elityre has done work on this for BERI, suggesting >30 questions.
Regarding the question metatype, Allan Dafoe has offered a set of desiderata in the appendix to his AI governance research agenda.
If true, sounds like a bug and not a feature of lw.
So: habryka did say “anyone” in the original description, and so he will pay both respondents who completed the bounty according to original specifications (which thereby excludes gjm). I will only pay Radamantis as I interpreted him as “claiming” the task with his original comment.
I suggest you pm with payment details.
I’ll PM habryka about what to do with the bounty given that there were two respondents.
Overall I’m excited this data and analysis was generated and will sit down to take a look and update his weekend. :)
What’s your “reasonable sounding metric” of success?
I add $30 to the bounty.
There are 110 items in the list. So 25% is ~28.
I hereby set the random seed as whatever will be the last digit and first two decimals (3 digits total) of the S&P 500 Index price on January 7, 10am GMT-5, as found in the interactive chart by Googling “s&p500”.
For example, the value of the seed on 10am January 4 was “797”.
[I would have used the NIST public randomness beacon (v2.0) but it appears to be down due to government shutdown :( ].
Instructions for choosing the movements
Let the above-generated seed be n.
indices = sorted(random.sample([i for i in range(1,111)], 28))
I’m confused: why doesn’t variability cause any trouble in the standard models? It seems that if producers are risk-averse, it results in less production than otherwise.
I’m confused. Do you mean “worlds” as in “future trajectories of the world” or as in “subcommunities of AI researchers”? And what’s a concrete example of gains from trade between worlds?
The link to “many rigorous well-controlled studies” is broken.
Suppose your goal is not to maximise an objective, but just to cross some threshold. This is plausibly the situation with existential risk (e.g. “maximise probability of okay outcome”). Then, if you’re above the threshold, you want to minimise variance, whereas if you’re below it, you want to maximise variance. (See this for a simple example of this strategy applied to a game.) If Richard believes we are currently above the x-risk threshold and Ben believes we are below it, this might be a simple crux.
However, I think the distribution of success is often very different from the distribution of impact, because of replacement effects. If Facebook hadn’t become the leading social network, then MySpace would have. If not Google, then Yahoo. If not Newton, then Leibniz (and if Newton, then Leibniz anyway).
I think this is less true for startups than for scientific discoveries, because of bad Nash equilibrium stemming from founder effects. The objective which Google is maximising might not be concave. It might have many peaks, and which you reach might be quite arbitrarily determined. Yet the peaks might have very different consequences when you have a billion users.
For lack of a concrete example… suppose a webapp W uses feature x, and this influences which audience uses the app. Then, once W has scaled and depend on that audience for substantial profit they can’t easily change x. (It might be that changing x to y wouldn’t decrease profit, but just not increase it.) Yet, had they initially used y instead of x, they could have grown just as big, but they would have had a different audience. Moreover, because of network effects and returns to scale, it might not be possible for a rivalling company to build their own webapp which is basically the same thing but with y instead.
I don’t want to base my argument on that video. It’s based on the intuitions for philosophy I developed doing my BA in it at Oxford. I expect to be able to find better examples, but don’t have the energy to do that now. This should be read more as “I’m pointing at something that others who have done philosophy might also have experienced”, rather than “I’m giving a rigorous defense of the claim that even people outside philosophy might appreciate”.
That still seems too vague to be useful. I don’t have the slack to do the work of generating good examples myself at the moment.
This would have been more readable if you gave concrete examples of each kind of problem. It seems like your claim might be a useful dichotomy, but in its current state it’s likely not going to cause me to analyse problems differently or take different actions.
Somewhat tangential, but...
You point to the following process:
Generation --> Evaluation --> Acceptance/rejection.
However, generation is often risky, and not everyone has the capacity to absorb that risk. For example, one might not have the exploratory space needed to pursue spontaneous 5-hour reading sprints while working full-time.
Hence, I think much of society looks like this:
Justification --> Evaluation --> Generation --> Evaluation --> Acceptance/rejection.
I think we some very important projects never happen because the people who have taken all the inferential steps necessary to understand them are not the same as those evaluating them, and so there’s an information-asymmetry.
That’s one of California’s hidden advantages: the mild climate means there’s lots of marginal space. In cold places that margin gets trimmed off. There’s a sharper line between outside and inside, and only projects that are officially sanctioned — by organizations, or parents, or wives, or at least by oneself — get proper indoor space. That raises the activation energy for new ideas. You can’t just tinker. You have to justify.
(This is one of the problems I model impact certificates as trying to solve.)
Have you written any of the upcoming posts yet? If so, can you link them?
How causally important was Dutch-book theorems in suggesting to you that market behaviour could be used for logical induction? This seems like the most “non sequitur” part of the story. Suddenly, what seems like a surprising insight was just there.
I predict somewhere between “very” and “crucially” important.