bfinn

• bfinn 18 Feb 2019 15:35 UTC

  3 points

  in reply to: Raemon's comment on: Some dis­junc­tive rea­sons for ur­gency on AI risk

  FWIW an­other rea­son, some­what similar to the low hang­ing fruit point, is that be­cause the re­main­ing prob­lems are in­creas­ingly spe­cial­ized, they re­quire more years’ train­ing be­fore you can tackle them. I.e. not just harder to solve once you’ve started, but it takes longer for some­one to get to the point where they can even start.

  Also, I won­der if the in­creas­ing spe­cial­iza­tion means there are more prob­lems to solve (albeit ever more niche), so peo­ple are be­ing spread thin­ner among them. (Though con­versely there are more peo­ple in the world, and many more sci­en­tists, than a cen­tury or two ago.)

• bfinn 18 Feb 2019 15:14 UTC

  9 points

  in reply to: RobinHanson's comment on: Some dis­junc­tive rea­sons for ur­gency on AI risk

  In soft­ware de­vel­op­ment, a per­haps rele­vant kind of prob­lem solv­ing, ex­tra re­sources in the form of more pro­gram­mers work­ing on the same pro­ject doesn’t speed things up much. My guessti­mate is out­put = time x log pro­gram­mers. I as­sume the main rea­son be­ing be­cause there’s a limit to the ex­tent that you can di­vide a pro­ject into in­de­pen­dent par­allel pro­gram­ming tasks. (Cf 9 women can’t make a baby in 1 month.)

  Ex­cept that if the peo­ple are work­ing in in­de­pen­dent smaller teams, each try­ing to crack the same prob­lem, and *if* the solu­tion re­quires a sin­gle break­through (or a few?) which can be made by a smaller team (e.g. pub­lic key en­cryp­tion, as op­posed to land­ing a man on the moon), then pre­sum­ably it’s pro­por­tional to the num­ber of teams, be­cause each has an in­de­pen­dent prob­a­bil­ity of mak­ing the break­through. And it seems plau­si­ble that solv­ing AI threats might be more like this.

• bfinn 12 Feb 2019 17:01 UTC

  3 points

  in reply to: shminux's comment on: Some Thoughts on Metaphilosophy

  But philoso­phers are good at propos­ing an­swers—they all do that, usu­ally just af­ter iden­ti­fy­ing a flaw with an ex­ist­ing pro­posal.

  What they’re not good at is con­vinc­ing ev­ery­one else that their solu­tion is the right one. (And pre­sum­ably this is be­cause mul­ti­ple solu­tions are plau­si­ble. And maybe that’s be­cause of the na­ture of proof—it’s im­pos­si­ble to prove some­thing defini­tively, and dis­prov­ing typ­i­cally in­volves find­ing a coun­terex­am­ple, which may be hard to find.)

  I’m not con­vinced philos­o­phy is much less good at find­ing ac­tual an­swers than say physics. It’s not as if physics is com­pletely solved, or even par­tic­u­larly sta­ble. Per­haps its most promis­ing pe­riod of sta­bil­ity was speci­fi­cally the laws of mo­tion & grav­ity af­ter New­ton—though for less than two cen­turies. Physics seems bet­ter than philos­o­phy at form­ing a tem­po­rary con­sen­sus; but that’s no use (and in­deed is coun­ter­pro­duc­tive) un­less the solu­tion is ac­tu­ally right.

  Cf a rare ex­am­ple of con­sen­sus in philos­o­phy: knowl­edge was ‘solved’ for 2300 years with the the­ory that it’s a ‘true jus­tified be­lief’. Un­til Get­tier thought of coun­terex­am­ples.

• bfinn 10 Feb 2019 23:13 UTC

  2 points

  in reply to: Yoav Ravid's comment on: Ur­gent & im­por­tant: How (not) to do your to-do list

  Thanks—yes I think there is a case for hav­ing a ‘Shouldn’t’ list. As you im­ply, it should only be for things you know are use­less/​harm­ful, not for things you’ve merely de­cided not to do be­cause they are low im­por­tance (e.g. paint the bath­room). Hence ‘shouldn’t do’ not merely ‘don’t do’.

• bfinn 2 Feb 2019 20:11 UTC

  2 points

  in reply to: Bucky's comment on: Ur­gent & im­por­tant: How (not) to do your to-do list

  Some­times you can del­e­gate things to your boss—e.g. by de­clin­ing work he tries to del­e­gate to you (say “I’m too busy”).

• bfinn 2 Feb 2019 11:24 UTC

  1 point

  in reply to: Bucky's comment on: Ur­gent & im­por­tant: How (not) to do your to-do list

  Thanks—glad you like it. I don’t know how the Eisen­hower Box is usu­ally taught, but from refer­ences to it on­line in e.g. blogs peo­ple don’t seem to ques­tion its val­idity. But in prac­tice they can’t be fol­low­ing it that liter­ally, e.g. they won’t be do­ing all the things it tells them to do and del­e­gat­ing all the things it tells them to del­e­gate, etc. So I sup­pose they must be treat­ing it as just a rough-and-ready guide.

Ur­gent & im­por­tant: How (not) to do your to-do list

• bfinn 1 Feb 2019 16:38 UTC

  2 points

  in reply to: Adele Lopez's comment on: Edi­tor Mini-Guide

  Looks like there is, but they must be LessWrong mem­bers.

• bfinn 30 Nov 2018 15:57 UTC

  1 point

  on: Jobs In­side the API

  Th­ese agents are noth­ing but a stupid in­ter­face layer be­tween me and the flight man­age­ment sys­tem.

  I sug­gest a pos­si­ble term for this is MUI: Meat User In­ter­face. The cus­tomer in­ter­acts with the MUI, and the MUI in­ter­acts with the GUI.

• bfinn 19 Nov 2018 15:07 UTC

  1 point

  in reply to: Sinal's comment on: Ra­tion­al­ity of demon­strat­ing & voting

  Yes I fol­low your ar­gu­ment, though I’m a bit doubt­ful about a re­sult that pro­duces a large differ­ence be­tween util­ity func­tion and moral credit.

  Re your Supreme Court ex­am­ple (and I agree this is a clearer way of think­ing about it), I don’t quite fol­low the ar­gu­ment. It’s true that if the other jus­tices had voted differ­ently, more of them would have had to vote differ­ently (‘flip’) had you done so, but as it’s a given that you knew how ev­ery­one else was go­ing to vote, flip­ping is ruled out—their votes are set in stone.

  And re ‘still each jus­tice’s prefer­ence… mat­ters’, I wasn’t clear if this is the same point or a sep­a­rate point—i.e. a sig­nal­ling or similar ar­gu­ment that the size of the ma­jor­ity mat­ters, e.g. poli­ti­cally.

• bfinn 13 Nov 2018 11:23 UTC

  0 points

  in reply to: norswap's comment on: Ra­tion­al­ity of demon­strat­ing & voting

  A lit­tle bit of al­tru­ism still seems to make it ra­tio­nal even if you care al­most en­tirely about your­self—see the ex­am­ple calcu­la­tions.

  I used to think that mak­ing vot­ing manda­tory was a good solu­tion, but nowa­days I think it’s a dra­co­nian mea­sure. Be­cause what if you dis­ap­prove for ex­am­ple of the par­tic­u­lar vot­ing sys­tem (First Past the Post in the UK/​US)? Then forc­ing you to com­ply with it, per­haps only sym­bol­i­cally (as you can dis­com­ply in other ways like spoiling your bal­lot pa­per—un­less that will be crim­i­nal­ized too) is a waste of ev­ery­one’s time.

  Similarly if you don’t want to vote be­cause you are in­differ­ent be­tween the can­di­dates, or think you don’t know enough about the is­sues to choose a can­di­date, etc.

  Some­thing some­what similar to, but less dra­co­nian than, com­pul­sory vot­ing would be to pay peo­ple to vote, e.g. £5 /​ $5 in cash or vouch­ers as you exit the pol­ling sta­tion. Which would also some­what cor­rect the cur­rent skew in turnout—poorer peo­ple are cur­rently less likely to vote.

• bfinn 13 Nov 2018 10:26 UTC

  1 point

  in reply to: pranomostro's comment on: Pro­duc­tivity: In­stru­men­tal Rationality

  Hav­ing at least a plan for when to work, and be­ing strict about that, works for me. I set alarms on my phone to work in 1 hour fo­cussed bursts, with 15 minute breaks in be­tween, all morn­ing and late af­ter­noon—it seems most peo­ple do their best fo­cussed work in the morn­ing; there’s also that fa­mous vi­o­lin/​pi­ano stu­dent re­search which in­di­cates that the best stu­dents also prac­tice late af­ter­noon. I re­serve early/​mid af­ter­noon for light work (ad­min etc.)

  In ad­di­tion, I sug­gest you have a gen­eral plan for which pro­jects to work on dur­ing a week & month, and make a daily more spe­cific (though not nec­es­sar­ily de­tailed) plan first thing in the morn­ing, or (bet­ter) at the end of the work day for the next day.

• bfinn 11 Nov 2018 20:56 UTC

  1 point

  on: Pro­duc­tivity: In­stru­men­tal Rationality

  Yes, I’ve been track­ing my pro­duc­tivity daily for over 6 years. I do it us­ing a sim­ple iPhone app called ATracker, which lets you define pro­jects & cat­e­gories and hit a but­ton when­ever you start/​stop them.

  I use about a dozen cat­e­gories (for differ­ent types of work & broad types of leisure, also broad lo­ca­tions). Every week or two I ex­port the data into a spread­sheet and pro­duce some pretty charts and also many met­rics, e.g. about how my time us­age matches up to var­i­ous tar­gets.

  It’s kind of use­ful but I’m not that rigor­ous in keep­ing to the tar­gets. Nonethe­less if I start get­ting lax, then af­ter a few days or weeks I can’t pre­tend it’s not hap­pen­ing, and the data helps nudge me back into be­ing more pro­duc­tive.

  By the way, I think you’re be­ing overly am­bi­tious aiming at 12 hours of proper deep work per day. I think it’s very hard to av­er­age more than about 6 hours per day over long pe­ri­ods.

  If you start do­ing a similar kind of track­ing, I’d be happy to share with you the kinds of charts, met­rics etc. I pro­duce, some of which aren’t ob­vi­ous.

• bfinn 9 Nov 2018 11:14 UTC

  2 points

  in reply to: Elo's comment on: Help Me Re­fac­tor My­self I am Lost

  Yes, I found my thought pro­cesses im­proved dra­mat­i­cally re­cently when I stopped listen­ing to the ra­dio af­ter wak­ing in the morn­ing, and in the shower. I now have ex­cel­lent think­ing & ideas at that time of day. Silence and no dis­trac­tions are golden. (No won­der so many peo­ple have good ideas in the shower.)

  I also recom­mend hav­ing a notepad by your bed. I’ve done this for years. Some­times ideas (or things I for­got to do) oc­cur to me shortly be­fore go­ing to sleep, or oc­ca­sion­ally when wak­ing in the night, and I write them down in the dark. It gets them out of your head, which helps you sleep too.

• bfinn 9 Nov 2018 11:05 UTC

  1 point

  in reply to: Davidmanheim's comment on: Sub­sys­tem Alignment

  Thanks, I’ll read those with in­ter­est.

  I didn’t think it likely that busi­ness has solved any of these prob­lems, but I just won­der if it might in­clude par­tic­u­lar failures not found in other fields that help point to­wards a solu­tion; or even if it has found one or two pieces of a solu­tion.

  Busi­ness may at least provide some in­struc­tive ex­am­ples of how things go wrong. Be­cause busi­ness in­volves lots of peo­ple & mo­tives & co­op­er­a­tion & con­flict, things which I sus­pect hu­mans are par­tic­u­larly good at think­ing & rea­son­ing about (as op­posed to more ab­stract things).

• bfinn 9 Nov 2018 10:43 UTC

  1 point

  in reply to: ChristianKl's comment on: Ra­tion­al­ity of demon­strat­ing & voting

  I.e. hence you can’t tell how effec­tive a pres­i­dent will be from their party’s poli­cies, be­cause some­times their most effec­tive ac­tions are fol­low­ing their op­po­nents’ poli­cies.

  Yes, could be. It’s in line with the Pu­tanu­monit ar­gu­ments that you just can’t tell which party will be bet­ter for the coun­try.

  I can’t think of par­tic­u­lar in­stances of this in the UK, so I don’t know if this is more of a US thing. What quite of­ten hap­pens in the UK (par­tic­u­larly since Tony Blair) is par­ties steal­ing each oth­ers’ poli­cies, even some­times in stronger form than the other party. But pre­sum­ably that’s just them try­ing to tempt vot­ers across from the other side with oc­ca­sional juicy lit­tle morsels. I.e. both par­ties con­verg­ing on the me­dian voter. [ADDED] Though similar to your point that the other party may im­ple­ment your party’s poli­cies, and per­haps more effec­tively, which makes it harder to pre­dict which party would run the coun­try bet­ter.

• bfinn 9 Nov 2018 10:28 UTC

  2 points

  in reply to: Andaro's comment on: Ra­tion­al­ity of demon­strat­ing & voting

  Yes, in­ter­est­ing points. I haven’t re­ally given any thought to vot­ing as a re­ward/​pun­ish­ment, but many vot­ers do this. Though of course it’s mixed up with for­ward-look­ing vot­ing, since (for many peo­ple) you vote against a poli­ti­cian who did some­thing bad so that they won’t be around to do more bad things.

  And poli­ti­ci­ans an­ti­ci­pate pun­ish­ment-vot­ing as a de­ter­rent to them do­ing bad things, since there isn’t much other de­ter­rent (ex­cept the law).

  Also an in­ter­est­ing point re vot­ing as re­cip­ro­ca­tion to similar vot­ers as a kind of soli­dar­ity group. (Par­ties are them­selves soli­dar­ity groups, but so of course are spe­cial in­ter­est groups and other sup­port­ers of par­tic­u­lar poli­cies.)

  I’m not sure whether or how all this af­fects the calcu­lus. Eliezer wrote an ar­ti­cle on vot­ing a while back in which if I re­call his line was some­thing like ‘it’s all too com­pli­cated to model, so just stick to sim­ple rea­son­ing’.

  Re your pen­to­bar­bital ex­am­ple, this could be some­thing where the 0.7 cents di­rect effect on you is big­ger—though it would in­deed have to be some­thing ap­proach­ing a $1 billion effect to count (since the ex­pected benefit to you is this /​ 3 mil­lion, in the UK). Though that said al­most all is­sues like this af­fect quite a few other peo­ple too, so al­tru­ism makes it worth­while any­way.

• bfinn 9 Nov 2018 10:07 UTC

  1 point

  in reply to: Nicholas Conrad's comment on: Ra­tion­al­ity of demon­strat­ing & voting

  I sus­pect that the chances of a 3rd party win­ning are or­ders of mag­ni­tude lower than a 1st or 2nd, so the ex­pected value from you hav­ing the de­cid­ing vote would be too small. But in terms of policy in­fluence, if the 3rd party does un­usu­ally well (with­out win­ning), I agree that can be sig­nifi­cant. In­deed I re­call an ex­am­ple of this hap­pen­ing in the UK in the 1990s, when in one na­tional elec­tion the Green party (then the 4th or 5th party) did un­ex­pect­edly well, albeit still only get­ting a few % of the vote, which im­me­di­ately made the ma­jor par­ties start say­ing how im­por­tant the en­vi­ron­ment was and an­nounc­ing new poli­cies.

• bfinn 8 Nov 2018 14:45 UTC

  2 points

  in reply to: Nicholas Conrad's comment on: Ra­tion­al­ity of demon­strat­ing & voting

  Yes, but since on my num­bers the benefits of vot­ing are so huge, a tiny differ­ence be­tween par­ties can still jus­tify it. E.g. near the end of the ar­ti­cle I calcu­late that in a UK gen­eral elec­tion, if the differ­ence be­tween the two main par­ties equals 10% of gov­ern­ment spend­ing (in benefit to the coun­try, not nec­es­sar­ily ac­tual spend), that equals 7% of Brexit or about $7,000 to a marginal voter.

  So even if it’s only worth 0.1% of gov­ern­ment spend­ing (e.g. a small con­fi­dence that one party will make a small ex­e­cu­tion im­prove­ment on a few poli­cies), that’s $70 - enough to jus­tify vot­ing.

• bfinn 7 Nov 2018 23:04 UTC

  1 point

  in reply to: Jacobian's comment on: Ra­tion­al­ity of demon­strat­ing & voting

  [Re­sponse sub­stan­tially ed­ited:]

  If I un­der­stand you right, you’re say­ing that if Re­main were to hap­pen then Leavers would in­cur a large ac­tual loss (rel­a­tive to the Leave sce­nario), be­cause they reckon the benefits of leav­ing in terms of so­cial co­he­sion, se­cu­rity etc. will not oc­cur.

  Per­haps those aren’t the best ex­am­ples, as ar­guably those are mat­ters of fact, so Leavers could find out they were wrong if it turns out there is no loss in so­cial co­he­sion & se­cu­rity by re­main­ing; so they wouldn’t nec­es­sar­ily lose utils. A bet­ter ex­am­ple might be na­tional self-de­ter­mi­na­tion, which a Leave sup­porter would value come what may, and a Re­main sup­porter might put lit­tle value on. That is, Leavers aren’t merely pre­dict­ing that leav­ing the EU would make things bet­ter for the UK, they are ex­press­ing a (non-falsifi­able) prefer­ence for be­ing out of the EU.

  I haven’t thought of that, and that could be so—or per­haps more likely it’s a mix­ture of pre­dic­tion and prefer­ence. In which case Leavers would only get some nega­tive utils, still leav­ing tens of thou­sands of $ per ex­tra Re­main voter. (And still plenty enough to jus­tify vot­ing even af­ter ma­jor shrink­age by the un­cer­tainty that poli­cies will turn out/​be im­ple­mented as ex­pected.)

  Com­pli­cated by the fact that if Re­main hap­pens, Leave sup­port­ers would always feel things would have been bet­ter if Leave had hap­pened, even if their pre­dic­tions were un­know­ingly false, be­cause they never get to try out & com­pare both sce­nar­ios. I.e. Leavers will never be satis­fied if the UK re­mains, and Re­main­ers will never be satis­fied if the UK leaves, re­gard­less of how the other pos­si­ble world would have been. (Maybe that’s your main point here.) But I reckon that dis­satis­fac­tion is small com­pared with the eco­nomic harm caused by leav­ing (if the me­dian GDP pre­dic­tions are true).

  By the way, I’m not con­vinced vot­ing is ra­tio­nal (hence I have never voted in my life), and be­lieved that it wasn’t, un­til the al­tru­ism calcu­la­tion oc­curred to me a year or so ago. My cur­rent sus­pi­cion is about the val­idity of mul­ti­ply­ing a very small prob­a­bil­ity by a very large benefit to get a jus­tifi­ca­tion; but I haven’t yet read/​thought of a strong ar­gu­ment against this.

  (PS Ah, you’re Ja­cob F—good to meet you! I en­joy your blog.)