Posts
Comments
I've got a few people interested in an effective altruism version of this, plus a small database of cards. Suggestions on how to proceed?
I got the same piece of advice - to think about things in terms of "wants" rather than "shoulds" or "have tos" - from someone outside the LW bubble, but in the context of things like doing my tax returns.
She teaches social skills to nerds at CFAR workshops. She has an incredibly positive view of humanity and of what people are capable of, and meeting her massively increased the size of my reference class for what a rational person is like.
LW Women Submissions
a call for anonymous submissions by the women on LW
Seven women submitted
uh... could this be rephrased?
I'm a very satisfied customer from the March workshop. The biggest win for me has been with social skills - it turns out that anxiety had been making me stupid, and that if I de-stress then whole new parts of my brain spring into action. And that was just one of a large number of practical insights. I was amazed at both how much CFAR know about how we can use our brains effectively, and at how much they were able to teach over 4 days. Really impressive, well-run effort with a buzz of "this is where it's happening".
I promised I'd write about this in more detail, so stay tuned!
The best thing about this was that there was very little status dynamic within the CFAR house - we were all learning together as equals.
Agree with purchasing non-sketchiness signalling and utilons separately. This is especially important if like jkaufman a lot of your value comes as an effective altruist role model
Agree that if diversification is the only way to get the elephant to part with its money then it might make sense.
Similarly, if you give all your donations to a single risky organization and they turn out to be incompetent then it might demotivate your future self. So you should hedge against this, which again can be done separately from purchasing the highest-expected-value thing.
Confused about what to do if we know we're in a situation where we're behaving far from rational agents but aren't sure exactly how. I think this is the case with purchasing xrisk reduction, and with failure to reach Aumann agreement between aspiring effective altruists. To what extent do the rules still apply?
Lots of valid reasons for diversification can also serve as handy rationalizations. Diversification feels like the right thing to do - and hey, here are the reasons why! I feel like diversification should feel like the wrong thing to do, and then possibly we should do it anyway but sort of grudgingly.
I can imagine having a preference for saving at least X lives
I feel like you've got a point here but I'm not quite getting it. Our preferences are defined over outcomes, and I struggle to see how "saving X lives" can be seen as an outcome - I see outcomes more along the lines of "X number of people are born and then die at age 5, Y number of people are born and then die at age 70". You can't necessarily point to any individual and say whether or not they were "saved".
I generally think of "the utility of saving 6 lives" as a shorthand for something like "the difference in utility between (X people die at age 5, Y people die at age 70) and (X-6 people die at age 5, Y+6 people die at age 70)".
We'd have to use more precise language if that utility varies a lot for different choices of X and Y, of course.
I felt like I gained one insight, which I attempted to summarize in my own words in this comment.
It also slightly brought into focus for me the distinction between "theoretical decision processes I can fantasize about implementing" and "decision processes I can implement in practice by making minor tweaks to my brain's software". The first set can include self-less models such as paperclip maximization or optimizing those branches where I win the lottery and ignoring the rest. It's possible that in the second set a notion of self just keeps bubbling up whatever you do.
One and a half insights is pretty good going, especially on a tough topic like this one. Because of inferential distance, what feels like 10 insights to you will feel like 1 insight to me - it's like you're supplying some of the missing pieces to your own jigsaw puzzle, but in my puzzle the pieces are a different shape.
So yeah, keep hacking away at the edges!
I can imagine that if you design an agent by starting off with a reinforcement learner, and then bolting some model-based planning stuff on the side, then the model will necessarily need to tag one of its objects as "self". Otherwise the reinforcement part would have trouble telling the model-based part what it's supposed to be optimizing for.
This is like a whole sequence condensed into a post.
Incidentally, in case it's useful to anyone... The way I originally processed the $112M figure (or $68M as it then was), was something along the lines of:
- $68M pledged
- apply 90% cynicism
- that gives $6.8M
- that's still way too large a number to represent actual ROI from $170K worth of volunteer time
- how can I make this inconvenient number go away?
aha! This is money that's expected to roll in over the next several decades. We really have no idea what the EA movement will turn into over that time, so should apply big future discounting when it comes to estimating our impact
(note it looks like Will was more optimistic, applying 67% cynicism to get from $400 to $130)
This implies immediately that 75-80% haven't, and in practise that number will be higher care of the self-reporting. This substantially reduces the likely impact of 80,000 hours as a program.
Reduces it from what? There's a point at which it's more cots effective to just find new people than carrying on working to persuade existing ones. My intuition doesn't say much about whether this happy point is above or below 25%.
Good point about self-reporting potentially exaggerating the impact though.
The pledging back-of-the-envelope calculation got me curious, because I had been assuming GWWC wouldn't flat out lie about how much had been pledged (they say "We currently have 291 members ... who together have pledged more than 112 million dollars" which implies an actual total not an estimate).
On the other hand, it's just measuring pledges, it's not an estimate of how much money anyone expects to actually materialise. It hadn't occurred to me that anyone would read it that way - I may be mistaken here though, in which case there's a genuine issue with how the number is being presented.
Anyway, I still wasn't sure the pledge number made sense so I did my own back-of-the-envelope:
£72.68M pledged 291 members £250K pledged per person over the course of their life 40 years average expected time until retirement (this may be optimistic. I get the impression most members are young though) £6.2K average pledged per member per year
That would mean people are expecting to make £62K per year averaged over their entire remaining career, which still seems very optimistic. But:
- some people will be pledging more than 10%
- there might be some very high income people mixed in there, dragging the mean up.
So I think this passes the laugh test for me, as a measure of how much people might conceivably have pledged, not how much they'll actually deliver.
I love the landmine metaphor - it blows up in your face and it's left over from some ancient war.
Did he mean if they're someone else's fault then you have to fix the person?
You also know your own results aren't fraudulent.
That experiment has changed Latham's opinion of priming and has him wondering now about the applications for unconscious primes in our daily lives.
He seems to have skipped right over the part where he wonders why he and Bargh see one thing and other people see something different. Do people update far more strongly on evidence if it comes from their own lab?
Also, yay priming! (I don't want this comment to sound negative about priming as such)
2 sounds wrong to me - like you're trying to explain why having a consistent internal belief structure is important to someone who already believes that.
The things which would occur to me are:
- If both of you are having reactions like this then you're dealing with status, in-group and out-group stuff, taking offense, etc. If you can make it not be about that and be about the philosophical issues - if you can both get curious - then that's great. But I don't know how to make that happen.
- Does your friend actually have any contradictory beliefs? Do they believe that they do?
- You could escalate - point out every time your friend applies a math thing to social justice. "2000 people? That's counting. You're applying a math thing there." "You think this is better than that? That's called a partial ordering and it's a math thing". I'm not sure I'd recommend this approach though.
This may appear self-evident to you, but not necessarily to your "socially progressive" friend. Can you make a convincing case for it?
Remember you have to make a convincing case without using stuff like logic
Not that I know of
Any advice on how to set one up? In particular how to add entries to it retrospectively - I was thinking about searching the comments database for things like "I intend to", "guard against", "publication bias" etc. and manually find the relevant ones. This is somewhat laborious, but the effect I want to avoid is "oh I've just finished my write-up (or am just about to), now I'll go and add the original comment to the anti-publication bias registry".
On the other hand it seems like anyone can safely add anyone else's comment to the registry as long as it's close enough in time to when the comment was written.
Any advice? (I figured if you're involved at CFAR you might know a bit about this stuff).
This is interesting. People who are vulnerable to the donor illusion either have some of their money turned into utilons, or are taught a valuable lesson about the donor illusion, possibly creating more utilons in the long term.
This is useful to me as I'll be attending the March workshop. If I successfully digest any of the insights presented here then I'll have a better platform to start from. (Two particular points are the stuff about the parasympathetic nervous system, which I'd basically never heard of before, and the connection between the concepts of "epistemic rationality" and "knowing about myself" which is more obvious-in-retrospect).
Thanks for the write-up!
And yes, I'll stick up at least a brief write-up of my own after I'm done. Does LW have an anti-publication-bias registry somewhere?
There's probably better stuff around, but it made me think of Hanson's comments in this thread:
There's probably better stuff around, but it made me think of Hanson's comments in this thread:
just because a computer is doing something that a human could not do without understanding does not mean the computer must be understanding it as well
I think linking this concept in my mind to the concept of the Chinese Room might be helpful. Thanks!
More posts like this please!
As part of Singularity University's acquisition of the Singularity Summit, we will be changing our name and ...
OK, this is big news. Don't know how I missed this one.
Appoint a chief editor. Chief's most important job would be to maintain a list of what most urgently needs adding or expanding in the wiki, and posting a monthly Discussion post reminding people about these. (Maybe choosing a different theme each month and listing a few requested edits in that category, together with a link to the wiki page that contains the full list).
When people make these changes, they can add a comment and chief editor (or some other high status figure) will respond with heaps of praise.
People will naturally bring up any other topics they'd like to see on the wiki or general comments about the wiki. Chief editor should take account of these and where relevant bring them up with the relevant people (e.g. the programmers).
Do you think "ugh" should be listed as a response to survey questions? (Or equivalently a check box that says "I've left some answers blank due to ugh field rather than due to not reading the question" - not possible with the current LW software, just brainstorming)
This might be helpful - thanks.
My answer for Exercise would be "I am trying this hack right now and so the results haven't come in yet" (so I answered "write-in").
I answered "I feel I should try" for lukeprog's algorithm, but it's really more of a "I'll put it on my list of hacks to try at some point, but with low priority as there's a whole bunch of others I should try first".
I like the title too, especially as it gives no information about what the survey is going to be about. (Still might be distorted as people's productivity experiences might correlate with how much time they spend filling in surveys on LW... but not sure there's much that we can do about that)
and if the AI can tell if its in a simulation vs the real world then its not really a test at all.
The AI would probably assign at least some probability to "the humans will try to test me first, but do a poor job of it so I can tell whether I'm in a sim or not"
If I understand Will's response correctly (under "Earmarking"), it's best to think of GWWC, 80K, EAA and LYCS as separate organizations (at least in terms of whose money will be used for what, which is what really matters). I don't know if this addresses your concern though.
I admit it makes the actual physical donation process look slightly clunky (no big shiny donate button), but my impression is they're not targeting casual donors so much so this may not be such a problem.
This is really detailed, and exceeded my expectations! Thank you!
Oh wow, totally wasn't expecting you to go ahead and answer that particular list of questions. Thanks for being so proactive!
Questions 7-11 aren't really relevant to FHI. Question 16 is relevant (at least the the "are there other orgs similar to you?" part) but I'm guessing you'd answer no to that?
The other answers are helpful, thanks!
Actually, the relevant thing isn't whether it's superlinear but whether a large AI/firm is more innovative than a set of smaller ones with the same total size. I was assuming that the latter would be linear, but it's probably actually sublinear as you'd expect different AIs/firms to be redundantly researching the same thing.
Big thank you to Hanson for helping illuminate what it is he thinks they're actually disagreeing about, in this comment:
Eliezer, it may seem obvious to you, but this is the key point on which we've been waiting for you to clearly argue. In a society like ours, but also with one or more AIs, and perhaps ems, why would innovations discovered by a single AI not spread soon to the others, and why would a non-friendly AI not use those innovations to trade, instead of war?
Just a thought: given a particular state-of-the-art, does an AI's innovation rate scale superlinearly with its size? If it does, an AI could go something like foom even if it chose to trade away all of its innovations, as it would stay more productive than all of its smaller competitors and just keep on growing.
The analogy with firms would suggest it's not like this; the analogy with brains is less clear. Also I get the sense that this doesn't correctly describe Yudkowsky's foom (which is somehow more meta-level than that).
the only person so far to actually answer the goddamn prompt
What's worse is I wasn't even consciously aware that I was doing that. I'll try and read posts more carefully in the future!
OK - I wasn't too sure about how these ones should be worded.
Another dimension: value discovery.
- Fantastic: There is a utility function representing human values (or a procedure for determining such a function) that most people (including people with a broad range of expertise) are happy with.
- Pretty good: Everyone's values are different (and often contradict each other), but there is broad agreement as to how to aggregate preferences. Most people accept FAI needs to respect values of humanity as a whole, not just their own.
- Sufficiently good: Many important human values contradict each other, with no "best" solution to those conflicts. Most people agree on the need for a compromise but quibble over how that compromise should be reached.
I like how you've partitioned things up into IA/government/status/memes/prediction/xrisk/security and given excellent/good/ok options for each. This helps imagine mix-and-match scenarios, e.g. "FAI team has got good at security but sucks at memes and status".
A few quick points:
The fantastic list has 8 points and the others have 7 (as there are two "government" points). This brings me on to...
Should there be a category for "funding"? The fantastic/good/ok options could be something like:
- Significant government funding and/or FAI team use their superior rationality to acquire very large amounts of money through business
- Attracts small share of government/philanthropy budgets and/or a lot of small donations from many individuals
- Shoestring budget, succeeds anyway because the problem turns out to not be so hard after all once you've acquired the right insights
Does it have to be the FAI team implementing the "other xrisk reduction efforts" or can it just be "such institutions exist"?
Another donation opportunity came up recently, which I responded to with a big long list of questions and I'll put the answers up when I get them. People seemed to like this approach - can we do something similar for the FHI?
Some thoughts:
- Are the people at FHI going to be too busy to answer this kind of stuff?
- Are they likely to be limited in how candid they can be with their answers if the answers are going to be made public?
- I'm guessing Stuart or Sean would be the people you'd recommend talking to?
Note that the Tides Foundation is not the same thing as CEA. I'm not sure what CEA's exact relationship is with the Tides Foundation - I'll add this to the list of questions.
My guess would be that the relationship to Tides is necessary in order to get US tax deductability (CEA is based in the UK), and that splitting off 80K and GWWC from each other wouldn't help with that. I will ask though.
Yes - fair enough.
How many of them wouldn't believe it if it wasn't working?
Just to be clear, I assume we're talking about the second order Peano axioms here?
Heh - I'm amazed at how many things in this post I alternately strongly agree or strongly disagree with.
It’s important to distinguish between the numeral “2″, which is a formal symbol designed to be manipulated according to formal rules, and the noun “two”, which appears to name something
OK... I honestly can't comprehend how someone could simultaneously believe that "2" is just a symbol and that "two" is a blob of pure meaning. It suggests the inferential distances are actually pretty great here despite a lot of surface similarities between what Landsburg is saying and what I would say.
Instead, the point of the Peano axioms was to model what mathematicians do when they’re talking about numbers
I'm happy with this (although the next sentence suggests I may have been misinterpreting it slightly). Another way I would put it is that the Peano axioms are about making things precise and getting everyone to agree to the same rules so that they argue fairly.
Like all good models, the Peano axioms are a simplification that captures important aspects of reality without attempting to reproduce reality in detail.
I'd like an example here.
I’d probably start with something like Bertrand Russell’s account of numbers: We say that two sets of objects are “equinumerous” if they can be placed in one-one correspondence with each other
This is defining numbers in terms of sets, which he explicitly criticizes Yudkowsky for later. I'll take the charitable interpretation though, which would be "Y thinks he can somehow avoid defining numbers in terms of sets... which it turns out he can't... so if you're going to do that anyway you may as well take the more straightforward Russell approach"
the whole point of logic is that it is a mechanical system for deriving inferences from assumptions, based on the forms of sentences without any reference to their meanings
I really like this definition of logic! It doesn't seem to be completely standard though, and Yudkowksy is using it in the more general sense of valid reasoning. So this point is basically about the semantics of the word "logic".
But Yudkowsky is trying to derive meaning from the operation of inference, which won’t work because in second-order logic, meaning comes first.
I think this is a slight strawmanning of Yudkowsky. Here Yudkowsky is trying to define the meaning of one particular system - the natural numbers - not define the entire concept of meaning.
he’s effectively resorted to taking sets as primitive objects
So when studying logic, model theory etc. we have to make a distinction between the system of reasoning that we are studying and "meta-reasoning". Meta-reasoning can be done in natural language or in formal mathematics - generally I prefer when it's a bit more mathy because of the thing of precision/agreeing to play by the rules that I mentioned earlier. I don't see math and natural language as operating at different levels of abstraction though - clearly Landsburg disagrees (with the whole 2/two thing) but it's hard for me to understand why.
Anyway, if you're doing model theory then your meta-reasoning involves sets. If you use natural language instead though, you're probably still using sets at least by implication.
full-fledged Platonic acknowledgement
I think this is what using natural language as your meta-reasoning feels like from the inside. Landsburg would say that the natural numbers exist and that "two" refers to a particular one; in model theory this would be "there exists a model N with such-and-such properties, and we have a mapping from symbols "0", "1", "2" etc. to elements of N".
It seems enormouosly more plausible to me that numbes are “just out there” than that physical objects are “just out there”
So he's saying that the existence of numbers implies the existence of the physical universe, and therefore that the existence of numbers is more likely?? Or is there some quality of "just out there"ness that's distinct from existence?
I'm not quite sure what you're saying here - that "Numbers" don't exist as such but "the even naturals" do exist?
I assume that if a statistically significant number of people noticed that they were trying sympathetic magic and it was working, then the simulation would have to be restarted or tweaked since it could alter the history of the world in significant ways. So you might want to plan out that aspect of your strategy before collecting any data.