agree that that holds, but don’t find it very interesting/relevant.

What you disagree with is an argument from ‘anything smart’ to ‘has goals’, which seems to be what is needed for the AI risk argument to apply to any superintelligent agent.

Is that right?

If so, I think it’s helpful to distinguish between ‘weakly has goals’ and ‘strongly has goals’:

1. Weakly has goals: ‘has some sort of drive toward something, at least sometimes' (e.g. aspects of outcomes are taken into account in decisions in some way)
2. Strongly has goals: ’pursues outcomes consistently and effectively' (i.e. decisions maximize expected utility)

So that the full argument I currently take you to be responding to is closer to:

1. By hypothesis, we will have superintelligent machines
2. They will weakly have goals (for various reasons, e.g. they will do something, and maybe that means ‘weakly having goals’ in the relevant way? Probably other arguments go in here.)
3. Anything that weakly has goals has reason to reform to become an EU maximizer, i.e. to strongly have goals
4. Therefore we will have superintelligent machines that strongly have goals

In that case, my current understanding is that you are disagreeing with 2, and that you agree that if 2 holds in some case, then the argument goes through. That is, creatures that are weakly goal directed are liable to become strongly goal directed. (e.g. an agent that twitches because it has various flickering and potentially conflicting urges toward different outcomes is liable to become an agent that more systematically seeks to bring about some such outcomes) Does that sound right?

If so, I think we agree. (In my intuition I characterize the situation as ‘there is roughly a gradient of goal directedness, and a force pulling less goal directed things into being more goal directed. This force probably doesn’t exist out at the zero goal directness edges, but it unclear how strong it is in the rest of the space—i.e. whether it becomes substantial as soon as you move out from zero goal directedness, or is weak until you are in a few specific places right next to ‘maximally goal directed’.)

Good points. Though I claim that I do hold the same facial expression for long periods sometimes, if that's what you mean by 'not moving'. In particular, sometimes it is very hard for me not to screw up my face in a kind of disgusted frown, especially if it is morning. And sometimes I grin for so long that my face hurts, and I still can't stop.

(Lesswrong version here: https://www.lesswrong.com/posts/JJxxoRPMMvWEYBDpc/why-does-applied-divinity-studies-think-ea-hasn-t-grown )

I respond here: https://worldspiritsockpuppet.com/2021/03/09/why-does-ads-think-ea-hasnt-grown.html

It doesn't seem that hard to wash your hands after putting away groceries, say. If I recall, I was not imagining getting many touches during such a trip. I'm mostly imagining that you put many of the groceries you purchase in your fridge or eat them within a couple of days, such that they are still fairly contaminated if they started out contaminated, and it is harder to not touch your face whenever you are eating recently acquired or cold food.

Yes - I like 'application' over 'potentially useful product' and 'my more refined writing skills' over 'my more honed writing', in its first one, for instance.

I grab the string and/or some beads I don't want to move together between my thumb and finger on one hand, and push the bead I do want to move with my thumb and finger of the other hand. (I don't need to see it because I can feel it and the beads don't move with my touching it.) I can also do it more awkwardly with one hand.

Thanks for further varieties! I hadn't seen the ring, and have had such a clicker but have not got the hang of using it non-awkwardly (where do you put it? With your keys? Who knows where those are? In your pocket? Who reliably has a pocket that fits things in? In your bag? Then you have to dig it out..)

Good point regarding wanting to know what number you have reached. I only want to know the exact number very occasionally, like with a bank account, but I agree that's not true of many use cases.

I haven't read Zvi's post, but would have thought that the good of slack can be cashed out in efficiency, if you are optimizing for the right goals (e.g. if you have a bunch of tasks in life which contribute to various things, it will turn out that you contribute to those things better overall if you have spare time between the tasks).  

If you aren't in the business of optimizing for the ultimately right goals though, I'd think you could also include slack as one of your instrumental goals, and thus mostly avoid serious conflict e.g. instead of turning out as many cookies per minute as I can, aim to turn out as many cookies as I can while spending half my time not working on it, and setting aside a bag of each ingredient. Perhaps the thought is that this doesn't work because 'slack' is hard to specify, so if you just say that I have to be away from the cookie cutters, I might spend my time strategizing about cookies instead, and that might be somehow not slack in the right way? Plus part of the point is that if things go awry, you want me to be able to put all my time back into cookies briefly?

Thanks. What kind of gloves do you suggest?

I actually know very little about my plants at present, so cannot help you.

It is irrigation actually, not moisture sensors. Or rather, I think it irrigates based on the level of moisture, using a combination of tiny tubes and clay spikes that I admittedly don't fully understand. (It seems to be much better at watering my plants than I am, even ignoring time costs!) I do have to fill up the water container sometimes.

I meant: conditional on it growing faster, why expect this is attributable to a small number of technologies, given that when it accelerated previously it was not like that (if I understand)?

If throughout most of history growth rates have been gradually increasing, I don't follow why you would expect one technology to cause it to grow much faster, if it goes back to accelerating.

Thanks for contributing data! :D 

They are meant to be chewy, not crumbly.

Making them tastier, though not confident about this - originally motivated by not having normal flour, and then have done some of each, and thought the gluten free ones were better, but much randomness at play. 

I did mean 'white' by 'wheat'; sorry (I am a foreigner). I haven't tried anything other than the gluten free one mentioned and white wheat flour.

>Someone's cognitive labor went into making the rabbit mold, and everything from there on out is eliminating the need to repeat that labor, and to reduce the number of people who need to have that knowledge.

Yeah, that's the kind of thing I had in mind in the last paragraph.

In such a case, you might get many of the benefits without the covid risks from driving to very close to the ER, then hanging out there and not going in and risking infection unless worse symptoms develop, but being able to act very fast if they do.

1) Even if it counts as a DSA, I claim that it is not very interesting in the context of AI. DSAs of something already almost as large as the world are commonplace. For instance, in the extreme, the world minus any particular person could take over the world if they wanted to. The concern with AI is that an initially tiny entity might take over the world.

2) My important point is rather that your '30 year' number is specific to the starting size of the thing, and not just a general number for getting a DSA. In particular, it does not apply to smaller things.

3) Agree income doesn't equal taking over, though in the modern world where much purchasing occurs, it is closer. Not clear to me that AI companies do better as a fraction of the world in terms of military power than they do in terms of spending.

The time it takes to get a DSA by growing bigger depends on how big you are to begin with. If I understand, you take your 30 years from considering the largest countries, which are not far from being the size of the world, and then use it when talking about AI projects that are much smaller (e.g. a billion dollars a year suggests about 1/100,000 of the world). If you start from a situation of an AI project being three doublings from taking over the world say, then most of the question of how it came to have a DSA seems to be the question of how it grew the other seventeen doublings. (Perhaps you are thinking of an initially large country growing fast via AI? Do we then have to imagine that all of the country's resources are going into AI?)

This sounds great to me, and I think I would be likely to sign up for it if I could, but I haven't thought about it for more than a few minutes, am particularly unsure about the implications for culture, and am maybe too enthusiastic in general for things being 'well organized'.

Oh yeah, I think I get something similar when my sleep schedule gets very out of whack, or for some reason when I moved into my new house in January, though it went back to normal with time. (Potentially relevant features there: bedroom didn't seem very separated from common areas, at first was sleeping on a pile of yoga mats instead of a bed, didn't get out much.)

I think random objects might work in a similar way. e.g. if talking in a restaurant, you grab the ketchup bottle and the salt to represent your point. I've only experimented with this once, with ultimately quite an elaborate set of condiments, tableware and fries involved. It seemed to make things more memorable and followable, but I wasn't much inclined to do it more for some reason. Possibly at that scale it was a lot of effort beyond the conversation.

Things I see around me sometimes get involved in my thoughts in a way that seems related. For instance, if I'm thinking about the interactions of two orgs while I'm near some trees, two of the trees will come to represent the two orgs, and my thoughts about how they should interact will echo ways that the trees are interacting, without me intending this.

No, never heard of it, that I know of.

I'm pretty unsure how much variation in experience there is—'not much' seems plausible to me, but why do you find it so probable?

I also thought that at first, and wanted to focus on why people join groups that are already large. But yeah, lack of very small groups to join would entirely explain that. Leaving a group signaling not liking the conversation seems like a big factor from my perspective, but I'd guess I'm unusually bothered by that.

Another random friction:

• If you just sit alone, you don't get to choose the second person who joins you. I think a thing people often do rather than sitting alone is wander alone, and grab someone else also wandering, or have plausible deniability that they might be actually walking somewhere, if they want to avoid being grabbed. This means both parties get some choice.

Aw, thanks. However I claim that this was a party with very high interesting people density, and that the most obvious difference between me and others was that I ever sat alone.

I share something like this experience (food desirability varies a lot based on unknown factors and something is desirable for maybe a week and then not desirable for months) but haven't checked carefully that it is about nutrient levels in particular. If you have, I'd be curious to hear more about how.

(My main alternative hypothesis regarding my own experience is that it is basically imaginary, so you might just have a better sense than me of which things are imaginary..)

A page number or something for the 'more seasoned' link might be useful. The document is very long and doesn't appear to contain 'season-'.

The 'blander' link doesn't look like it supports the claim much, though I am only looking at the abstract. It says that 'in many instances' there have been reductions in crop flavor, but even this appears to be background that the author is assuming, rather than a claim that the paper is about. If the rest of the paper does contain more evidence on this, could you quote it or something, since the paper is expensive to see?

I am somewhat hesitant to share simple intuition pumps about important topics, in case those intuition pumps are misleading.

This sounds wrong to me. Do you expect considering such things freely to be misleading on net? I expect some intuition pumps to be misleading, but for considering all of the intuitions that we can find about a situation to be better than avoiding them.

Thanks for your thoughts!

I don't quite follow you on the intelligence explosion issue. For instance, why does a strong argument against the intelligence explosion hypothesis need to show that a feedback loop is unlikely? Couldn't we believe that it is likely, but not likely to be very rapid for a while? For instance, there is probably a feedback loop in intelligence already, where humans with better thoughts and equipment are effectively smarter, and can then devise better thoughts and equipment. But this has been true for a while, and is a fairly slow process (at least for now, relative to our ability to deal with things).

My example for high status/small was an esteemed teacher unexpectedly dropping in to see to see their student perform, and entering silently and at the last minute, then standing quietly at the back of the room by the door.

I also think they are probably wrong, but this kind of argument is a substantial part of why. So I want to see if they can be rescued from it, since that would affect their probability of being right from my perspective.

Do you think there are more compelling arguments that they are wrong, such that we need not consider ones like this? (Also just curious)

Interested in things like this, presently have a partial version that is good.

In my experience this has been less of a problem than you might expect: our landlord likes us because we are reasonable and friendly and only destroy parts of the house when we want to make renovations with our own money and so on. So they would prefer more of us to many other candidates. And since we would also prefer they have more of us, we can make sure our landlord and more of us are in contact.

I and friends have, but pretty newly; there are currently two houses two doors apart, and more friends in the process of moving into a third three doors down. I have found this good so far, and expect to continue to for now, though i agree it might be unstable long term. As an aside, there is something nice about being able to wander down the street and visit one's neighbors, that all living in one house doesn't capture.

Bostrom quotes a colleague saying that a Fields medal indicates two things: that the recipient was capable of accomplishing something important, and that he didn't. Should potential Fields medalists move into AI safety research?

The claim on p257 that we should try to do things that are robustly positive seems contrary to usual consequentialist views, unless this is just a heuristic for maximizing value.

Does anyone know of a good short summary of the case for caring about AI risk?

Did you disagree with anything in this chapter?