cole_haus

Sum­mary of Cartwright and Hardie’s “Ev­i­dence-based Policy: A prac­ti­cal guide to do­ing it bet­ter”

• cole_haus 21 May 2019 18:41 UTC

  6 points

  on: EA Fo­rum: Foot­notes are live, and other updates

  Nice!

  Per­haps you all have con­sid­ered this already, but I think there’s a lot to like about side­notes over foot­notes, es­pe­cially on the web (e.g. foot­notes aren’t always in sight at the bot­tom of a phys­i­cal page).

• cole_haus 10 May 2019 22:02 UTC

  4 points

  on: Struc­ture EA or­ga­ni­za­tions as WSDNs?

  How would you ex­pect EA WSDNs to differ from cur­rent EA orgs con­cretely?

  When it comes to worker co­op­er­a­tives, I see the differ­ences as all flow­ing from re­duc­ing con­flict­ing in­ter­ests. That is, in stan­dard firms, own­ers are ul­ti­mately in­ter­ested in prof­its and only in­stru­men­tally in­ter­ested in work­ing con­di­tions while work­ers are ul­ti­mately in­ter­ested in work­ing con­di­tions (broadly con­strued) and only in­stru­men­tally in­ter­ested in prof­its. Worker co­op­er­a­tives re­solve this ten­sion by mak­ing agents prin­ci­pals and prin­ci­pals agents.

  This is an ideal­iza­tion, but it seems like the in­ter­ests of all rele­vant ac­tors in EA orgs (and non­prof­its more gen­er­ally?) are more al­igned. The board and the work­ers are (at least in the­ory) largely (if not solely) mo­ti­vated by the same do-good­ing goal.

• cole_haus 4 May 2019 4:10 UTC

  3 points

  on: What is the cur­rent best es­ti­mate of the cu­mu­la­tive elas­tic­ity of chicken?

  Con­sid­er­a­tion 1: Economists of­ten con­sider small ac­tors in com­pet­i­tive mar­kets to be price-tak­ers mean­ing that they can­not in­fluence prices on their own. This seems like a pretty plau­si­ble char­ac­ter­i­za­tion of any in­di­vi­d­ual food buyer.

  Con­sid­er­a­tion 2: “He rea­soned that eco­nomics says a drop in de­mand for some com­mod­ity should cause prices to fall for that com­mod­ity, and over­all con­sump­tion re­mains the same.” This is not cor­rect. In in­ward shift in the de­mand curve (“a drop in de­mand”) (for or­di­nary down­ward slop­ing de­mand curves and up­ward slop­ing sup­ply curves), causes both equil­ibrium price and quan­tity to de­crease. I’d guess the thing he’s try­ing to get at is that for a good which is unit elas­tic, a small drop in price is offset by a small in­crease in quan­tity which leads to to­tal rev­enue be­ing un­changed.

  So our first op­tion is to re­gard in­di­vi­d­ual ac­tors as too small to in­fluence the price. If we re­ject this and think they do have an effect, their effect would be to shift the de­mand curve in—drop­ping equil­ibrium price and quan­tity.

  Aside: I’m rea­son­ably well-in­formed about eco­nomics and don’t re­call hav­ing ever heard the term “cu­mu­la­tive elas­tic­ity” be­fore.

• cole_haus 24 Apr 2019 2:00 UTC

  7 points

  on: Why does EA use QALYs in­stead of ex­pe­rience sam­pling?

  I don’t re­ally see ESM as be­ing in op­po­si­tion to QALYs. It seems like it’s a method that you would use as an in­put in QALY weight de­ter­mi­na­tions. Wikipe­dia lists some of the cur­rent meth­ods for de­riv­ing QALY weights as:

  Time-trade-off (TTO): Re­spon­dents are asked to choose be­tween re­main­ing in a state of ill health for a pe­riod of time, or be­ing re­stored to perfect health but hav­ing a shorter life ex­pec­tancy.

  Stan­dard gam­ble (SG): Re­spon­dents are asked to choose be­tween re­main­ing in a state of ill health for a pe­riod of time, or choos­ing a med­i­cal in­ter­ven­tion which has a chance of ei­ther restor­ing them to perfect health, or kil­ling them.

  Vi­sual analogue scale (VAS): Re­spon­dents are asked to rate a state of ill health on a scale from 0 to 100, with 0 rep­re­sent­ing be­ing dead and 100 rep­re­sent­ing perfect health. This method has the ad­van­tage of be­ing the eas­iest to ask, but is the most sub­jec­tive.

  There’s also the “day re­con­struc­tion method” (DRM). The Oxford Hand­book of Hap­piness talks about ESM, DRM and oth­ers rele­vant mea­sure­ment ap­proaches at var­i­ous points.

  I’d guess the trou­ble with us­ing ESM, DRM and some other meth­ods like them for QALY weights is it’s hard to iso­late the causal effect of par­tic­u­lar con­di­tions us­ing these meth­ods.

• cole_haus 18 Apr 2019 23:29 UTC

  7 points

  in reply to: Robert_Wiblin's comment on: Thoughts on 80,000 Hours’ re­search that might help with job-search frustrations

  Ah, I see that now. Thanks.

  FWIW, I was speci­fi­cally look­ing for a dis­claimer and it didn’t quickly come to my at­ten­tion. It looks like a few other peo­ple in these sub­threads may have also missed the dis­claimer.

• cole_haus 18 Apr 2019 20:06 UTC

  10 points

  in reply to: lexande's comment on: Thoughts on 80,000 Hours’ re­search that might help with job-search frustrations

  Yeah, I hadn’t re­al­ized it was more or less de­p­re­cated. (The page it­self doesn’t seem to give any in­di­ca­tion of that. Edit: Ah, it does. I missed the sec­ond para­graph of the side­note when I quickly scanned for some dis­claimer.)

  Also, ap­par­ently un­for­tu­nately, it’s the first sub­link un­der the 80,000 Hours site on Google if you search for 80,000 Hours.

• cole_haus 16 Apr 2019 20:06 UTC

  15 points

  in reply to: MichaelPlant's comment on: Thoughts on 80,000 Hours’ re­search that might help with job-search frustrations

  It seems quite pos­si­ble to me have a “pa­ram­e­ter­ized list”. That is, recom­men­da­tions can take the shape “If X is true of you, Y and Z are good op­tions.” And in fact 80,000 Hours does do this to some de­gree (via, for ex­am­ple, their ca­reer quiz). While this isn’t en­tirely per­son­al­ized (it’s based only on cer­tain at­tributes that 80,000 Hours high­lights), it’s also far from a sin­gle, defini­tive list. So it doesn’t seem to be that there’s any in­sol­u­ble ten­sion be­tween tak­ing ac­count of in­di­vi­d­ual differ­ence and com­mu­ni­cat­ing the same mes­sage to a broad au­di­ence—you just have to rely on the au­di­ence to do some in­ter­pret­ing.

• cole_haus 10 Apr 2019 5:23 UTC

  1 point

  in reply to: Habryka's comment on: Long-Term Fu­ture Fund: April 2019 grant decisions

  I don’t par­tic­u­larly want to try to re­solve the dis­agree­ment here, but I’d think value per dol­lar is pretty differ­ent for dol­lars at EA in­sti­tu­tions and for dol­lars with (many) EA-al­igned peo­ple [1]. It seems like the whole fil­ter­ing/​se­lec­tion pro­cess of grant­ing is pred­i­cated on this as­sump­tion. Maybe you be­lieve that peo­ple at CEA are the type of peo­ple that would make very good use of money re­gard­less of their in­sti­tu­tional af­fili­a­tion?

  [1] I’d ex­pect it to vary from per­son to per­son de­pend­ing on their al­ign­ment, com­mit­ment, com­pe­tence, etc.

• cole_haus 10 Apr 2019 1:55 UTC

  23 points

  in reply to: Habryka's comment on: Long-Term Fu­ture Fund: April 2019 grant decisions

  I am not OP but as some­one who also has (minor) con­cerns un­der this head­ing:

  • Some peo­ple judge HPMoR to be of lit­tle artis­tic merit/​low aes­thetic quality

  • Some peo­ple find the sub­cul­tural af­fili­a­tions of HPMoR off-putting (fan­fic­tion in gen­eral, co­pi­ous refer­ences to other ar­guably low-sta­tus fan­doms)

  If the re­cip­i­ents have nega­tive im­pres­sions of HPMoR for rea­sons like the above, that could re­sult in (un­nec­es­sar­ily) nega­tive im­pres­sions of ra­tio­nal­ity/​EA.

  Clearly, there also many peo­ple that like HPMoR and don’t have the above con­cerns. The key ques­tion is prob­a­bly what frac­tion of re­cip­i­ents will have pos­i­tive, neu­tral and nega­tive re­ac­tions.

• cole_haus 10 Apr 2019 1:44 UTC

  5 points

  in reply to: Habryka's comment on: Long-Term Fu­ture Fund: April 2019 grant decisions

  It’s not at all clear to me why the whole $150k of a coun­ter­fac­tual salary would be counted as a cost. The most rea­son­able (sim­ple) model I can think of is some­thing like: ($150k * .1 + $60k) * 1.5 = $112.5k where the $150k*.1 term is the amount of salary they might be ex­pected to donate from some coun­ter­fac­tual role. This then gives you the to­tal “EA dol­lars” that the po­si­tions cost whereas your model seems to com­bine “EA dol­lars” (CEA costs) and “per­sonal dol­lars” (their to­tal salary).

• cole_haus 10 Apr 2019 1:42 UTC

  5 points

  in reply to: Habryka's comment on: Long-Term Fu­ture Fund: April 2019 grant decisions

  I think you have some math er­rors:

  • $150k * 1.5 + $60k = $285k rather than $295k

  • Pre­sum­ably, this should be ($150k + $60k) * 1.5 = $315k ?

• cole_haus 5 Apr 2019 20:51 UTC

  20 points

  in reply to: G Gordon Worley III's comment on: Most im­por­tant un­fulfilled role in the EA ecosys­tem?

  I have a pretty averse re­ac­tion to all the peo­ple you named, ex­pect I would feel similarly about some­one in that mold in EA, and ex­pect many other peo­ple in EA would feel similarly. I don’t think charis­matic lead­er­ship fits all that well with the other el­e­ments of EA in ways both im­por­tant and in­ci­den­tal.

• cole_haus 4 Apr 2019 19:11 UTC

  4 points

  on: [Link] The Op­ti­mizer’s Curse & Wrong-Way Reductions

  I don’t know how promis­ing oth­ers think this is, but I quite liked Con­cepts for De­ci­sion Mak­ing un­der Se­vere Uncer­tainty with Par­tial Or­di­nal and Par­tial Car­di­nal Prefer­ences. It tries to out­line pos­si­ble de­ci­sion pro­ce­dures once you re­lax some of the sub­ject ex­pected util­ity the­ory as­sump­tions you ob­ject to. For ex­am­ple, it talks about the pos­si­bil­ity of hav­ing a credal set of be­liefs (if one ob­jects to the idea of as­sign­ing a sin­gle prob­a­bil­ity) and then do­ing max­imin on this i.e. se­lect­ing the out­come that has the best ex­pected util­ity ac­cord­ing to its least fa­vor­able cre­dences.

• cole_haus 4 Apr 2019 19:06 UTC

  10 points

  on: [Link] The Op­ti­mizer’s Curse & Wrong-Way Reductions

  There’s ac­tu­ally a thing called the Satis­ficer’s Curse (pdf) which is even more gen­eral:

  The Satis­ficer’s Curse is a sys­tem­atic over­val­u­a­tion that oc­curs when any un­cer­tain prospect is cho­sen be­cause its es­ti­mate ex­ceeds a pos­i­tive thresh­old. It is the most gen­eral ver­sion of the three curses, all of which can be seen as statis­ti­cal arte­facts.

• cole_haus 2 Apr 2019 5:02 UTC

  1 point

  on: Is any EA or­ga­ni­za­tion us­ing or con­sid­er­ing us­ing Bu­terin et al.’s mechanism for match­ing funds?

  IIRC, the mechanism has prob­lems with col­lu­sion/​dis­sem­bling. For ex­am­ple, one backer with $46 dol­lars and 4 back­ers with $1 each will get sig­nifi­cantly bet­ter re­sults by split­ting their money into 5 con­tri­bu­tions of $10 each. This seems like a prob­lem that’s ac­tu­ally mod­er­ately likely to arise in prac­tice.

• cole_haus 30 Mar 2019 4:25 UTC

  2 points

  on: a black swan en­ergy prize

  It looks like the case you’re mak­ing in the “a prize” sec­tion is that prizes are more open to “out­siders” than grants which seems gen­er­ally plau­si­ble to me. On the other hand, grants can ac­tu­ally fund the re­search it­self while con­tes­tants for a prize need some source of fund­ing. If it’s cap­i­tal-in­ten­sive to mount a se­ri­ous at­tempt at the prize, this cre­ates a fund­ing and vet­ting prob­lem again (con­tes­tants will need money to bankroll their at­tempt).

• cole_haus 30 Mar 2019 4:25 UTC

  2 points

  on: a black swan en­ergy prize

  I’m not con­vinced that a prize is par­tic­u­larly helpful in this case. I think of prizes as use­ful for in­duc­ing in­vest­ment in things like pub­lic goods where pri­vate re­turns are limited. That doesn’t seem to be the case here; suc­cess­fully cre­at­ing “rad­i­cally bet­ter en­ergy gen­er­a­tion” seems like it would be wildly re­mu­ner­a­tive. The promise of vast wealth seems like it ought to be suffi­cient in­cen­tive re­gard­less of a prize.

  OTOH, that’s all very first-prin­ci­ples and the his­tory of in­no­va­tion prizes doesn’t seem to re­ally pay much at­ten­tion to this line of crit­i­cism. Maybe prizes make par­tic­u­lar prob­lems more salient, etc.

• cole_haus 14 Mar 2019 18:06 UTC

  2 points

  on: How to Un­der­stand and Miti­gate Risk (Cross­post from LessWrong)

  This is in­ter­est­ing! I think it would also be use­ful to talk about the stan­dard ter­minol­ogy in the field. Some of those terms are:

  • Aleatoric and epistemic uncertainty

  • De­ci­sions un­der risk vs de­ci­sions un­der ignorance

  Rea­sons I think it’s use­ful to talk about stan­dard ter­minol­ogy:

  • Allows you to con­verse with oth­ers and un­der­stand their work more easily

  • Allows read­ers to fol­low up and con­nect with a larger body of work

  • Com­mu­ni­cates to ex­perts that you’ve se­ri­ously en­gaged with the field and un­der­stand it

  In this par­tic­u­lar case, I’d be in­ter­est­ing in hear­ing how your cat­e­gories map to the stan­dard ones. Or, if you think they don’t, it would be in­ter­est­ing to hear why that is. What are the in­ad­e­qua­cies of the stan­dard terms and cat­e­gories?

• cole_haus 20 Feb 2019 7:12 UTC

  3 points

  on: Im­pact Prizes as an al­ter­na­tive to Cer­tifi­cates of Impact

  This seems very re­lated to so­cial im­pact bonds: “So­cial Im­pact Bonds are a type of bond, but not the most com­mon type. While they op­er­ate over a fixed pe­riod of time, they do not offer a fixed rate of re­turn. Re­pay­ment to in­vestors is con­tin­gent upon speci­fied so­cial out­comes be­ing achieved.”