I’m Harry Altman. I do strange sorts of math.
Posts I’d recommend:
A summary of Savage’s foundations for probability and utility—if the arguments used to ground probability and utility seem circular to you, here’s a non-circular way of doing it.
Sniffnoy
Karma: 4,170
Unfortunately, no. I see that I have a largely complete draft written from 7 years ago, but I’m not sure if I’d still want to post it in that form. Maybe worth reviewing at some point.
More on policy arguments and the AB problem
This doesn’t appear to be what’s usually meant by “consequentialism”?
Oh, nice! I guess this is basically saying the same thing as that, I’ll add a link.
Oh wow I’d forgotten about that!
Suggesting some revisions to Graham’s hierarchy of disagreement
Workflow vs interface vs implementation
Quick thoughts on the difficulty of widely conveying a non-stereotyped position
Doing principle-of-charity better
X as phenomenon vs as policy, Goodhart, and the AB problem
Consequentialism is for making decisions
Yes, I think I’d agree with that.
Whether you actually look up the answer seems irrelevant? Is it that misleading the Lonesome is easier if the Collective knows the answer?
I think the idea is that while you can lie about this, in reality things go fairly differently in cases where you looked it up in a way that’s noticeable.
Does it seem like it’s working as practice?
Hm, not really I guess. (Oops, I guess I forgot to mention that part!)
I’ve run this several times at OBNYC, it’s gone pretty well. Generally we didn’t bother with scoring. One issue with scoring is needing to come up with what counts for numerical questions. Although we tried to do that anyway, because we wanted to score individual questions even if we weren’t keeping score overall. For many things you can use “order of magnitude and first digit”, but that doesn’t work well for everything. Dates we generally did plus or minus 10 years. But it may need to vary a bit depending on just what the question. Maybe plus or minus some fixed percentage for many of them? (10%? 20%?) We did plus or minus an inch for a question about Conan O’Brien’s height.
One modification that got suggested at the most recent one was to say that on a 1, you look up the answer and lie; this is so that when you say “we looked it up” this is less informative. We never actually rolled a 1 after making this change, however. Perhaps one should add lookups on 5 as well if you’re doing this, to really make it uninformative? (So that the truth:lie ratio is 2:1 regardless of whether you’re doing a lookup or not.)
(At earlier ones we had for a while a “no talking about the die roll” rule that would make this unnecessary, but people didn’t like that.)
Having a good source of questions has been a little bit of a problem. The provided list isn’t that great—we’ve used questions from our copy of Wits & Wagers, or lists online, or just making ones up. Make sure you have some sort of question source!
We could also point to sleepwalkers of various sorts: even when executing complex actions (like murdering someone), I’ve never seen any accounts which mention deeply felt emotions. (WP emphasizes their dullness and apathetic affect.)
Nitpick: Sleepwalking proper apparently happens during non-REM sleep; acting out a dream during REM sleep is different and has its own name. Although it seems like sleepwalkers may also be dreaming somehow even though they aren’t in REM sleep? I don’t know—this is definitely not my area—and arguably none of this is relevant to the original point; but I thought I should point it out.
Ha! OK, that is indeed nasty. Yeah I guess CASes can solve this kind of problem these days, can’t they? Well—I say “these days” as if it this hasn’t been the case for, like, my entire life, I’ve just never gotten used to making routine use of them...
One annoying thing in reading Chapter 3 -- chapter 3 states that for l=2,4,8, the optimal scoring rules can be written in terms of elementary functions. However, you only actually give the full formula for the case l=8 (for l=2 you give it on half the interval). What are the formulas for the other cases?
(But also, this is really cool, thanks for posting this!)
I think some cases cases of what you’re describing as derivation-time penalties may really be can-you-derive-that-at-all penalties. E.g., with MWI and no Born rule assumed, it doesn’t seem that there is any way to derive it. I would still expect a “correct” interpretation of QM to be essentially MWI-like, but I still think it’s correct to penalize MWI-w/o-Born-assumption, not for the complexity of deriving the Born rule, but for the fact that it doesn’t seem to be possible at all. Similarly with attempts to eliminate time, or its distinction from space, from physics; it seems like it simply shouldn’t be possible in such a case to get something like Lorentz invariance.
Why do babies need so much sleep then?
Given that at the moment we don’t really understand why people need to sleep at all, I don’t think this is a strong argument for any particular claimed function.
Note probably best in general to avoid phrasing things in terms of digits, due to the possibility of a cascade of 9s. Here I guess since we’re not getting 9s it’s not an issue. But yes as you can say you can bound the errors and see that the ranges don’t overlap!