casebash

EA should wargame Coron­avirus

• casebash 12 Jan 2020 12:32 UTC

  2 points

  in reply to: SiebeRozendal's comment on: Fi­nal up­date on EA Nor­way’s Oper­a­tions Project

  What’s Good Growth?

• casebash 11 Jan 2020 20:33 UTC

  12 points

  in reply to: vaidehi_agarwalla's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  Yeah, EA is likely less com­pel­ling when this is defined as feel­ing mo­ti­vat­ing/​in­ter­est­ing to the av­er­age per­son at the mo­ment, al­though it is hard to judge since EA hasn’t been around for any­where near as long. Nonethe­less, many of the is­sues EAs care about seem way too weird for the av­er­age per­son, then again if you look at fem­i­nism, a lot of the ideas were only ever pre­sent in an overly aca­demic form. Part of the rea­son why they are so in­fluen­tial now is that they have filtered down into the gen­eral pop­u­la­tion in a sim­pler form (such as “girl power”, “feel­ing good, ra­tio­nal­ity bad”). Plus so­cial jus­tice is more likely to benefit the peo­ple sup­port­ing it in the here and now than EA which fo­cuses more on other coun­tries, other species and other times which is always a tough sell.

  SJ is an ex­tremely in­clu­sive move­ment (ba­si­cally by defi­ni­tion)

  I’m gen­er­ally wary of ar­gu­ment by defi­ni­tion. In­deed, SJ is very in­clu­sive to mem­bers of a racial minor­ity or those who are LGBTI, but is very much not when it comes to ide­olog­i­cal di­ver­sity. And some strands can be very un­wel­com­ing to mem­bers of ma­jori­ties. So it’s much more com­plex than that.

• casebash 11 Jan 2020 17:14 UTC

  6 points

  in reply to: Sean_o_h's comment on: I’m Cul­len O’Keefe, a Policy Re­searcher at OpenAI, AMA

  “There are definitely many who see these more in the move­ment/​tribe sense”—For mod­ern so­cial jus­tice this tends to fo­cus on who is a good or bad per­son, while for EA this tends to fo­cus more on who to trust. (There’s a less dom­i­nant strand of thought within so­cial jus­tice that says we shouldn’t blame in­di­vi­d­u­als for sys­tem­atic is­sues, but it’s rel­a­tively rare). EA makes some efforts to­wards be­ing anti-tribal, while so­cial jus­tice is less wor­ried about the down­sides of be­ing tribal.

• casebash 26 Nov 2019 10:46 UTC

  8 points

  in reply to: LarissaHeskethRowe's comment on: Up­dates from Lev­er­age Re­search: his­tory, mis­takes and new focus

  Greater knowl­edge of psy­chol­ogy would be pow­er­ful, but why should we ex­pect the sign to be pos­i­tive, in­stead of say mak­ing the world worse by im­prov­ing pro­pa­ganda and mar­ket­ing?

• casebash 22 Nov 2019 22:53 UTC

  24 points

  on: Up­dates from Lev­er­age Re­search: his­tory, mis­takes and new focus

  Why is Lev­er­age work­ing on psy­chol­ogy? What is it hop­ing to ac­com­plish?

• casebash 18 Nov 2019 0:50 UTC

  6 points

  on: “EA res­i­den­cies” as an out­reach activity

  This seems like a good idea and definitely the thing I’d con­sider once I learn enough about ai that this would be valuable for oth­ers.

• casebash 3 Nov 2019 8:34 UTC

  4 points

  on: Does 80,000 Hours fo­cus too much on AI risk?

  “It’s not clear that ad­vanced ar­tifi­cial in­tel­li­gence is go­ing to ar­rive any time within the next sev­eral decades”—On the other hand, it’s seems, at least to me, most likely that it will. Even if sev­eral more break­throughs would be re­quired to reach gen­eral in­tel­li­gence, those may still come rel­a­tively fast as deep learn­ing has now fi­nally be­come use­ful enough in a wide enough ar­ray of ap­pli­ca­tions that there is far more money and tal­ent in the field than there ever was be­fore by or­ders of mag­ni­tude. Now this by it­self wouldn’t nec­es­sar­ily guaran­tee fast ad­vance­ment in a field, but AI re­search is still the kind of area where a sin­gle in­di­vi­d­ual can push the re­search for­ward sig­nifi­cantly just by them­selves. And gov­ern­ments are be­gin­ning to re­al­ise the strate­gic im­por­tance of AI, so even more re­sources are flood­ing the field.

  “One of the top AI safety or­ga­ni­za­tions, MIRI, has now gone pri­vate so now we can’t even in­spect whether they are do­ing use­ful work.”—this is not an un­rea­son­able choice and we have their past record to go on. Nonethe­less, there are more open op­tions if this is im­por­tant to you.

  “Pro­duc­tive AI safety re­search work is in­ac­cessible to over 99.9% of the pop­u­la­tion, mak­ing this ad­vice al­most use­less to nearly ev­ery­one read­ing the ar­ti­cle.”—Not nec­es­sar­ily. Even if be­com­ing good enough to be a re­searcher is very hard, it prob­a­bly isn’t nearly as hard to be­come good enough at a par­tic­u­lar area to help men­tor other peo­ple.

• casebash 19 Oct 2019 20:41 UTC

  5 points

  on: How can I hire an EA re­search as­sis­tant?

  I’m definitely in favour of this kind of pro­ject since I feel more EAs should be ex­per­i­ment­ing with small pro­jects.

• casebash 11 Oct 2019 21:29 UTC

  2 points

  in reply to: NunoSempere's comment on: Shap­ley val­ues: Bet­ter than counterfactuals

  “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”—true, but the NGO’s coun­ter­fac­tual im­pact is re­duced when I feel it’s fairer for the NGO to be able to claim the full amount (though of course you’d never know the gov­ern­ment’s true mo­ti­va­tions in real life)

• casebash 11 Oct 2019 8:45 UTC

  16 points

  on: Shap­ley val­ues: Bet­ter than counterfactuals

  The or­der in­differ­ence of Shapely val­ues only makes sense from a per­spec­tive where there is perfect knowl­edge of what other play­ers will do, but if you don’t have that, a party that spent a huge amount of money on a pro­ject that was al­most cer­tainly go­ing to be waste­ful and ended up be­ing saved when by sheer hap­pen­stance an­other party ap­peared to save the pro­ject was not mak­ing good spend­ing de­ci­sions. Similarly, many agents won’t be op­ti­mis­ing for Shapely value, say a gov­ern­ment which spends money on in­fras­truc­ture not car­ing about whether it’ll be used or not just to win poli­ti­cal points, so they don’t prop­erly de­serve a share of the gains when some­one else in­ter­venes with no­tifi­ca­tions to make the pro­ject ac­tu­ally effec­tive.

  I feel that this ar­ti­cle pre­sents Shap­ley value as just plain su­pe­rior, when in­stead a com­bi­na­tion of both Shap­ley value and coun­ter­fac­tual value will likely be a bet­ter met­ric. Beyond this, what you re­ally want to use is some­thing more like FDT where you take into ac­count the fact that the de­ci­sions of some agents are sub­junc­tively linked to you and that the de­ci­sions of some other agents aren’t. Even though my cur­rent the­ory is that very, very few agents are ac­tu­ally sub­junc­tively linked to you, I sus­pect that think­ing about prob­lems in this fash­ion is likely to work rea­son­ably well in prac­tise (I would need to ded­i­cate a solid cou­ple of hours in or­der to be able to write out my rea­sons for be­liev­ing this more con­cretely)

• casebash 15 Sep 2019 0:22 UTC

  12 points

  on: case­bash’s Shortform

  If we run any more anony­mous sur­veys, we should en­courage peo­ple to pause and con­sider whether they are con­tribut­ing pro­duc­tively or just vent­ing. I’d still be in favour of shar­ing all the re­sponses, but I have enough faith in my fel­low EAs to be­lieve that some would take this to heart.

[Question] Why were peo­ple skep­ti­cal about RAISE?

• casebash 27 Aug 2019 14:12 UTC

  23 points

  on: Move­ment Col­lapse Scenarios

  I’m most con­cerned about at­tempts to poli­ti­cise the move­ment as un­like most of the other risks, this risk is ad­ver­sar­ial. EA has to thread the nee­dle of op­er­at­ing and main­tain­ing our rep­u­ta­tion in a poli­ti­cised en­vi­ron­ment with­out let­ting this dis­tort our way of think­ing.

• casebash 27 Aug 2019 8:05 UTC

  2 points

  on: case­bash’s Shortform

  I sus­pect that it could be im­pact­ful to study say a mas­ters of AI or com­puter sci­ence even if you don’t re­ally need it. Univer­sity pro­vides one of the best op­por­tu­ni­ties to meet and deeply con­nect with peo­ple in a par­tic­u­lar field and I’d be sur­prised if you couldn’t per­suade at least a cou­ple of peo­ple of the im­por­tance of AI safety with­out re­ally try­ing. On the other hand, if you went in with the in­ten­tion of net­work­ing as much as pos­si­ble, I think you could have much more suc­cess.

• casebash 22 Aug 2019 22:42 UTC

  3 points

  on: Effec­tive Altru­ism Lon­don Strat­egy 2019

  In­ter­est­ing read­ing your strat­egy, par­tic­u­larly what you aren’t fo­cus­ing on. The one part I’d be some­what skep­ti­cal of is de­creas­ing up­skil­ling. Peo­ple, par­tic­u­larly the peo­ple that we want to join our com­mu­nity, want to grow and im­prove. It’s im­por­tant to be re­al­is­tic about how much some­one can up­skill in a limited amount of time, but these kinds of events seem like a key draw.

• casebash 21 Aug 2019 11:17 UTC

  5 points

  on: case­bash’s Shortform

  One of the vague ideas spin­ning around in my head is that maybe in ad­di­tion to EA which is a fairly open, loosely co-or­di­nated, big-tent move­ment with sev­eral differ­ent cause ar­eas, there would also in value in a more se­lec­tive, tightly co-or­di­nated, nar­row move­ment fo­cus­ing just on the long term fu­ture. In­ter­est­ingly, this would be an ac­cu­rate de­scrip­tion of some EA orgs, with the key differ­ence be­ing that these orgs tend to rely on paid staff rather than vol­un­teers. I don’t have a solid idea of how this would work, but just thought I’d put this out there...

case­bash’s Shortform

• casebash 8 Aug 2019 1:59 UTC

  3 points

  on: Why has poverty wor­ld­wide fallen so lit­tle in re­cent decades out­side China?

  That is pretty con­cern­ing. I would love an ex­pla­na­tion of this as well!

• casebash 4 Aug 2019 22:23 UTC

  18 points

  on: What types of or­ga­ni­za­tions would be ideal to be dis­tribut­ing fund­ing for EA? (Fel­low­ships, Or­ga­ni­za­tions, etc)

  I’m strongly in favour of cre­at­ing a fel­low­ship with a fancy name and web­site in or­der to al­low peo­ple to build ca­reer cap­i­tal; or at least make ac­cept­ing these fel­low­ships not a step back­wards. EA Grant doesn’t ex­actly sound pres­ti­gious.