Posts

NunoSempere's Shortform 2020-03-22T19:58:54.830Z · score: 4 (1 votes)
Shapley Values Reloaded: Philantropic Coordination Theory & other miscellanea. 2020-03-10T17:36:54.114Z · score: 31 (12 votes)
A review of two books on survey-making 2020-03-01T19:11:13.828Z · score: 27 (14 votes)
A glowing review of two free online MIT Global Poverty courses 2020-01-15T11:40:41.519Z · score: 20 (14 votes)
[Part 1] Amplifying generalist research via forecasting – models of impact and challenges 2019-12-19T18:16:04.299Z · score: 53 (16 votes)
[Part 2] Amplifying generalist research via forecasting – results from a preliminary exploration 2019-12-19T16:36:10.564Z · score: 31 (11 votes)
Shapley values: Better than counterfactuals 2019-10-10T10:26:24.220Z · score: 82 (36 votes)
Why do social movements fail: Two concrete examples. 2019-10-04T19:56:02.028Z · score: 87 (43 votes)
EA Mental Health Survey: Results and Analysis. 2019-06-13T19:55:37.127Z · score: 46 (23 votes)

Comments

Comment by nunosempere on Effective Altruism and Free Riding · 2020-03-27T22:28:24.587Z · score: 4 (3 votes) · EA · GW

Interesting. Reminds me of this post by Paul Christiano on moral public goods

Comment by nunosempere on Are selection forces selecting for or against altruism? Will people in the future be more, as, or less altruistic? · 2020-03-27T17:55:45.198Z · score: 3 (3 votes) · EA · GW

The classical answer to this is that altruism towards strangers is not evolutionarily adaptative. This is because the altruistic give ressources benefit their own and others' descendants equally, while the nonaltruistic also get those benefits for their descendants without having to pay the cost. See also the tragic story of George R. Price.

Comment by nunosempere on AMA: Leah Edgerton, Executive Director of Animal Charity Evaluators · 2020-03-26T11:55:07.435Z · score: 1 (1 votes) · EA · GW

Either / Both

Comment by nunosempere on NunoSempere's Shortform · 2020-03-22T19:58:55.058Z · score: 2 (5 votes) · EA · GW

Testing shortform

Comment by nunosempere on AMA: "The Oxford Handbook of Social Movements" · 2020-03-22T12:13:22.819Z · score: 1 (1 votes) · EA · GW

Whatever happened to the Technology Assessment movement?

... In the early 70s, there was an academic Technology Assessment movement. They wanted to do detailed analysis of incoming technologies, and figure out how technological development could be planned, developed in a better order, and at a better rate. This is relevant not only to EAs who care about tech risks, but also to anyone who cares about tech and its impacts in general... (source)

Comment by nunosempere on AMA: Leah Edgerton, Executive Director of Animal Charity Evaluators · 2020-03-19T15:53:47.407Z · score: 2 (3 votes) · EA · GW

How useful is reducing animal farming for reducing pandemic risk?

Question taken from: here

Comment by nunosempere on AMA: Elie Hassenfeld, co-founder and CEO of GiveWell · 2020-03-19T15:51:36.889Z · score: 1 (1 votes) · EA · GW

Is there a point at which "maximizing total impact" and "doing the most cost-effective things" start being subtly different? How does it look like when that happens?

Comment by nunosempere on AMA: "The Oxford Handbook of Social Movements" · 2020-03-18T09:23:54.307Z · score: 1 (1 votes) · EA · GW

From this list of project ideas

What creates effective social movements?
Kerry Vaughan
"Social movements like effective altruism have the potential to unlock the abilities of large numbers of people by making it easier for them to coordinate and by providing social infrastructure to support altruistic activities. Yet, our understanding of what makes social movements effective or how to improve them is poor. Of particular interest is research into what makes social movements collapse and how to prevent the collapse of valuable movements. Conducting solid research and figuring out how to use its findings could be a multiplier of the EA movements as a whole."

Comment by nunosempere on AMA: "The Oxford Handbook of Social Movements" · 2020-03-18T07:14:04.678Z · score: 2 (2 votes) · EA · GW

What are the longest lasting social movements, and do they have any shared characteristics? Similarly. what is the half-time of a social movement, or the average duration of one?

Comment by nunosempere on AMA: "The Oxford Handbook of Social Movements" · 2020-03-18T07:13:08.401Z · score: 8 (6 votes) · EA · GW

What are the top 10 social movements nearest in think-space to EA, and did they succeed?

Comment by nunosempere on AMA: "The Oxford Handbook of Social Movements" · 2020-03-18T07:12:31.828Z · score: 4 (3 votes) · EA · GW

Is there an example of a movement which doesn't give its members things to do, yet succeeds? / What do these movements do with people?

Comment by nunosempere on AMA: Leah Edgerton, Executive Director of Animal Charity Evaluators · 2020-03-17T21:34:02.090Z · score: 11 (4 votes) · EA · GW

What are some questions regarding EAA which are amenable to being forecasted?

Comment by nunosempere on AMA: Leah Edgerton, Executive Director of Animal Charity Evaluators · 2020-03-17T21:32:10.533Z · score: 11 (6 votes) · EA · GW

Is there a piece of technology which would make your work significantly easier, but which doesn't exist yet?

Comment by nunosempere on AMA: Leah Edgerton, Executive Director of Animal Charity Evaluators · 2020-03-17T21:30:21.637Z · score: 2 (3 votes) · EA · GW

In which ways, if any, are the ethical stances which you take counterintuitive to you? Are there ways in which you expect your ethical stances to be counterintuitive to others, and any ways you adjust for that?

Comment by nunosempere on AMA: Toby Ord, author of "The Precipice" and co-founder of the EA movement · 2020-03-17T21:21:31.982Z · score: 7 (2 votes) · EA · GW

What's up with Pascal's Mugging? Why hasn't this pesky problem just been authoritatively solved? (and if it has, what's the solution?) What is your preferred answer? / Which bullets do you bite (e.g., bounded utility function, assigning probability 0 to events, a decision-theoretical approach cop-out, etc.)?

Comment by nunosempere on AMA: Toby Ord, author of "The Precipice" and co-founder of the EA movement · 2020-03-17T21:21:04.888Z · score: 8 (6 votes) · EA · GW

This is a genuine question. The framing is that if Toby Ord wants to get in touch with a high ranking member of government, get an article published in a prominent newspaper, direct a large number of man hours to a project he finds worthy, etc. he probably can; just the association to Oxford will open doors in many cases.

This is in opposition to a box in a basement which produces the same research he would, and some of these differences stem from him being endorsed by some prestigious organizations, and there being some social common knowledge around his person. The words "public intellectual" come to mind.

I'm wondering how the powers-of-being-different-from-a-box-which-produces-research will pan out.

Comment by nunosempere on AMA: Toby Ord, author of "The Precipice" and co-founder of the EA movement · 2020-03-17T21:00:50.467Z · score: 5 (4 votes) · EA · GW

I like how you operationalized the second question.

Comment by nunosempere on AMA: Toby Ord, author of "The Precipice" and co-founder of the EA movement · 2020-03-17T16:04:57.702Z · score: 6 (5 votes) · EA · GW

Suppose your life's work ended up having negative impact. What is the most likely scenario under which this could happen?

Comment by nunosempere on AMA: Toby Ord, author of "The Precipice" and co-founder of the EA movement · 2020-03-17T08:27:59.098Z · score: 6 (5 votes) · EA · GW

As a sharp mind, respected scholar, or prominent member in the EA community, you have a certain degree of agency, an ability to start new projects and make things happen, a no small amount of oomph and mojo. How are you planning to use this agency in the coming decades?

Comment by nunosempere on Open Thread #46 · 2020-03-13T10:59:53.421Z · score: 4 (3 votes) · EA · GW

EA Mental health using SlateStarCodex data.

Running some analysis on the SlateStarCodex Survey results of this year. EAs are less mentally ill than non-EAs, who are less mentally ill than respondents who identify as "sort of" EA. From this we should conclude that all three groups are all basically the same, because the difference is not significant at all.

Link to original comment & code, which I think nobody saw: Here.

Comment by nunosempere on Launching Utilitarianism.net: An Introductory Online Textbook on Utilitarianism · 2020-03-10T10:46:32.052Z · score: 1 (8 votes) · EA · GW

we, in the modern era, may also be unknowingly guilty of ...

Maybe substitute "guilty" for "responsible"?

Utilitarianism cares not only about the wellbeing of humans, but also about the wellbeing of non-human animals. Consequently, utilitarianism rejects speciesism, a form of discrimination against those who do not belong to a certain species.

There is a part of me which dislikes you presenting utilitarism which includes animals as the standard form of utilitarism. I think that utilitarianism + nonspeciesm falls under the "right but not trivial" category, and that a lot of legwork has to be done before you can get people to accept it, and further that this legwork must be done, instead of sliding over the inferential distance. Because of this, I'd prefer you to disambiguate between versions of utilitarianism which aggregate over humans, and those who aggregate over all sentient/conscious beings, and maybe point out how this developed over time (i.e., Peter Singer had to come and make the argument forcefully, because before it was not obvious)? For example, the Wikipedia entry on utilitarianism has a whole section on Humans alone, or other sentient beings?.

Similarly, maybe you would also want to disambiguate a little bit more between effective altruism and utilitarianism, and explicitly mention it when you're linking it to effective altruism websites, or use effective altruism examples?

Also, what's up with attributing the veil of ignorance to Harsanyi but not mentioning Rawls?

The section on Multi-level Utilitarianism Versus Single-level Utilitarianism seems exceedingly strange. In particular, you can totally use utilitarianism as a decision procedure (and if you don't, what's the point?). The fact that you don't have the processing power of a supercomputer and perfect information doesn't mean that you can't approximate it as best you can.

For example, if I buy eggs which come from less shitty farmers, or if I decide to not buy eggs in order to reduce factory farming, I'm using utilitarianism as a decision procedure. Even though I can't discern the exact effects of the action, I can discern that the action has positive expected value.

I don't fall into recursive loops trying to compute how much compute I should use to compute the expected value of an action because I'm not an easily disabled robot in a film. But I do sometimes go up several levels of recursion, depending on the importance of the decision. I use heuristics like I use low degree Taylor polynomials.

(I also don't always instantiate utilitarianism. But when I do, I do use it as a decision procedure)

In contrast, to our knowledge no one has ever defended single-level utilitarianism [i.e., that utilitarianism should be a decision procedure] You know what, I defend single-level utilitarianism as a straightforward application of utilitarianism + bounded computing power / bounded rationality, and have the strong intuition that if utilitarianism isn't a decision rule, then there's no point to it. Fight me. (But also, feel free not to if you calculate that you have better things to do).

A common objection to multi-level utilitarianism is that it is self-effacing. A theory is said to be (partially) self-effacing if it (sometimes) directs its adherents to follow a different theory. Multi-level utilitarianism often forbids using the utilitarian criterion when we make decisions, instead recommending to act in accordance with non-utilitarian heuristics. However, there is nothing inconsistent about saying that your criterion of moral rightness comes apart from the decision procedure it recommends, and it does not mean that the theory fails.

I have different intuitions which strongly go in the other direction.

Comment by nunosempere on Epistea Workshop Series: Epistemics Workshop, May 2020, UK · 2020-02-28T22:31:57.829Z · score: 2 (2 votes) · EA · GW

Having attended a previous workshop, I'd wager that this iteration would likely be interesting, and the team, which seems to have remained constant since then, is brilliant.

They might, however, want to write all the information from the workshop webpage directly into this post, which right now is perhaps too much on the terse side.

Comment by nunosempere on AI Impacts: Historic trends in technological progress · 2020-02-12T15:38:58.342Z · score: 5 (4 votes) · EA · GW

I've been following this with interest.

Re: Telecommunications performance, the red telephone might also be a a discontinuity in practical terms.

The 1962 Cuban Missile Crisis made the hotline a priority. During the standoff, official diplomatic messages typically took six hours to deliver; unofficial channels, such as via television network correspondents, had to be used too as they were quicker

That is, even though faster systems existed, they hadn't been implemented in the area of communications between the Soviet Union and the USA (pretty huge blindspot), but could be implemented more or less immediately, once both regimes actually bothered.

Also of interest to readers might be: some other discontinuities, one in passenger ship length and the other one on time needed to circumnavigate the Earth. AI impacts also has a couple of other discontinuities on their webpage, not mentioned/linked above:

Comment by nunosempere on EA Mental Health Survey: Results and Analysis. · 2020-01-31T08:52:40.433Z · score: 2 (2 votes) · EA · GW

Running the results again on the SlateStarCodex Survey results of this year. EAs are less mentally ill than non-EAs, who are less mentally ill than respondents who identify as "sort of" EA. From this we should conclude that all three groups are all basically the same, because the difference is not significant at all.

The code used to find that out is, in R:

setwd("my/directory") ## directory in which the 2020ssc_public.xlsx file resides, downloadable from https://slatestarcodex.com/2020/01/20/ssc-survey-results-2020/
install.packages("openxlsx", dependencies = TRUE)
library(openxlsx)

D <- read.xlsx("2020ssc_public.xlsx")

mentally_ill_strict =  c("I have a formal diagnosis of this condition")
mentally_ill_loose =  c("I have a formal diagnosis of this condition", "I think I might have this condition, although I have never been formally diagnosed")

m <- mentally_ill_strict
D$MentallyIll_1 = D$Depression %in% m | D$Anxiety %in% m | D$OCD %in% m  | D$Eatingdisorder %in% m  | D$PTSD %in% m  | D$Alcoholism %in% m  | D$Drugaddiction %in% m  | D$Borderline %in% m  | D$Bipolar %in% m

m <- mentally_ill_loose
D$MentallyIll_2 = D$Depression %in% m | D$Anxiety %in% m | D$OCD %in% m  | D$Eatingdisorder %in% m  | D$PTSD %in% m  | D$Alcoholism %in% m  | D$Drugaddiction %in% m  | D$Borderline %in% m  | D$Bipolar %in% m

summary(lm(D$MentallyIll_1 ~ D$EAID))
summary(glm(D$MentallyIll_1 ~ D$EAID, family = "binomial")) ## This is a logistic regression

summary(lm(D$MentallyIll_1 ~ D$EAID))
summary(glm(D$MentallyIll_1 ~ D$EAID, family = "binomial")) ## This is a logistic regression

And the results are

Call:
lm(formula = D$MentallyIll_1 ~ D$EAID)

Residuals:
    Min      1Q  Median      3Q     Max 
-0.5256 -0.5151  0.4744  0.4849  0.5325 

Coefficients:
            Estimate Std. Error t value Pr(>|t|)    
(Intercept)  0.46753    0.04027  11.609   <2e-16 ***
D$EAIDNo     0.04757    0.04109   1.158    0.247    
D$EAIDSorta  0.05811    0.04154   1.399    0.162    
D$EAIDYes    0.03499    0.04328   0.808    0.419    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.4998 on 7335 degrees of freedom
Multiple R-squared:  0.0004201,	Adjusted R-squared:  1.124e-05 
F-statistic: 1.027 on 3 and 7335 DF,  p-value: 0.3792

> summary(glm(D$MentallyIll_1 ~ D$EAID, family = "binomial")) ## This is a logistic regression

Call:
glm(formula = D$MentallyIll_1 ~ D$EAID, family = "binomial")

Deviance Residuals: 
   Min      1Q  Median      3Q     Max  
-1.221  -1.203   1.134   1.152   1.233  

Coefficients:
            Estimate Std. Error z value Pr(>|z|)
(Intercept)  -0.1301     0.1615  -0.805    0.421
D$EAIDNo      0.1905     0.1648   1.156    0.248
D$EAIDSorta   0.2327     0.1666   1.397    0.162
D$EAIDYes     0.1401     0.1735   0.807    0.419

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 10167  on 7338  degrees of freedom
Residual deviance: 10164  on 7335  degrees of freedom
AIC: 10172

Number of Fisher Scoring iterations: 3

> summary(lm(D$MentallyIll_1 ~ D$EAID))

Call:
lm(formula = D$MentallyIll_1 ~ D$EAID)

Residuals:
    Min      1Q  Median      3Q     Max 
-0.5256 -0.5151  0.4744  0.4849  0.5325 

Coefficients:
            Estimate Std. Error t value Pr(>|t|)    
(Intercept)  0.46753    0.04027  11.609   <2e-16 ***
D$EAIDNo     0.04757    0.04109   1.158    0.247    
D$EAIDSorta  0.05811    0.04154   1.399    0.162    
D$EAIDYes    0.03499    0.04328   0.808    0.419    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.4998 on 7335 degrees of freedom
Multiple R-squared:  0.0004201,	Adjusted R-squared:  1.124e-05 
F-statistic: 1.027 on 3 and 7335 DF,  p-value: 0.3792

> summary(glm(D$MentallyIll_1 ~ D$EAID, family = "binomial")) ## This is a logistic regression

Call:
glm(formula = D$MentallyIll_1 ~ D$EAID, family = "binomial")

Deviance Residuals: 
   Min      1Q  Median      3Q     Max  
-1.221  -1.203   1.134   1.152   1.233  

Coefficients:
            Estimate Std. Error z value Pr(>|z|)
(Intercept)  -0.1301     0.1615  -0.805    0.421
D$EAIDNo      0.1905     0.1648   1.156    0.248
D$EAIDSorta   0.2327     0.1666   1.397    0.162
D$EAIDYes     0.1401     0.1735   0.807    0.419

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 10167  on 7338  degrees of freedom
Residual deviance: 10164  on 7335  degrees of freedom
AIC: 10172

Number of Fisher Scoring iterations: 3

Comment by nunosempere on The EA Hotel is now the Centre for Enabling EA Learning & Research (CEEALAR) · 2020-01-30T10:04:56.136Z · score: 5 (4 votes) · EA · GW

Center for Assisting EA Study And Research: CAESAR.

Assisting ~ Aiding ~ Abetting

Comment by nunosempere on AMA: Rob Mather, founder and CEO of the Against Malaria Foundation · 2020-01-23T14:08:28.353Z · score: 3 (2 votes) · EA · GW

I started somewhat by accident...

Would you give a guess as to the probability you would have assigned to ending up doing something like this before you in fact did? How unlikely was this class of outcomes?

Comment by nunosempere on EA interest: Google Trends [fixed] · 2020-01-17T10:06:48.281Z · score: 14 (12 votes) · EA · GW

[The images don't load for me. What I usually do in these cases is to upload the image to github, and then add it to the EA forum with ![](imageURL)]

Comment by nunosempere on Why do social movements fail: Two concrete examples. · 2019-12-19T21:22:53.660Z · score: 4 (3 votes) · EA · GW

Curiously enough, 'twas me who fulfilled the bounty.

Comment by nunosempere on Opinion: Estimating Invertebrate Sentience · 2019-12-13T10:56:01.009Z · score: 13 (6 votes) · EA · GW

You could also consider providing your probabilities in the form of distributions, for example by answering a question like "What probability would you assign to tribbles [1] being sentient after 1000 more hours of research?" This would perhaps solve part of the problem of communicating the uncertainty which you want to communicate.

Some example answers might be:

In the first one your uncertainty is significant. You have a probability of ~45%, but you consider it likely that you will update a lot. You just don't know in which direction: You consider it equally likely that you will end up at 30% or at 70%.

In the second one, research has mostly been done and you've already mostly made up your mind. You have a probability of ~45%, and you believe that further research is most likely to move you to 42%, or to 47%, but not that much further. You'd be shocked if you ended up with a probability of more than 60%.

The third one is the distributional equivalent of you making a shrug. Your probability is 45%, but really, tribbles being so underexplored means that your distribution looks pretty much uniform.

Note that it's even possible that all three distributions have the same mean (~45%). This would mean that in all three cases you'd think that a bet of 45:55 would be a fair bet (that is, that it has an expected value of 0).

You could also predict the standard deviation of your distribution (how broad it is) after 1000h of research, 2000h, 5000h, etc., aggregate the distributions of all your researchers, and do other nice things.

[1] Fictional species.

Comment by nunosempere on EA Leaders Forum: Survey on EA priorities (data and analysis) · 2019-12-09T14:00:22.070Z · score: 4 (3 votes) · EA · GW

Hey, regarding your question "What types of talent do you currently think [your organization // EA as a whole] will need more of over the next 5 years? (Pick up to 6)", I think you might want to word it somewhat differently, and perhaps disambiguate between:

  • On a scale 0-10, how much [X] talent will [EA/your organization] need over the next 5 years?".
  • On a scale 0-10, how much will the need for [X] talent in [EA/your organization] increase over the next 5 years?".

This mainly has the advantage of allowing for more granular comparison. For example, maybe management is always a solid 10, whereas government expertise is a solid 7, but both always fall in the top 6, and the difference might sometimes be important. I also think that the second wording is somewhat easier to read / requires less cognitive labor.

Two book recommendations are The Power of Survey Design: A User's Guide for Managing Surveys, Interpreting Results, and Influencing Respondents and Improving survey questions - Design and Evaluation. I should have a short review/summary and some checklists somewhere, if you're interested.

Comment by nunosempere on How do we create international supply chain accountability? · 2019-12-02T09:30:52.785Z · score: 1 (1 votes) · EA · GW

I researched this at some point; will try to write something up this week, though I might do so in a separate post. Thanks for asking the question!

Comment by nunosempere on How do we create international supply chain accountability? · 2019-12-02T09:27:52.822Z · score: 1 (1 votes) · EA · GW

I disagree; I see "supply chain accountability" as a very specific, almost technical term (the word international is a little bit redundant, though), which contains enough information for me to figure out what it means.

Comment by nunosempere on I'm Buck Shlegeris, I do research and outreach at MIRI, AMA · 2019-11-16T08:12:31.597Z · score: 19 (9 votes) · EA · GW

What would you be working on if you were working on something else?

Comment by nunosempere on I'm Buck Shlegeris, I do research and outreach at MIRI, AMA · 2019-11-16T08:12:16.235Z · score: 14 (8 votes) · EA · GW

Do you have the intuition that a random gifted person can contribute to technical research on AI safety?

Comment by nunosempere on Shapley values: Better than counterfactuals · 2019-10-31T14:06:21.389Z · score: 3 (2 votes) · EA · GW

Hey Chris! It was nice seeing you at the EA Hotel, and I'm glad we could talk about this. I'm writing down some of my notes from our conversations. Is there anything I've forgotten, or which you'd like to add?

a. What are you using Shapley values / counterfactual values for?

You might want to use different tools depending on what your goal is; three different goals migh be: Coordination / Analysis / Reward / Award.

For example, you might want a function which is easier to understand when announcing an award. If you're rewarding a behavior, you might want to make sure you're incentivizing the right thing.

b. The problem of choosing who to count is more complicated than I originally thought, and you should in fact exclude some agents from your calculations.

The example of: "If a bus driver falls off a cliff and Superman rescues them and brings them safely to their destination, earlier, the bus driver gets half the credit" is silly, but made the thing really crisp for me.

Hearing that, we then thought that:

  • Yes, the driver gets half the credit under Shapley values, but the same value as Superman under counterfactual value.
  • (also, if the driver distracts Superman from saving a different bus, then the driver gets 0 or negative value in both cases)
  • (if the driver was intelligent enough to know that Superman wasn't doing anything important, he might actually get half the credit, but only of getting there earlier. In this scenario, had there been no Superman, the driver wouldn't have fallen off the cliff.).
  • (if the driver was a paperclip maximizer who didn't know that Superman was going to be around, then Superman should take all the credit).

So the answer would seem to be something like: -Counting only over people who are broadly similar to you?

  • Who are optimizing over the same thing, or whose decisions can be changed because of yours? It seems like this is more of a case of causal, rather than subjunctive dependence.

c. Shapley values and uncertainty

How do SVs deal with uncertainty? Can you do expected value over SVs? [Yes, you can]. For example, if you have a 1% chance of a SV of 100, you can say that the E[SV] = 1. Even thought the SV formalism is more complicated than the counterfactual, it still works elegantly / is well-defined, etc.

Comment by nunosempere on Attempt at understanding the role of moral philosophy in moral progress · 2019-10-28T23:27:38.042Z · score: 2 (2 votes) · EA · GW

Regarding "to what extent have individuals influenced history?", I have found Great Founder Theory, valuable, but it doesn't answer that exact question.

Comment by nunosempere on Shapley values: Better than counterfactuals · 2019-10-15T07:37:56.339Z · score: 2 (2 votes) · EA · GW
  1. would surprise me; can you think of a source?
Comment by nunosempere on Shapley values: Better than counterfactuals · 2019-10-11T10:15:08.578Z · score: 2 (2 votes) · EA · GW

Fair point re: uncertainty. The situation seems pretty symmetric, though: if a politician builds roads just to get votes, and an NGO steps in and does something valuable with that, the politician's counterfactual impact is still the same as the NGO's, so both the Shapley value and counterfactuals have that problem (?). Maybe one can exclude agents acording to how close their goals are to yours, e.g., totally exclude a paperclip maximizer from both counterfactual and Shapley value calculations, and apply order indifference to allies only (?). This is something I haven't though about; thanks for pointing it out.

Fair point re: epistemic status. Changed my epistemic status.

Comment by nunosempere on Shapley values: Better than counterfactuals · 2019-10-11T10:08:26.209Z · score: 1 (1 votes) · EA · GW

I don't exactly claiming to have identified a problem with the counterfactual function, in itself. The counterfactual is perfectly well defined, and I like it, and it has done nothing wrong. I understand this. It is clear to me that it can't be added just like that. The function, per se, is fine.

What I'm claiming is that, because it can't be aggregated, it is not the right function to think about in terms of assigning impact to people in the context of groups. I am arguing about the area of applicability of the function, not about the function. I am claiming that, if you are optimizing for counterfactual impact in terms of groups, pitfalls may arise.

It's like, when you first see for the same time: -1 = sqrt(-1)*sqrt(-1) = sqrt((-1)*(-1)) = sqrt(1) = 1, therefore -1 = 1, and you can't see the mistake. It's not that the sqrt function is wrong, it's that you're using it outside it's limited fiefdom, so something breaks. I hope the example proved amusing.

I'm not only making statements about the counterfactual function, I'm also making statements about the concept which people have in your head which is called "impact", and how that concept doesn't map to counterfactual impact some of the time, and about how, if you had to map that concept to a mathematical function, the Shapley value is a better candidate.

Comment by nunosempere on Shapley values: Better than counterfactuals · 2019-10-11T09:48:17.198Z · score: 4 (3 votes) · EA · GW

1.

I have thought about this, and I'm actually biting the bullet. I think that a lot of people get impact for a lot of things, and that even smallish projects depend on a lot of other moving parts, in the direction of You didn't build that.

I don't agree with some of your examples when taken literally, but I agree with the nuanced thing you're pointing at with them, e.g., building good roads seems very valuable precisely because it helps other projects, if there is high nurse absenteeism then the nurses who show up take some of the impact...

I think that if you divide the thing's impact by, say 10x, the ordering of the things according to impact remains, so this shouldn't dissuade people from doing high impact things. The interesting thing is that some divisors will be greater than others, and thus the ordering will be changed. I claim that this says something interesting.

2.

Not really. If 10 people have already done it, your Shapley value will be positive if you take that bargain. If the thing hasn't been done yet, you can't convince 10 Shapley-optimizing altruists to do the thing for 0.5m each, but you might convince 10 counterfactual impact optimizers. As @casebach mentioned, this may have problems when dealing with uncertainty (for example: what if you're pretty sure that someone is going to do it?).

3.

You're right. The example, however, specified that the EAs were to be "otherwise idle", to simplify calculations.

Comment by nunosempere on Shapley values: Better than counterfactuals · 2019-10-11T08:42:48.260Z · score: 2 (2 votes) · EA · GW

In my mind, that gets a complexity penalty. Imagine that instead of ten people, there were 10^10 people. Then for that hack to work, and for everyone to be able to say that they convinced all the others, there has to be some overhead, which I think that the Shapley value doesn't require.

Comment by nunosempere on Shapley values: Better than counterfactuals · 2019-10-11T07:23:40.149Z · score: 2 (2 votes) · EA · GW

Good point!

Comment by nunosempere on Shapley values: Better than counterfactuals · 2019-10-10T17:10:31.941Z · score: 4 (3 votes) · EA · GW

What you say seems similar to a Stag hunt. Consider, though, that if the group is optimizing for their individual counterfactual impact, they'll want to coordinate to all do the 100 utility project. If they were optimizing their Shapley value, they'd instead want to coordinate to do 10 different projects, each worth 20 utility. 20*10 = 200 >100.

Comment by nunosempere on Why do social movements fail: Two concrete examples. · 2019-10-09T20:09:45.284Z · score: 2 (2 votes) · EA · GW

What concrete examples were you thinking of?

Comment by nunosempere on Why do social movements fail: Two concrete examples. · 2019-10-09T20:06:44.291Z · score: 7 (3 votes) · EA · GW

Regarding Spanish Enlightenment, I can't answer as decisively, because the sources I used were in Spanish, and they were combined with me just knowing things about Spanish literature and history, which made hypothesis generation much easier and much faster.

That being said, the English Wikipedia page Enlightenment in Spain might be a good starting point.

Comment by nunosempere on Why do social movements fail: Two concrete examples. · 2019-10-09T19:48:12.017Z · score: 22 (6 votes) · EA · GW

Here is, additionally, a list of behaviors/techniques recommended by General Semantics which stood out to me for some reason. The problem is, though, that I find it difficult to say whether they're representative; for that, see the first link in my other comment: An overview of general semantics . With that in mind:

Extensional devices:

  • Indexing : Muslim(1) is not Muslim(2); Feminist(1) is not Feminist(2);. Remember to look for the differences even among a group or category that presume similarities.
  • Dating : Steve(2008) is not Steve(1968); Steve’s-views-on-abortion(2008) are not Steve’s-views-on-abortion(1988). Remember that each person and each ‘thing’ we experience changes over time, even though the changes may not be apparent to us.
  • Quotes : ‘truth’, ‘reality’, ‘mind’, ‘elite’. Use quotes around terms as a caution to indicate you’re aware that there is an opportunity for misunderstanding if the term is particularly subject to interpretation, or if you’re being sarcastic, ironic, or facetious. o hyphen : mind-body, thinking-feeling. Use to join terms that we can separate in language, but can’t actually separate in the ‘real’ world. Remember that we can talk in terms that don’t accurately reflect the world ‘out there.’
  • etc.: Remember that our knowledge and awareness of anything is limited. We can’t sense or experience or talk about all of something, so we should maintain an awareness that “more could be said.”

Variations of English:

  • E-Prime: eliminate or reduce forms of the to be verbs (is, are, were, am, being, etc.). In particular, reduce those that we consider is of identity (ex. John is a liberal) and is of predication (ex. The rose is red.)
    • “What’s this? What’s that?” Don’t answer, “It’s a table,” but, “We call it a table.”
  • English Minus Absolutisms (EMA): eliminate or reduce inappropriate generalizations or expressions that imply allness or absolute attitudes. Examples include: all, none, every, totally, absolutely, perfect, without a doubt, certain, completely.

Holding a stone

Bruce Kodish led the sessions dealing with experiencing on the silent level. One exercise was seemingly quite simple. We were told to pick out a stone, bring it to class, then for a few minutes simply experience the stone on the silent level. In other words, to use our senses without verbalizing our reactions to our senses. My inability to accomplish this simple task was enlightening. It emphasized to me how language can get in the way of our moment-to-moment experiences with “what is going on.” It also demonstrated the extent to which I generate meanings for things. While I was unsuccessful in shutting off my verbalizing, I was quite proficient in coming up with all kinds of thoughts-and-feelings-and-meanings about an ordinary, arbitrary rock. If I can ‘make up’ so much meaning for a random inanimate object, perhaps it would be appropriate for me to be hesitant and inquiring in my future evaluations of relationships with more animate beings.

Ladders

General Semantics has several ladders, which illustrate different levels of abstraction. For example:

A)

  1. Something is going on
  2. I experience what’s going on
  3. I evaluate my experience of what’s going on
  4. From my evaluation of my experience of what’s going on, I respond to and give meaning to what is going.

Example: You misunderstood what I was trying to say / You didn't write clearly enough benefits from that.

B)

What Happens ≠ What I Sense ≠ How I Respond ≠ “What It Means”

C)

What we sense is not what happened - What we describe is not what we sense - What it means is not what we describe.

D)

E)

Here is an example of these ladders being used:

What this GS stuff meant to me, at that particular time, was that I didn’t have to be consumed with guilt over the fact that I had decided to end my marriage. Divorce didn’t have a predetermined meaning — our daughter wasn’t forever doomed to be neglected and miserable; I didn’t have to walk forever with my head bowed, ashamed of taking actions to further my own personal happiness; my wife didn’t have to forever grieve over what I had ‘done’ to her. It was certainly possible that each of these outcomes could occur, but they were not unavoidable consequences of the event called divorce. Source: Here is something about general semantics, by Steven Stockdale, who was once director of the Insititute of Semantics.

Note that CBT says something similar

Comment by nunosempere on Why do social movements fail: Two concrete examples. · 2019-10-09T19:45:48.112Z · score: 9 (4 votes) · EA · GW

Fortunately, I keep notes, so here is a list of links with respect to general semantics which kind of answer your questions.

An overview of general semantics

Reflections by a general semanticist
History of general semantics
A Brief History of General Semantics
A Brief History of General Semantics II
Drama because of having two organizations
A book length introduction
The Wikipedia page

Comment by nunosempere on Editing available for EA Forum drafts · 2019-07-24T19:25:32.838Z · score: 1 (3 votes) · EA · GW

As a datapoint, this would have been highly useful to me before writing this: EA Mental Health Survey: Results and Analysis. Can you have a look at it regardless?

Interestingly, regardless of when you make the offer publicly, there is someone who will have published something at t-1, so I don't feel too bad.

Comment by nunosempere on How Europe might matter for AI governance · 2019-07-14T16:23:30.885Z · score: 13 (6 votes) · EA · GW

You can post an image using standard markdown syntax:

![](link to the image)

For example, to insert the above image, I wrote:
![](https://nunosempere.github.io/ea/AI-Europe.png)
Comment by nunosempere on EA Mental Health Survey: Results and Analysis. · 2019-07-14T12:08:39.067Z · score: 4 (3 votes) · EA · GW

[this comment is also archived in my blog, here without indentation and thus easier to read]

Re: @Peter_Hurford. The 2019 SSC Survey does have an EA_ID question. Using that:

If you run some regressions, you get a significant correlation between EA affiliation and mental conditions; respondents who identified as EA differed from non-EAs by ~2-4% (see below). Note that the SSC Survey is subject to fewer biases than the EA Mental Health survey, and also note that it's still difficult to extract causal conclusions. Data available here

Plots:

Diagnosed + Intuited

                                                                            x   y         %
1                                                                      EA Yes 959 100.00000
2         Has been diagnosed with a mental condition, or thinks they have one 580  60.47967
3 Has not been diagnosed with a mental condition, and does not think they any 347  36.18352
4                                                          NA / Didn't answer 125  13.03441
                                                                            x    y          %
1                                                                    EA Sorta 2223 100.000000
2         Has been diagnosed with a mental condition, or thinks they have one 1354  60.908682
3 Has not been diagnosed with a mental condition, and does not think they any  795  35.762483
4                                                          NA / Didn't answer  167   7.512371
                                                                            x    y          %
1                                                                       EA No 4158 100.000000
2         Has been diagnosed with a mental condition, or thinks they have one 2416  58.104858
3 Has not been diagnosed with a mental condition, and does not think they any 1587  38.167388
4                                                          NA / Didn't answer  248   5.964406

Diagnosed

                                               x   y         %
1                                         EA Yes 959 100.00000
2     Has been diagnosed with a mental condition 314  32.74244
3 Has not been diagnosed with a mental condition 613  63.92075
4                             NA / Didn't answer 125  13.03441
                                               x    y          %
1                                       EA Sorta 2223 100.000000
2     Has been diagnosed with a mental condition  718  32.298695
3 Has not been diagnosed with a mental condition 1431  64.372470
4                             NA / Didn't answer  167   7.512371
                                               x    y          %
1                                          EA No 4158 100.000000
2     Has been diagnosed with a mental condition 1183  28.451178
3 Has not been diagnosed with a mental condition 2820  67.821068
4                             NA / Didn't answer  248   5.964406

Regressions

Linear

> # D$mentally_ill = Number of diagnosed mental ilnesses
> # D$mentally_ill2= Number of mental ilnesses, diagnosed + intuited
> summary(lm(D$mentally_ill ~ D$`EA ID`))

Call:
lm(formula = D$mentally_ill ~ D$`EA ID`)

Residuals:
    Min      1Q  Median      3Q     Max 
-0.5717 -0.5514 -0.4689  0.4486 10.4283 

Coefficients:
               Estimate Std. Error t value Pr(>|t|)    
(Intercept)     0.46890    0.01424  32.935  < 2e-16 ***
D$`EA ID`Sorta  0.08252    0.02409   3.426 0.000617 ***
D$`EA ID`Yes    0.10284    0.03283   3.132 0.001742 ** 
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.9008 on 7076 degrees of freedom
  (354 observations deleted due to missingness)
Multiple R-squared:  0.002421,	Adjusted R-squared:  0.002139 
F-statistic: 8.587 on 2 and 7076 DF,  p-value: 0.0001884
> summary(lm(D$mentally_ill2 ~ D$`EA ID`))

Call:
lm(formula = D$mentally_ill2 ~ D$`EA ID`)

Residuals:
    Min      1Q  Median      3Q     Max 
-1.3711 -1.2638 -0.2638  0.7362  9.6289 

Coefficients:
               Estimate Std. Error t value Pr(>|t|)    
(Intercept)     1.26380    0.02243  56.343   <2e-16 ***
D$`EA ID`Sorta  0.09637    0.03795   2.539   0.0111 *  
D$`EA ID`Yes    0.10729    0.05173   2.074   0.0381 *  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 1.419 on 7076 degrees of freedom
  (354 observations deleted due to missingness)
Multiple R-squared:  0.001216,	Adjusted R-squared:  0.0009338 
F-statistic: 4.308 on 2 and 7076 DF,  p-value: 0.0135
> summary(lm(D$mentally_ill>0 ~ D$`EA ID`))

Call:
lm(formula = D$mentally_ill > 0 ~ D$`EA ID`)

Residuals:
    Min      1Q  Median      3Q     Max 
-0.3387 -0.3341 -0.2955  0.6659  0.7045 

Coefficients:
               Estimate Std. Error t value Pr(>|t|)    
(Intercept)    0.295528   0.007323  40.354  < 2e-16 ***
D$`EA ID`Sorta 0.038581   0.012391   3.114  0.00186 ** 
D$`EA ID`Yes   0.043199   0.016889   2.558  0.01055 *  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.4633 on 7076 degrees of freedom
  (354 observations deleted due to missingness)
Multiple R-squared:  0.001835,	Adjusted R-squared:  0.001553 
F-statistic: 6.505 on 2 and 7076 DF,  p-value: 0.001505
> summary(lm(D$mentally_ill2>0 ~ D$`EA ID`))

Call:
lm(formula = D$mentally_ill2 > 0 ~ D$`EA ID`)

Residuals:
    Min      1Q  Median      3Q     Max 
-0.6301 -0.6036  0.3699  0.3965  0.3965 

Coefficients:
               Estimate Std. Error t value Pr(>|t|)    
(Intercept)    0.603547   0.007692  78.466   <2e-16 ***
D$`EA ID`Sorta 0.026513   0.013014   2.037   0.0417 *  
D$`EA ID`Yes   0.022127   0.017738   1.247   0.2123    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.4867 on 7076 degrees of freedom
  (354 observations deleted due to missingness)
Multiple R-squared:  0.0006657,	Adjusted R-squared:  0.0003832 
F-statistic: 2.357 on 2 and 7076 DF,  p-value: 0.09481

Logistic

> summary(glm(D$mentally_ill>0 ~ D$`EA ID`, family=binomial(link='logit')))

Call:
glm(formula = D$mentally_ill > 0 ~ D$`EA ID`, family = binomial(link = "logit"))

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-0.9095  -0.9018  -0.8370   1.4807   1.5614  

Coefficients:
               Estimate Std. Error z value Pr(>|z|)    
(Intercept)    -0.86868    0.03464 -25.078  < 2e-16 ***
D$`EA ID`Sorta  0.17902    0.05737   3.120  0.00181 ** 
D$`EA ID`Yes    0.19971    0.07756   2.575  0.01003 *  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 8797.8  on 7078  degrees of freedom
Residual deviance: 8784.8  on 7076  degrees of freedom
  (354 observations deleted due to missingness)
AIC: 8790.8

Number of Fisher Scoring iterations: 4
> summary(glm(D$mentally_ill2>0 ~ D$`EA ID`, family=binomial(link='logit')))

Call:
glm(formula = D$mentally_ill2 > 0 ~ D$`EA ID`, family = binomial(link = "logit"))

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.4103  -1.3603   0.9612   1.0049   1.0049  

Coefficients:
               Estimate Std. Error z value Pr(>|z|)    
(Intercept)     0.42027    0.03231  13.007   <2e-16 ***
D$`EA ID`Sorta  0.11221    0.05514   2.035   0.0419 *  
D$`EA ID`Yes    0.09344    0.07517   1.243   0.2139    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 9439.1  on 7078  degrees of freedom
Residual deviance: 9434.4  on 7076  degrees of freedom
  (354 observations deleted due to missingness)
AIC: 9440.4

Number of Fisher Scoring iterations: 4