jacobjacob

• jacobjacob 22 Mar 2019 15:10 UTC

  4 points

  in reply to: Pablo_Stafforini's comment on: Un­der­stand­ing in­for­ma­tion cascades

  more re­cent data sug­gests that the suc­cesses of the ex­trem­iz­ing al­gorithm dur­ing the fore­cast­ing tour­na­ment were a fluke.

  Do you have a link to this data?

• jacobjacob 14 Mar 2019 15:45 UTC

  8 points

  in reply to: Jan_Kulveit's comment on: Un­der­stand­ing in­for­ma­tion cascades

  I haven’t looked through your links in much de­tail, but wanted to re­ply to this:

  Over­all I would sug­gest to ap­proach this with some in­tel­lec­tual hu­mil­ity and study ex­ist­ing re­search more, rather then try to rein­vent large part of net­work sci­ence on LessWrong. (My guess is some­thing like >2000 re­search years were spent on the topic of­ten by quite good peo­ple.)

  I ei­ther dis­agree or am con­fused. It seems good to use re­sources to out­source your abil­ity to do liter­a­ture re­views, dis­til­la­tion or ex­trap­o­la­tion, to some­one with higher com­par­a­tive ad­van­tage. If the LW ques­tion fea­ture can en­able that, it will make the mar­ket for in­tel­lec­tual progress more effi­cient; and I wanted to test whether this was so.

  I am not try­ing to rein­vent net­work sci­ence, and I’m not that in­ter­ested in the large amount of the­o­ret­i­cal work that has been done. I am try­ing to 1) ap­ply these in­sights to very par­tic­u­lar prob­lems I face (re­lat­ing to fore­cast­ing and more); and 2) think about this from a cost-effec­tive­ness per­spec­tive.

  I am very happy to trade money for my time in an­swer­ing these ques­tions.

  (Nei­ther 1) nor 2) seems like some­thing I ex­pect the ex­ist­ing liter­a­ture to have been very in­ter­ested in. I be­lieve this for similar rea­sons to those Holden Karnofsky ex­press here.)

• jacobjacob 14 Mar 2019 15:20 UTC

  6 points

  in reply to: Kaj_Sotala's comment on: Un­der­stand­ing in­for­ma­tion cascades

  Seems like a sen­si­ble worry, and we did con­sider some ver­sion of it. My rea­son­ing was roughly:

  1) The ques­tions fea­ture is quite new, and if it will be very valuable, most use-cases and the proper UI haven’t been dis­cov­ered yet (these can be hard to pre­dict in ad­vance with­out get­ting users to play around with differ­ent things and then talk­ing to them).

  No one has yet at­tempted to use mul­ti­ple ques­tions. So it would be valuable for the LW team and the com­mu­nity to ex­per­i­ment with that, de­spite pos­si­ble coun­ter­vailing con­sid­er­a­tions (any good ex­per­i­ment will have suffi­cient un­cer­tainty that such con­sid­er­a­tions will always ex­ist).

  2) Ques­tions ¹⁄₂, 3 and 4 are quite differ­ent, and it seems good to be able to do re­search on one sub-prob­lem with­out tak­ing mind­share from ev­ery­one work­ing on any sub­prob­lem.

[Question] For­mal­is­ing con­tin­u­ous info cas­cades? [Info-cas­cade se­ries]

[Question] How large is the harm from info-cas­cades? [Info-cas­cade se­ries]

[Question] How can we re­spond to info-cas­cades? [Info-cas­cade se­ries]

[Question] Distri­bu­tion of info-cas­cades across fields? [Info-cas­cade se­ries]

[Question] Un­der­stand­ing in­for­ma­tion cascades

• jacobjacob 13 Mar 2019 10:46 UTC

  3 points

  on: Un­der­stand­ing in­for­ma­tion cascades

  See this post for a good, sim­ple math­e­mat­i­cal de­scrip­tion of the dis­crete ver­sion of the phe­nomenon.

• jacobjacob 13 Mar 2019 10:26 UTC

  11 points

  on: How large is the harm from info-cas­cades? [Info-cas­cade se­ries]

  Me and Ben Pace (with some help from Niki Shams) made a Guessti­mate model of how much in­for­ma­tion cas­cades is cost­ing sci­ence in terms of wasted grant money. The model is largely based on the ex­cel­lent pa­per “How cita­tion dis­tor­tions cre­ate un­founded au­thor­ity: anal­y­sis of a cita­tion net­work” (Green­berg, 2009), which traces how an un­cer­tain claim in biomedicine is in­flated to es­tab­lished knowl­edge over a pe­riod of 15 years, and used to jus­tify ~$10 mil­lion in grant money from the NIH (we calcu­lated the num­ber our­selves here).

  There are many open ques­tions about some of the in­puts to our model as well as how this gen­er­al­ises out­side of academia (or even out­side of biomedicine). How­ever, we see this as a “Jel­ly­baby” in Dou­glas Hub­bard’s sense—it’s a first data-point and stab at the prob­lem which brings us from “no idea idea how big or small the costs of info-cas­cades are”, to at least “it is plau­si­ble though very un­cer­tain that the costs can be on the or­der of mag­ni­tude of billions of dol­lars, yearly, in aca­demic grant money”.

• jacobjacob 13 Mar 2019 10:26 UTC

  5 points

  on: How large is the harm from info-cas­cades? [Info-cas­cade se­ries]

  This might be an in­ter­est­ing poin­ter.

  In Note-8 in the sup­ple­men­tary ma­te­ri­als, Green­berg be­gins to quan­tify the prob­lem. He defines an am­plifi­ca­tion mea­sure for pa­per P as the num­ber of cita­tion-paths origi­nat­ing at P and ter­mi­nat­ing at all other pa­pers, ex­cept for paths of length 1 flow­ing di­rectly to pri­mary data pa­pers. The am­plifi­ca­tion den­sity of a net­work is the mean am­plifi­ca­tion across its pa­pers.

  Green­berg then finds that, in the par­tic­u­lar net­work analysed, you can achieve am­plifi­ca­tion den­sity of about 1000 over a 15 year time-frame. This den­sity grows ex­po­nen­tially with a dou­bling time of very roughly 2 years.

• jacobjacob 13 Mar 2019 10:17 UTC

  18 points

  on: Un­der­stand­ing in­for­ma­tion cascades

  Here’s a quick bibliog­ra­phy we threw to­gether.

  Back­ground:

  • In­for­ma­tion Cas­cades and Ra­tional Herd­ing: An An­no­tated Bibliog­ra­phy and Re­source Refer­ence (Bikchan­dani et al. 2004). The best re­source on the topic, see in par­tic­u­lar the ini­tial pa­pers on the sub­ject.

  • Y2K Bibliog­ra­phy of Ex­per­i­men­tal Eco­nomics and So­cial Science In­for­ma­tion Cas­cades and Herd Effects (Holt, 1999. Less thor­ough, but catches some pa­pers the first one misses.

  • “In­for­ma­tion cas­cade” from Wikipe­dia. An ex­cel­lent in­tro­duc­tion.

  • “Un­der­stand­ing In­for­ma­tion Cas­cades” from In­vesto­pe­dia.

  Pre­vi­ous LessWrong posts refer­ring to info cas­cades:

  • In­for­ma­tion cas­cades, by Johni­cholas, 2009

  • In­for­ma­tion cas­cades in sci­en­tific prac­tice, by RichardKen­n­away, 2009

  • In­for­ma­tion cas­cades, LW Wiki

  And then here are all the LW posts we could find that used the con­cept (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11) . Not sure how rele­vant they are, but might be use­ful in ori­ent­ing around the con­cept.

• jacobjacob 10 Mar 2019 11:24 UTC

  1 point

  on: [Er­ror com­mu­ni­cat­ing with LW2 server]

  SPOILER WARNING

  Schel­ling cafe? What Schel­ling cafe?

  One you went to be­fore, maybe?

  And who on earth do you talk to? Maybe the guy who sat with us in a cab af­ter EA Global Lonon last fall...

• jacobjacob 27 Feb 2019 17:38 UTC

  17 points

  in reply to: johnswentworth's comment on: Un­con­scious Economies

  I found my­self in a situ­a­tion like: “if this is com­mon knowl­edge within econ, writ­ing an ex­pla­na­tion would sig­nal I’m not part of econ and hence my econ opinions are low sta­tus”, but de­cided to go ahead any­way.

  It’s good you found it helpful. I’m won­der­ing if equil­ibria like the above is a mechanism pre­vent­ing im­por­tant stuff from be­ing dis­til­led.

• jacobjacob 27 Feb 2019 17:34 UTC

  8 points

  in reply to: strangepoop's comment on: Un­con­scious Economies

  I re­ally ap­pre­ci­ate you cit­ing that.

  I should have made it clearer, but for refer­ence, the works I’ve been ex­posed to:

  • Hal Var­ian’s un­der­grad textbook

  • Marginal Revolu­tion University

  • Some amount of listen­ing to Econ Talk, read­ing In­vesto­pe­dia and Wikipe­dia articles

  • MSc de­gree at LSE

Un­con­scious Economies

• jacobjacob 27 Feb 2019 11:50 UTC

  3 points

  on: Less Com­pe­ti­tion, More Mer­i­toc­racy?

  For sec­tion III. it would be re­ally helpful to con­cretely work through what hap­pens in the ex­am­ples of di­vorce, nu­clear war, gov­ern­ment de­fault, etc. What’s a plau­si­ble thought pro­cess of the agents in­volved?

  My cur­rent model is some­thing like “my mar­riage is worse than I find tol­er­able, so I have noth­ing to loose. Now that di­vorce is le­gal, I might as well gam­ble my sav­ings in the cas­ino. If I win we could move to a bet­ter home and maybe save the re­la­tion­ship, if I lose we’ll get di­vorced.”

  Peo­ple who have noth­ing to lose start tak­ing risks which fill up the merely pos­si­bly bad out­comes un­til they start mat­ter­ing.

• jacobjacob 12 Feb 2019 10:37 UTC

  4 points

  in reply to: Donald Hobson's comment on: How im­por­tant is it that LW has an un­limited sup­ply of karma?

  In the broader econ­omy, it’s not the case that “If buy­ing things re­duced your in­come, peo­ple stop buy­ing things, and even­tu­ally money stops flow­ing al­to­gether”.

  So the only way that makes sense to me is if you model con­tent as a pub­lic good which no user is in­cen­tivised to con­tribute to main­tain­ing.

  Spec­u­la­tively, this might be avoided if votes were pub­lic: be­cause then vot­ing would be a costly sig­nal of one’s epistemic val­ues or other things.

• jacobjacob 12 Feb 2019 10:32 UTC

  2 points

  in reply to: Pattern's comment on: How im­por­tant is it that LW has an un­limited sup­ply of karma?

  though I’m not sure how that is calcu­la­teed from one’s karma

  I be­lieve it’s pro­por­tional to the log of your user karma. But I’m not sure.

  One can get high karma from a small amount of con­tent that a small num­ber of suffi­ciently high karma users that dou­ble up vote it.

  There is still an in­cen­tive gra­di­ent to­wards “least pub­lish­able units”.

  Sup­pose you have a piece of work worth 18 karma to high-karma user U. How­ever, U’s strong up­vote is only worth 8 karma.

  If you just post one piece of work, you get 8 karma. If you split your work into three pieces, each of which U val­ues at 6 karma, you’re bet­ter off. U might strong-up­vote all of them (they’d rather al­lo­cate a lit­tle too much karma than way too lit­tle), and you get 24 karma.

  To the ex­tend the metaphor in the origi­nal ques­tion: maybe if the world econ­omy ran on the equiv­a­lent of strong up­votes there would still be cars around, yet no one could buy air­planes.

• jacobjacob 12 Feb 2019 10:17 UTC

  2 points

  in reply to: Raemon's comment on: How im­por­tant is it that LW has an un­limited sup­ply of karma?

  Do you have de­tails on when and why that was re­moved? Or past posts dis­cussing that sys­tem?