Paul Christiano has a notion of competitiveness, which seems relevant. Directions and desiderata for AI control seems to be the the place it's stated most clearly.
The following quote (emphasis in the original) is one of the reasons he gives for desiring competitiveness, and seems to be in the same ballpark as the reason you gave:
...You can’t unilaterally use uncompetitive alignment techniques; we would need global coordination to avoid trouble. If we _don’t know how to build competitive benign AI, then users/designers of AI systems have to compr_omise effic
This seems like an important worry. I've updated the main post to state that I'm now unclear whether reports are good or bad (because it seems like most of the effect comes from how others' use the information in the reports, and it's unclear to me whether they will mostly improve or worsen their judgement).
I do think that (a) people will discount lottery winners at least a bit relative to donors of the same size and (b) it's good to introduce input on funding evaluation from someone with errors that are (relatively) uncorrelated with major funding bodies' errors.
That the use of the funds will be worse when writing a report is plausible. Do you also think that reports change others' giving either negligibly or negatively?
I guess it depends on the details of the returns to scale for donors. If there are returns to scale across the whole range of possible values of the donor lottery, as long as one person who would do lots of work/has good judgment joins the donor lottery, we should be excited about less conscientious people joining as well.
To be more concrete, imagine the amount of good you can do with a donation goes with the square of the donation. Let's suppose one person who will be a good donor joins the lottery with $1. Everyone else in the lottery will make a neutral
...I don't have to if it doesn't seem worth the opportunity cost
Thanks for highlighting in this comment. It don't think I made that prominent enough in the post itself
Sorry, I didn't communicate what I meant well there.
It might be the case that DALYs somewhat faithfully track both (a) the impact of conditions on subjective wellbeing and (b) the impact of conditions on economic contribution, even if they're not explicitly intended to track (b). It might also be the case that efforts to extend DALYs to more faithfully track (a) for things that are worse than death would mean that they tracked (b) less well in those cases.
Then, it could be the case that it's better to stick with the current way of doing things.
I don't actu
...I agree that this seems important.
If I remember/understand correctly, the normal instruments fail to deliver useful answers for very bad conditions. For example, if you administer a survey asking how many years of healthy life the survey-taker thinks a year where they suffer X is worth, very bad situations generate incredibly broad answers.
Some people say those years are valueless (so just at 0), some say they have huge disvalue (so they'd rather die now than face one year with the condition and then the rest of their life in good health), and some say tha
...This has gotten a few downvotes. My best guess as to the causes are the section headers and the picture, but I'm not sure. So I'm going to add four subcomments to this: "I downvoted and the section headers were a reason why", "I downvoted and the picture was a reason why", "I downvoted and something else stylistic was a reason why", "I downvoted and the claims or arguments were a reason why". If you downvote, I'd be grateful if you indicate a reason why (either by commenting or voting on a comment)
unfortunately I wasn’t able to include a table of contents
on the greaterwrong version of the EA forum, there's an automatically generated TOC. So that's an option for people who would strongly prefer a TOC
I have been feeling the siren song of agent-based models recently (I think it seems a natural move in a lot of cases, because we are actually modelling agents), but your criticisms of them reminded me that they often don't pay for their complexity in better predictions. It seems quite a general and useful point, and perhaps could be extracted to a st
...I split out the comments into areas of concern over on lesswrong. I think it would be a bit too noisy to duplicate that over here, but do feel free to bring up any issues!
Seems right, though I don't know to what extent Paul's view is representative of OpenAI's overall view.