My Mistakes

Things I have changed my mind about.
personal, philosophy, predictions, Bitcoin, survey, Bayes
2011-09-152018-02-08 in progress certainty: unlikely importance: 7

“One does not care to ac­knowl­edge the mis­takes of one’s youth.”

, 1

It is salu­tary for the soul to re­view past events and per­haps keep a list of things one no longer be­lieves, since such crises are rare2 and so eas­ily pass from mem­ory (there is no feel­ing of be­ing wrong, only of hav­ing been wrong3). One does not need an elab­o­rate rit­ual (fun as they are to read about) to change one’s mind, but the changes must hap­pen. If you are not chang­ing, you are not grow­ing4; no one has won the be­lief lot­tery and has a mo­nop­oly on truth5. To the hon­est in­quir­er, all sur­prises are pleas­ant ones6.


Only the most clever and the most stu­pid can­not change.7

This list is not for spe­cific facts of which there are too many to record, nor is it for fal­si­fied pre­dic­tions like my be­lief that George W. Bush would not be elected (for those see or my Pre­dic­tion­Book.­com page), nor mis­takes in my pri­vate life (which go into a pri­vate file), nor things I never had an ini­tial strong po­si­tion on (Win­dows vs Lin­ux, Java vs Haskel­l). The fol­low­ing are some ma­jor ideas or sets of ideas that I have changed my mind about:


For I count be­ing re­futed a greater good, in­so­far as it is a greater good to be rid of the great­est evil from one­self than to rid some­one else of it. I don’t sup­pose that any evil for a man is as great as false be­lief about the things we’re dis­cussing right now…8

I think re­li­gion was the first sub­ject in my life that I took se­ri­ous­ly. As best as I can re­call at this point, I have no “de­con­ver­sion story” or tale to tell, since I don’t re­mem­ber ever se­ri­ously be­liev­ing9 - the sto­ries in the Bible or at my Catholic church were in­ter­est­ing, but they were ob­vi­ously fic­tion to some de­gree. I was­n’t go­ing to re­ject re­li­gion out of hand be­cause some of the sto­ries were made-up (any more than I be­lieved George Wash­ing­ton did­n’t ex­ist be­cause the story of him chop­ping down an ap­ple tree was made-up), but the big claims did­n’t seem to be pan­ning out ei­ther:

  1. My prayers re­ceived no an­swers of any kind, not even a voice in my head
  2. I did­n’t see any mir­a­cles or in­ter­ces­sions like I ex­pected from a om­nipo­tent lov­ing god

The lat­ter was prob­a­bly due to the car­toons I watched on TV, which seemed quite sen­si­ble to me: a pow­er­ful fig­ure like a god would act in all sorts of ways. If there re­ally was a god, that was some­thing that ought to be quite ob­vi­ous to any­one who ‘had eyes to see’. I had more ev­i­dence that Santa or China ex­isted than did God, which seemed back­wards. Ex­pla­na­tions for the ab­sence of di­vine ac­tion ranged from the strained to so lu­di­crously bad that they cor­roded what lit­tle faith I pos­sessed10. I would later rec­og­nize my own doubts in pas­sages of skep­ti­cal au­thors like Ed­ward Gib­bon and his De­cline:

…From the first of the fa­thers to the last of the popes, a suc­ces­sion of bish­ops, of saints, of mar­tyrs, and of mir­a­cles, is con­tin­ued with­out in­ter­rup­tion; and the progress of su­per­sti­tion was so grad­u­al, and al­most im­per­cep­ti­ble, that we know not in what par­tic­u­lar link we should break the chain of tra­di­tion. Every age bears tes­ti­mony to the won­der­ful events by which it was dis­tin­guished, and its tes­ti­mony ap­pears no less weighty and re­spectable than that of the pre­ced­ing gen­er­a­tion, till we are in­sen­si­bly led on to ac­cuse our own in­con­sis­ten­cy, if in the eighth or in the twelfth cen­tury we deny to the ven­er­a­ble Be­de, or to the holy Bernard, the same de­gree of con­fi­dence which, in the sec­ond cen­tu­ry, we had so lib­er­ally granted to Justin or to Ire­naeus. If the truth of any of those mir­a­cles is ap­pre­ci­ated by their ap­par­ent use and pro­pri­ety, every age had un­be­liev­ers to con­vince, heretics to con­fute, and idol­a­trous na­tions to con­vert; and suffi­cient mo­tives might al­ways be pro­duced to jus­tify the in­ter­po­si­tion of Heav­en. And yet, since every friend to rev­e­la­tion is per­suaded of the re­al­i­ty, and every rea­son­able man is con­vinced of the ces­sa­tion, of mirac­u­lous pow­ers, it is ev­i­dent that there must have been some pe­riod in which they were ei­ther sud­denly or grad­u­ally with­drawn from the Chris­t­ian church. What­ever aera is cho­sen for that pur­pose, the death of the apos­tles, the con­ver­sion of the Ro­man em­pire, or the ex­tinc­tion of the Ar­ian heresy, the in­sen­si­bil­ity of the Chris­tians who lived at that time will equally afford a just mat­ter of sur­prise. They still sup­ported their pre­ten­sions after they had lost their pow­er. Credulity per­formed the office of faith; fa­nati­cism was per­mit­ted to as­sume the lan­guage of in­spi­ra­tion, and the effects of ac­ci­dent or con­trivance were as­cribed to su­per­nat­ural caus­es. The re­cent ex­pe­ri­ence of gen­uine mir­a­cles should have in­structed the Chris­t­ian world in the ways of Prov­i­dence, and ha­bit­u­ated their eye (if we may use a very in­ad­e­quate ex­pres­sion) to the style of the di­vine artist…What­ever opin­ion may be en­ter­tained of the mir­a­cles of the prim­i­tive church since the time of the apos­tles, this un­re­sist­ing soft­ness of tem­per, so con­spic­u­ous among the be­liev­ers of the sec­ond and third cen­turies, proved of some ac­ci­den­tal ben­e­fit to the cause of truth and re­li­gion. In mod­ern times, a la­tent and even in­vol­un­tary scep­ti­cism ad­heres to the most pi­ous dis­po­si­tions. Their ad­mis­sion of su­per­nat­ural truths is much less an ac­tive con­sent than a cold and pas­sive ac­qui­es­cence.

I have seen these rea­sons mocked as sim­plis­tic and puerile, and I was cer­tainly aware that there were sub­tle ar­gu­ments which in­tel­li­gent philoso­phers be­lieved re­solved the (such as , which is valid but I do not con­sider it sound since it re­quires the mean­ing­less con­cept of free will) and that Chris­tians of var­i­ous stripes had var­i­ous com­pli­cated ex­pla­na­tions for why this world was con­sis­tent with there be­ing a God (if for no other rea­son than that I ob­served there were the­ists as in­tel­li­gent or more in­tel­li­gent than me). But the ba­sic con­cept seemed con­fused, free will was an even more du­bi­ous plank to go on, and in gen­eral the en­tire com­plex of his­tor­i­cal claims, meta­physics, and ac­tiv­i­ties of re­li­gious peo­ple did not seem con­vinc­ing. (Richard Car­ri­er’s 2011 Why I Am Not A Chris­t­ian ex­presses the gen­eral tenor of my mis­giv­ings, es­pe­cially after I checked out every­thing the li­brary had on , , the Gnos­tics, and early Chris­tian­ity - Je n’avais pas be­soin de cette hy­pothèse-là, ba­si­cal­ly.)

So I never be­lieved (although it was ob­vi­ous enough that there was no point in dis­cussing this since it might just lead to me go­ing to church more and sit­ting on the hard wooden pews), but there was still the trou­bling mat­ter of Heaven & Hell: those in­fini­ties meant I could­n’t sim­ply dis­miss re­li­gion and con­tinue read­ing about di­nosaurs or Al­ca­traz. If I got re­li­gion wrong, I would have got­ten lit­er­ally the most im­por­tant pos­si­ble thing wrong! Noth­ing else was as im­por­tant - if you’re wrong about a round earth, at worst you will never be a good ge­o­g­ra­pher or as­tronomer; if you’re wrong about be­liev­ing in as­trol­o­gy, at worst you waste time and mon­ey; if you’re wrong about evo­lu­tion and bi­ol­o­gy, at worst you en­dan­ger your life; and so on. But if you’re wrong about re­li­gion, wast­ing your life is about the least of the con­se­quences. And every­one ac­cepts a re­li­gion or at least the le­git­i­macy of re­li­gious claims, so it would be un­speak­ably ar­ro­gant of a kid to dis­miss re­li­gion en­tirely - that sort of ev­i­dence is sim­ply not there1112. (Oddly enough, athe­ists - who are not im­me­di­ately shown to be mis­taken or fools - are even rarer in books and car­toons than they are in real life.)

Kids ac­tu­ally are kind of skep­ti­cal if they have rea­son to be skep­ti­cal, and like­wise will be­lieve all sorts of strange things if the source was pre­vi­ously trust­wor­thy13. This is as it should be! Kids can­not come prewired with 100% cor­rect be­liefs, and must be able to learn all sorts of strange (but true) things from re­li­able au­thor­i­ties; these strate­gies are ex­actly what one would ad­vise. It is not their fault that some of the most re­li­able au­thor­i­ties in their lives (their par­ents) are mis­taken about one ma­jor set of be­liefs. They sim­ply have bad epis­temic luck.

So I read the Bible, which veered from bor­ing to in­co­her­ent to dis­gust­ing. (I be­came a fan of the , how­ev­er, and still pe­ri­od­i­cally read the Book of Job, Ec­cle­si­astes, and Proverb­s.) That did­n’t help much. Well, maybe Chris­tian­ity was not the right re­li­gion? My el­e­men­tary school li­brary had a rather strange se­lec­tion of books which in­cluded var­i­ous East­ern texts or an­tholo­gies (I re­mem­ber in par­tic­u­lar one an­thol­ogy on med­i­ta­tion, which was a hodge-podge of re­li­gious in­struc­tion man­u­als, es­says, and sci­en­tific stud­ies on med­i­ta­tion - that took me a long time to read, and it was only in high school and col­lege that I re­ally be­came com­fort­able read­ing psy­chol­ogy pa­per­s). I con­tin­ued read­ing in this vein for years, in be­tween all my more nor­mal read­ings. The Ko­ran was in­ter­est­ing and in gen­eral much bet­ter than the Bible. Shinto texts were worth­less mythol­o­giz­ing. Tao­ism had some very good early texts (the Chuang-tzu in par­tic­u­lar) but then bizarrely de­gen­er­ated into alche­my. Bud­dhism was strange: I rather liked the gen­eral philo­soph­i­cal ap­proach, but there were many pop­ulist el­e­ments in Ma­hayana texts that both­ered me. Hin­duism had a strange beau­ty, but my re­ac­tion was sim­i­lar to that of the early trans­la­tors, who con­demned it for sloth and las­si­tude. I also con­sid­ered the Oc­cult se­ri­ously and be­gan read­ing the Skep­ti­cal lit­er­a­ture on that and re­lated top­ics (see the later sec­tion).

By this point in my read­ing, I had reached mid­dle school; this sum­mary makes my read­ing sound more sys­tem­atic than it was. I still had­n’t found any es­pe­cially good rea­son to be­lieve in God or any gods, and had a jaun­diced view of many texts I had read. was a shock when I fi­nally be­came ca­pa­ble of read­ing it and source-texts like Jose­phus: it’s amaz­ing just how un­cer­tain, vari­able, self­-con­tra­dic­to­ry, edit­ed, and his­tor­i­cally in­con­sis­tent both the Old and New Tes­ta­ments are. There are hun­dreds of ma­jor vari­ants of the var­i­ous books, there are count­less thou­sands of tex­tual vari­ants (many of key the­o­log­i­cal pas­sages) leav­ing traces of ide­o­log­i­cal fab­ri­ca­tion through­out be­sides the ca­sual fal­si­fi­ca­tion of many his­tor­i­cal events for dra­matic effect or faked co­her­ence with Old Tes­ta­ment ‘prophe­cies’ (we all know of false claims like the Mas­sacre of the In­no­cents or Je­sus be­ing born of a ‘vir­gin’ to fit an er­ro­neous trans­la­tion or the sun stop­ping or the Tem­ple veil be­ing ripped or the Ro­man cen­sus) but it’s dra­matic to find that Je­sus was an ut­ter no­body even in Jose­phus, where the only men­tion of Je­sus seems to be fal­si­fied (by Chris­tians of course) de­spite him name-drop­ping con­stantly - and speak­ing of Jose­phus, it’s hard not to be im­pressed that while one re­cites every Sun­day how Je­sus was “cru­ci­fied un­der Pon­tius Pi­late” as proof we are told of Je­sus’s his­toric­ity & that he was not a story or mythol­o­gy, the Pon­tius Pi­late in Jose­phus is a cor­rupt mer­ci­less Ro­man offi­cial who does­n’t hes­i­tate to get his hands bloody if nec­es­sary and who does­n’t match the Gospel story in the slight­est; in­deed, read­ing through Jose­phus’s ac­counts of con­stant tur­moil in Jerusalem of var­i­ous false prophets and mes­si­ahs and rebels (h­m­m….) and the diffi­cul­ties the au­thor­i­ties face, one can’t be­lieve the en­tire story of Pi­late be­cause to not ei­ther im­me­di­ately ex­e­cute Je­sus or leave the whole mat­ter un­til well after the very po­lit­i­cally sen­si­tive hol­i­day is to as­sume both the Ro­man and Jew­ish offi­cials suffered a sud­den at­tack of the stu­pids and a col­lec­tive am­ne­sia of how they usu­ally dealt with such prob­lems. The whole story is bla­tant rub­bish! Yet my re­li­gion teach­ers kept oc­ca­sion­ally em­pha­siz­ing how Je­sus was a his­tor­i­cal fig­ure. The mythi­cist case is not com­pelling but they do a good job of show­ing how el­e­ments of the story were stan­dard tropes for the an­cient world, with many parts of the nar­ra­tive hav­ing mul­ti­ple prece­dents or weird­ly-in­ter­preted Old Tes­ta­ment jus­ti­fi­ca­tions. In short, read­ing through higher Bib­li­cal crit­i­cism, it’s not a sur­prise that it is such anath­ema to mod­ern Chris­t­ian sects and scrip­tural in­errancy de­vel­oped in re­ac­tion to it.

At some point, I shrugged and gave up and de­cided I was an athe­ist14 be­cause cer­tainly I felt noth­ing15. The­ol­ogy was in­ter­est­ing to some ex­tent, but there were bet­ter things to read about. (My lit­er­ary in­ter­est in Tao­ism and philo­soph­i­cal in­ter­est in Bud­dhism re­main, but I put no stock in any su­per­nat­ural claims they might make.)

The American Revolution

In mid­dle school, we were as­signed a pro-con de­bate about the Amer­i­can Rev­o­lu­tion; I hap­pened to be on the pro side, but as I read through the ar­gu­ments, I be­came in­creas­ingly dis­turbed and even­tu­ally de­cided that the pro-Rev­o­lu­tion ar­gu­ments were weak or fal­la­cious.

The Rev­o­lu­tion was a blood­bath with ~100,000 ca­su­al­ties or fa­tal­i­ties fol­lowed by 62,000 Loy­al­ist/­Tory refugees for fear of re­tal­i­a­tion and their ex­pro­pri­a­tion (the ones who stayed did not es­cape per­se­cu­tion); this is a butcher’s bill that did not seem jus­ti­fied in the least by any­thing in Britain or Amer­i­ca’s sub­se­quent his­tory (what, were the British go­ing to ran­domly mas­sacre Amer­i­cans for fun?), even now with a pop­u­la­tion of >300 mil­lion, and much less back when the pop­u­la­tion was 1/100th the size. In­de­pen­dence was granted to sim­i­lar Eng­lish colonies at the smaller price of “wait­ing a while”: Canada was es­sen­tially au­tonomous by 1867 (less than a cen­tury lat­er) and Aus­tralia was first set­tled in 1788 with au­tonomous colonies not long be­hind and the cur­rent Com­mon­wealth formed by 1901. (Nor did Canada or Aus­tralia suffer worse at Eng­land’s hands dur­ing the wait­ing pe­riod than, say, Amer­ica in that time suffered at its own hand­s.) In the long run, in­de­pen­dence may have been good for the USA, but this would be due to sheer ac­ci­dent: the British were hold­ing the fron­tier at the Ap­palachi­ans (see ), and Napoleon likely would not have been will­ing en­gage in the with Eng­lish colonies inas­much as he was at war with Eng­land. (As­sum­ing we see this as a good thing: Bryan Ca­plan de­scribes that as re­mov­ing “the last real check on Amer­i­can ag­gres­sion against the In­di­ans”.)

Nei­ther of these is a very strong ar­gu­ment; the British could eas­ily have re­voked the Procla­ma­tion in face of the colo­nial re­sis­tance (and in prac­tice did16), and Napoleon could not hold onto New France for very long against the British fleets. The ar­gu­ment from ‘free­dom’ is a buzz­word or un­sup­ported by the facts - Canada and Aus­tralia are hardly hell­hole bas­tions of to­tal­i­tar­i­an­ism, and are ranked by as be­ing as free as the USA. (Steve Sailer asks “Yet how much real differ­ence did the very differ­ent po­lit­i­cal paths of Amer­ica and Canada make in the long run?”; could we have been Canada?)

And there are im­por­tant ar­gu­ments for the op­po­site, that Amer­ica would have been bet­ter off un­der British rule - Britain very early on and likely would have ended slav­ery in the colonies as well. (Some have ar­gued that with con­tin­ued con­trol of the south­ern colonies, Britain would have not been able to do this; but the usual ar­gu­ments for the Rev­o­lu­tion cen­ter on the tyranny of Britain - so was the dog wag­ging the tail or the tail the dog?) The South cru­cially de­pended on Eng­land’s tacit sup­port (see­ing the South as a coun­ter­weight to the dan­ger­ous North?), so the would ei­ther never have started or have been sup­pressed very quick­ly. The Civil War would also have lacked its in­tel­lec­tual jus­ti­fi­ca­tion of if the states had re­mained Crown colonies. The Civil War was so bloody and de­struc­tive17 that avoid­ing it is worth a great deal in­deed. And then there comes WWI and WWII. It is not hard to see how Amer­ica re­main­ing a colony would have been bet­ter for both Eu­rope and Amer­i­ca.

Aside from the bet­ter out­comes for slaves and In­di­ans, it’s been sug­gested that Amer­ica would have ben­e­fited from main­tain­ing a par­lia­men­tary con­sti­tu­tion­al-monar­chy democ­racy rather than in­vent­ing its par­tic­u­lar pres­i­den­t-ori­ented re­pub­lic (a view that has some more ap­peal in the 2000s, but is more broadly sup­ported by the pop­u­lar­ity of par­lia­men­tary democ­ra­cies glob­ally and their ap­par­ent greater sta­bil­ity & suc­cess com­pared to the more Amer­i­can-style sys­tems in un­sta­ble & coup-prone Latin Amer­i­ca).

Since that par­a­digm shift in mid­dle school, my view has changed lit­tle:

  • Crane Brin­ton’s con­firmed my be­liefs with sta­tis­tics about the eco­nomic class of par­tic­i­pants: naked fi­nan­cial self­-in­ter­est is not a very con­vinc­ing ar­gu­ment for plung­ing a coun­try into war, given that Eng­land had in­curred sub­stan­tial debt de­fend­ing and ex­pand­ing the colonies and their tax bur­den - that they end­lessly com­plained of - was com­i­cally tiny com­pared to Eng­land prop­er. One of the in­ter­est­ing points Brin­ton makes was that con­trary to the uni­ver­sal be­lief, rev­o­lu­tions do not uni­ver­sally tend to oc­cur at times of poverty or in­creas­ing wealth in­equal­i­ty; in­deed, be­fore the Amer­i­can rev­o­lu­tion, the colonists were less taxed, wealth­ier & more equal than the Eng­lish.

  • con­tin­u­ing the eco­nomic the­me, the bur­dens on the Amer­i­can colonists such as the are now con­sid­ered to not be bur­den­some at all, but neg­li­gi­ble or pos­i­tive, es­pe­cially com­pared to in­de­pen­dence. Famed Scot­tish econ­o­mist Adam Smith sup­ported the Nav­i­ga­tion Acts as a crit­i­cal part of the Em­pire’s de­fense18 (which in­cluded the Amer­i­can colonies; but see again the colonies’ grat­i­tude for the French-In­dian War). Their light bur­den has be­come eco­nomic his­tory con­sen­sus since the dis­cus­sion was sparked in the 1960s (eg. Thomas 1965, Thomas 1968): in 1994, 198 eco­nomic his­to­ri­ans were sur­veyed asked sev­eral ques­tions on this point find­ing that:

    1. 132 dis­agreed with the propo­si­tion “One of the pri­mary causes of the Amer­i­can Rev­o­lu­tion was the be­hav­ior of British and Scot­tish mer­chants in the 1760s and 1770s, which threat­ened the abil­i­ties of Amer­i­can mer­chants to en­gage in new or even tra­di­tional eco­nomic pur­suits.”
    2. 178 agreed or par­tially agreed that “The costs im­posed on the colonists by the trade re­stric­tions of the Nav­i­ga­tion Acts were small.”
    3. 111 dis­agreed that “The eco­nomic bur­den of British poli­cies was the spark to the Amer­i­can Rev­o­lu­tion.”
    4. 117 agreed or par­tially agreed that “The per­sonal eco­nomic in­ter­ests of del­e­gates to the Con­sti­tu­tional Con­ven­tion gen­er­ally had a [sub­stan­tial] effect on their vot­ing be­hav­ior.”
  • Men­cius Mold­bug dis­cussed good deal of pri­mary source ma­te­r­ial which sup­ported my in­ter­pre­ta­tion.

    I par­tic­u­larly en­joyed his de­scrip­tion of the Pulitzer-win­ning , a study of the pop­u­lar cir­cu­lars and es­says (of which Thomas Paine’s is only the most fa­mous): the au­thor finds that the rebels and their lead­ers be­lieved there was a con­spir­acy by Eng­lish elites to strip them of their free­doms and crush the Protes­tants un­der the yoke of the .

    Bai­lyn points out that no traces of any such con­spir­acy has ever been found in the di­aries or mem­o­ran­dums or let­ters of said elites. Hence the Found­ing Fa­thers were, as Mold­bug claimed, ex­actly anal­o­gous to or . Mold­bug fur­ther points out that re­al­ity has di­rectly con­tra­dicted their pre­dic­tions, as both the Monar­chy and Church of Eng­land have seen their power con­tin­u­ously de­creas­ing to their pre­sen­t-day cer­e­mo­nial sta­tus, a diminu­tion in progress long be­fore the Amer­i­can Rev­o­lu­tion.

  • Pos­si­bly on Mold­bug’s ad­vice, I then read vol­ume 1 of Mur­ray Roth­bard’s . I was unim­pressed. Roth­bard seems to think he is jus­ti­fy­ing the Rev­o­lu­tion as a no­ble lib­er­tar­ian thing (ex­cept for those other scoundrels who just want to take over); but all I saw were scoundrels.

  • At­tempt­ing to take an out­side view and ig­nore the cult built up around the Found­ing Fa­thers, view­ing them as a cyn­i­cal for­eigner might, the Fa­thers do not nec­es­sar­ily come off well.

    For ex­am­ple, one can com­pare George Wash­ing­ton to : both led a guer­rilla rev­o­lu­tion of British colonies against the coun­try which had built their colony up into a wealthy re­gional pow­er­house, and they or their al­lies em­ployed mobs and ter­ror­ist tac­tics; both over­saw hy­per­in­fla­tion of their cur­ren­cy; both ex­pro­pri­ated po­lit­i­cally dis­fa­vored groups, and en­gaged in give-aways to sup­port­ers (Mu­gabe re­dis­trib­uted land to black sup­port­ers, Wash­ing­ton ap­proved of states’ war-debts - an in­cred­i­ble wind­fall for the Hamil­ton-con­nected spec­u­la­tors, who sup­ported the ); both were over­whelm­ingly voted into office and com­manded mass pop­u­lar­ity even after ma­jor fail­ures of their poli­cies be­came ev­i­dent (e­co­nomic growth & hy­per­in­fla­tion for Mu­gabe, the Whiskey Re­bel­lion for Wash­ing­ton), be­ing hailed as fa­thers of their coun­tries; and both wound up one of, if not the, most wealthy men in the coun­try (Mu­gabe’s for­tune has been es­ti­mated at any­where from $3b to $10b; Wash­ing­ton, in in­fla­tion-ad­justed terms, has been es­ti­mated at $0.5b).

  • Je­remy Ben­tham amus­ingly evis­cer­ates the In­de­pen­dence’s com­plaints


In roughly mid­dle school as well, I was very in­ter­ested in eco­nomic in­jus­tice and guer­rilla war­fare, which nat­u­rally led me straight into the com­mu­nist lit­er­a­ture. I grew out of this when I re­al­ized that while I might not be able to pin­point the prob­lems in com­mu­nism, a lot of that was due to the sheer ob­scu­rity and bull­shit­ting in the lit­er­a­ture (I fi­nally gave up after read­ing twice, con­clud­ing the prob­lem was not me, Marx­ism was re­ally that in­tel­lec­tu­ally worth­less), and the prac­ti­cal re­sults with economies & hu­man lives spoke for them­selves: the ideas were tried in so many coun­tries by so many groups in so many differ­ent cir­cum­stances over so many decades that if there were any­thing to them, at least one coun­try would have suc­ceed­ed. In com­par­ison, even with the broad­est sam­ple in­clud­ing hell­holes like the Bel­gian Con­go, cap­i­tal­ism can still point to suc­cess sto­ries like Japan.

(Sim­i­lar ar­gu­ments can be used for sci­ence and re­li­gion: after early sci­ence got the ba­sic in­duc­tive em­pir­i­cal for­mula right, it took off and within 2 or 3 cen­turies had con­quered the in­tel­lec­tual world and as­sisted the con­quest of much of the real world too; in con­trast, 2 or 3 cen­turies after Chris­tian­ity be­gan, its texts were be­gin­ning to fi­nally con­geal into the be­gin­nings of a canon, it was mi­nor, and the Ro­mans were still mak­ing oc­ca­sional efforts to ex­ter­mi­nate this irk­some re­li­gion. Charles Mur­ray, in a book I oth­er­wise ap­prove of, at­tempts to ar­gue in Hu­man Ac­com­plish­ment that Chris­tian­ity was a key fac­tor in the great ac­com­plish­ments of West­ern sci­ence & tech­nol­ogy by some gib­ber­ish in­volv­ing hu­man dig­ni­ty; the ar­gu­ment is in­trin­si­cally ab­surd - Greek as­tron­omy and phi­los­o­phy were ac­tive when Chris­tian­ity start­ed, St. Paul lit­er­ally de­bated the Greek philoso­phers in Athens, and yet Chris­tian­ity did not spark any rev­o­lu­tion in the 100s, or 200s, or 300s, or for the next mil­len­ni­um, nor the next mil­len­nium and a half. It would lit­er­ally be fairer to at­tribute sci­ence to William the Con­queror, be­cause that’s a gap one-third the size and there’s at least a di­rect line from William the Con­queror to the Royal So­ci­ety! If we try to be fairer and say it’s late Chris­tian­ity as ex­em­pli­fied by the phi­los­o­phy of Thomas Aquinas - as in­flu­enced by non-Chris­t­ian thought like Aris­to­tle as it is - that still leaves us a gap of some­thing like 300-500 years. Let us say I would find Mur­ray’s ar­gu­ment of more in­ter­est if it were com­ing from a non-Chris­tian…)

The Occult

This is not a par­tic­u­lar er­ror but a whole class of them. I was sure that the over­all the­is­tic ex­pla­na­tions were false, but surely there were real phe­nom­e­non go­ing on? I’d read up on in­di­vid­ual things like Nos­tradamus’s prophe­cies or the Lance of Long­i­nus, check the skep­tics lit­er­a­ture, and dis­be­lieve; rinse and re­peat un­til I fi­nally dis­miss the en­tire area with some ex­cep­tions like the men­tal & phys­i­cal ben­e­fits of med­i­ta­tion. One might say my ex­pe­ri­ence was a lit­tle like ca­reer as re­counted in “The Elu­sive Open Mind: Ten Years of Neg­a­tive Re­search in Para­psy­chol­ogy”, sans the de­tailed ex­per­i­ments. (I am still an­noyed that I was un­able to dis­be­lieve the re­search on un­til I read more about the cor­rup­tion, de­cep­tion, and fal­si­fied pre­dic­tions of the TM or­ga­ni­za­tion it­self.) For­tu­nate­ly, I had ba­si­cally given up on oc­cult things by high school, be­fore I read Eco’s , so I don’t feel too cha­grined about this.


I spend most of my time read­ing; I also spent most of my time in el­e­men­tary, mid­dle, and high school read­ing. What has changed in what I read - I now read prin­ci­pally non­fic­tion (phi­los­o­phy, eco­nom­ics, ran­dom sci­ences, etc.), where I used to read al­most ex­clu­sively fic­tion. (I would in­clude one non­fic­tion book in my stacks of books to check out, on a sort of ‘veg­eta­bles’ ap­proach. Eat your veg­eta­bles and you can have dessert.) I, in fact, as­pired to be a nov­el­ist. I thought fic­tion was a no­ble task, the high­est pro­duc­tion of hu­man­i­ty, and writ­ers some of the best peo­ple around, pro­duc­ing im­mor­tal works of truth. Slowly this changed. I re­al­ized fic­tion changed noth­ing, and when it did change things, it was as oft as not for the worse. Fic­tion pro­moted sim­pli­fi­ca­tion, fo­cus on sym­pa­thetic ex­am­ples, and I rec­og­nized how much of my own in­fat­u­a­tion with the Oc­cult (a­mong other er­rors) could be traced to fic­tion. What a strange be­lief, that you could find truths in lies.19 And there are so many of them, too! So very many. (I wrote one es­say on this top­ic, .) I still pro­duce some fic­tion these days, but mostly when I can’t help it or as a writ­ing ex­er­cise.


I changed my mind about in 2011. I had nat­u­rally as­sumed, in line with the usual Amer­i­can cul­tural mes­sages, that there was noth­ing good about to­bacco and that smok­ing is deeply shame­ful, prov­ing that you are a selfish lazy short­-sighted per­son who is happy to com­mit slow sui­cide (tak­ing oth­ers with him via sec­ond-hand smoke) and cost so­ci­ety a for­tune in med­ical care. Then some men­tions of nico­tine as use­ful came up and I be­gan re­search­ing it. I’m still not a fan of smok­ing, and I re­gard any to­bacco with deep trep­i­da­tion, but the re­search lit­er­a­ture seems pretty clear: nico­tine en­hances men­tal per­for­mance in mul­ti­ple do­mains and may have some mi­nor health ben­e­fits to boot. Nico­tine sans to­bacco seems like a clear win. (It amuses me that of the changes listed here, this is prob­a­bly the one peo­ple will find most re­volt­ing and bizarre.)

Centralized darknet-markets

I over­es­ti­mated the sta­bil­ity of Bit­coin+­Tor dark­net mar­kets such as : I was aware that the cen­tral­iza­tion of the first-gen­er­a­tion DNMs (SR/BMR/Atlantis/Sheep) meant that the site op­er­a­tors had a strong temp­ta­tion to steal all de­posit & es­crows, but I thought that the value of fu­ture es­crow com­mis­sions pro­vided enough in­cen­tive to make rip-and-run scams rare - cer­tainly they were fairly rare dur­ing the Silk Road 1 era.

After Silk Road was shut down in Oc­to­ber 2013, SR turned out to be highly un­usu­al: both less hacked than most mar­kets, and it seems that what­ever his (many) other fail­ings, Ross Ul­bricht gen­uinely be­lieved his own ide­ol­ogy and so was run­ning Silk Road out of prin­ci­ple rather than greed (which also ex­plains why he did­n’t re­tire de­spite a for­tune larger than he could spend in a life­time). At­tracted by the sud­den void in a large mar­ket, and by the FBI’s press re­leases crow­ing over how many hun­dreds of mil­lions of dol­lars Silk Road had earned, dozens of new mar­kets sprang up to fill the void. Many then pro­ceeded to scam users (often tak­ing ad­van­tage of the stan­dard ‘seller bonds’: sell­ers would de­posit a large sum as a guar­an­tee against scam­ming buy­ers in the early pe­riod where they were ac­cept­ing or­ders but most pack­ages would not have ar­rived) or al­ter­nate­ly, be hacked due to the op­er­a­tors’ get-rich-quick in­com­pe­tence and rather than re­fund users from fu­ture profits, de­cide to steal every­thing the hacker did­n’t get. As of April 2014, it seems users have mostly learned cau­tion, and the shift to mul­ti­sig es­crow re­moves the need to trust mar­ket op­er­a­tors and hence the risk from the op­er­a­tors or hack­ers, so mat­ters may fi­nally be sta­bi­liz­ing.

I think my orig­i­nal point is still cor­rect that mar­kets can be trusted as long as the dis­counted present value of their fu­ture earn­ings ex­ceeds the amount they can steal. My mis­take here was over­es­ti­mat­ing the net present val­ue: I did­n’t re­al­ize that site op­er­a­tors had such high dis­count rates (one, PBF, pulled its scam after per­haps a few thou­sand dol­lars’ worth of Bit­coin had been de­posited de­spite pos­i­tive ini­tial re­views) and there was so much risk in­volved (the Bit­coin ex­change rate, ar­rest, hack­ing; all ex­ac­er­bated by the in­com­pe­tence of many site op­er­a­tors).

This mis­take lead to com­pla­cency on my part in archiv­ing the mar­kets & fo­rums: if you ex­pect a mar­ket to be around for years, there is no par­tic­u­lar need to try to mir­ror them week­ly. And so while I have good cov­er­age of the DNMs post-De­cem­ber-2013, I am miss­ing most of the mar­kets be­fore then.

Potential changes

The mind can­not fore­see its own ad­vance.20

There are some things I used to be cer­tain about, but I am no longer cer­tain ei­ther way; I await fu­ture de­vel­op­ments which may tip me one way or the oth­er.

Near Singularity

I am no longer cer­tain that is near.

In the 1990s, all the num­bers seem to be ever-ac­cel­er­at­ing. In­deed, I could feel with Kurzweil that . But an odd thing hap­pened in the 2000s (a dreary decade, dis­tracted by the dual dis­si­pa­tion of Afghanistan & Iraq). The hard­ware kept get­ting bet­ter mostly in line with Moore’s Law (trou­bling as the flight to par­al­lelism is), but the AI soft­ware did­n’t seem to keep up. I am only a lay­man, but it looks as if all the AI ap­pli­ca­tions one might cite in 2011 as progress are just old al­go­rithms now prac­ti­cal with newer hard­ware. And eco­nomic growth slowed down, and the stock mar­ket ticked along, barely main­tain­ing it­self. The Hu­man Genome Project com­pletely fiz­zled out, with in­ter­est­ing in­sights and not much else. (It’s great that genome se­quenc­ing has im­proved ex­actly as promised, but what about every­thing else? Where are our em­bryo se­lec­tions, our ger­m-line en­gi­neer­ing, our uni­ver­sal ge­netic ther­a­pies, our cus­tomized drugs?21) The phar­ma­ceu­ti­cal in­dus­try has reached such di­min­ish­ing re­turns that even the op­ti­mists have no­ticed the prob­lems in the drug pipeline, prob­lems so se­vere that it’s hard to wave them away as due to that drat­ted FDA or ig­no­rant con­sumers. As of 2007, the in­creases in longevity for the el­derly22 in the US has con­tin­ued to be less each year and , which is­n’t good news for those hop­ing to reach “es­cape ve­loc­ity”; and med­i­cine has been a re­peated dis­ap­point­ment even to fore­cast­ing-savvy pre­dic­tors (the ’90s and the ge­netic rev­o­lu­tion be­ing es­pe­cially re­mark­able for its lack of con­crete im­prove­ments). Kurzweil pub­lished an eval­u­a­tion of his pre­dic­tions up to ~2009 with great fan­fare and self­-con­grat­u­la­tion, but read­ing through them, I was struck by how many he weaseled out on (claim­ing as a hit any­thing that ex­isted in a lab or a mi­cro­scopic mar­ket seg­ment, even though in con­text he had clearly ex­pected it to be wide­spread) and how often they failed due to un­in­tel­li­gent soft­ware.

And there are many trou­bling long-term met­rics. I was deeply trou­bled to read ’s point­ing out a long-term de­cline in dis­cov­er­ies per capita (de­spite ever in­creas­ing sci­en­tists and artists per cap­i­ta!), even after he cor­rected for every­thing he could think of. I did­n’t see any ob­vi­ous mis­takes. twisted the knife fur­ther, and then I read Joseph Tain­ter’s The Col­lapse of Com­plex So­ci­eties. I have kept notes since and see lit­tle rea­son to ex­pect a gen­eral ex­po­nen­tial up­wards over all fields, in­clud­ing the ones min­i­mally con­nected to com­put­ing. ( “The End of the Fu­ture” makes a dis­tinc­tion be­tween “the progress in com­put­ers and the fail­ure in en­ergy”; he also makes an in­ter­est­ing link be­tween the lack of progress and the many re­cent spec­u­la­tive bub­bles in “The Op­ti­mistic Thought Ex­per­i­ment”.) The Sin­gu­lar­ity is still more likely than not, but these days, I tend to look to­wards em­u­la­tion of hu­man brains via scan­ning of as the cause. Whole brain em­u­la­tion is not likely for many decades, given the ex­treme com­pu­ta­tional de­mands (even if we are op­ti­mistic and take the Whole Brain Em­u­la­tion Roadmap fig­ures, one would not ex­pect a up­load un­til the 2030s) and it’s not clear how use­ful an up­load would be in the first place. It seems en­tirely pos­si­ble that the mind will run slow­ly, be able to self­-mod­ify only in triv­ial ways, and in gen­eral be a cu­rios­ity akin to the Space Shut­tle than a piv­otal mo­ment in hu­man his­tory de­serv­ing of the ti­tle Sin­gu­lar­i­ty.


Less­Wrong dis­cus­sion

I re­spect my own opin­ion, but at the same time I know I am not im­mune to com­mon be­liefs; so it both­ers me to see ‘stag­na­tion’ and pes­simistic ideas be­come more wide­spread be­cause this means I may just be fol­low­ing a trend. I did not like agree­ing with any of Wired’s hy­per­bolic fore­casts back in the 1990s, and I do not like agree­ing with Pe­ter Thiel or Neal Stephen­son now. One of Buffet’s clas­sic say­ings is “if they [in­vestors] in­sist on try­ing to time their par­tic­i­pa­tion in eq­ui­ties, they should try to be fear­ful when oth­ers are greedy and greedy when oth­ers are fear­ful.” What grounds do I have for be­ing ‘greedy’ now, when many are be­ing ‘fear­ful’? What Kah­ne­man-style pre-mortem would I give for ex­plain­ing why the Sin­gu­lar­ity might in­deed be Near?

First, one could point out that a num­ber of tech­no­log­i­cal mile­stones seem to be catch­ing up, after long stag­na­tions. From 2009-2012, there were a num­ber of un­ex­pected achieve­ments: Google’s ro­botic car as­tounded me, the long AI-re­sis­tant game of is falling to tech­niques are clos­ing in the Go mas­ters (I ex­pect com­put­ers to take world cham­pion by 2030), on­line ed­u­ca­tion seems to be start­ing to re­al­ize its promise (eg. the suc­cess of ), pri­vate space ex­ploita­tion is do­ing sur­pris­ingly well (as are Tesla elec­tric cars, which seem to be mov­ing from play­things to per­haps mass mar­ket cars), smart­phones - after a decade of be­ing crip­pled by tele­coms and lim­ited com­put­ing power - are be­com­ing ubiq­ui­tous and desk­top re­place­ments. Old dreams like and on­line have swung into ac­tion and given rise to ac­tive & grow­ing com­mu­ni­ties. In the larger pic­ture, the Long De­pres­sion be­gin­ning in 2008 has wreaked havoc on young peo­ple, but China has not im­ploded while con­tin­u­ing to move up the qual­ity chain (re­plac­ing la­bor­ers with ro­bots) and far from be­ing ‘the cri­sis of cap­i­tal­ism’ or ‘the end of cap­i­tal­ism’, in gen­er­al, global life is go­ing on. Even Africa, while the pop­u­la­tion size is ex­plod­ing, is grow­ing eco­nom­i­cally - per­haps thanks to uni­ver­sally avail­able cheap cell­phones. Peak Oil con­tin­ues to be de­layed by new de­vel­op­ments like frack­ing and re­sul­tant gluts of nat­ural gas (the US now ex­ports en­er­gy!), al­though the long-term sci­en­tific pro­duc­tiv­ity trends seem to still point down­wards.

What many of these points have in com­mon is that their fore­bears ger­mi­nated for a long time in niches and did not live up to the fore­casts of their pro­po­nents - smart­phones, for ex­am­ple, have been ex­pected to rev­o­lu­tion­ize every­thing since at least the 1980s (by mem­bers of the in par­tic­u­lar). And in­deed, they are rev­o­lu­tion­iz­ing every­thing, world­wide, 30 years lat­er. This ex­em­pli­fies a line at­trib­uted to Roy Ama­ra:

We tend to over­es­ti­mate the effect of a tech­nol­ogy in the short run and un­der­es­ti­mate the effect in the long run.

A num­ber of cur­rent but dis­ap­point­ing trends may be dis­ap­point­ing only in the “short run”, the flat part of their re­spec­tive ex­po­nen­tial or sig­moid de­vel­op­ment curves23. For ex­am­ple, DNA se­quenc­ing has been plum­met­ing and se­quenc­ing a whole hu­man genome will likely be <$100 by 2015; this has been an in­cred­i­ble boon for ba­sic re­search and our knowl­edge of the world, but so far the ap­pli­ca­tions have been fairly min­i­mal - but this may not be true forever, with new projects start­ing up tack­ling top­ics of the great­est mag­ni­tude like us­ing thou­sands of genomes to search for the thou­sands of al­le­les which each affect in­tel­li­gence a tiny bit. Were em­bryo se­lec­tion for in­tel­li­gence to be­come vi­able (as there is no rea­son to be­lieve would not be pos­si­ble once the right al­le­les have been iden­ti­fied) and every baby could be born with IQs >130, so­ci­ety would change.

Does this ap­ply to AI? At least two of the ex­am­ples seem clear-cut ex­am­ples of an ‘Amara effect’: Go-play­ing AIs were for decades toys eas­ily beaten by bad am­a­teurs un­til Monte Carlo Tree Search was in­tro­duced in 200624 and then a decade after that they were su­per­hu­man, while the first in was a de­ba­cle in which no car fin­ished and the best car man­aged to travel a whop­ping 7 miles be­fore get­ting stuck on a rock. 8 years lat­er, and now the con­ver­sa­tion has sud­denly shifted from “will Go AIs ever reach hu­man lev­el?” or “will self­-driv­ing cars ever be able to cope with the real world?” to the sim­ple ques­tion when?.

One of the ironies is that I am sure a ‘pure’ AI is pos­si­ble; but the AI can’t be de­vel­oped be­fore the com­put­ing power is avail­able (we hu­mans are just not good enough at math & pro­gram­ming to achieve it with­out run­ning code), which means the AI will be de­vel­oped ei­ther si­mul­ta­ne­ous with or after enough com­put­ing power be­comes avail­able. If the lat­ter, if the AI is not run at the ex­act in­stant that there is enough pro­cess­ing power avail­able, ever more com­put­ing power in ex­cess of what is needed (by de­fi­n­i­tion) builds up. It is like a dry for­est roast­ing in the sum­mer sun: the longer the wait un­til the match, the faster and hot­ter the wild­fire will burn25. Per­haps para­dox­i­cal­ly, the longer I live with­out see­ing an AI of any kind, the more schiz­o­phrenic my fore­casts will ap­pear to an out­sider who has­n’t care­fully thought about the is­sue - I will pre­dict with in­creas­ingly high con­fi­dence that the fu­ture will be bor­ing and nor­mal (be­cause the con­tin­ued non-ap­pear­ance makes it in­creas­ingly likely AI is im­pos­si­ble, see & AI), that AI is more likely to be cre­ated in the next year (be­cause the pos­si­bil­i­ties are be­ing ex­hausted as time pass­es) and the changes I pre­dict be­come ever more rad­i­cal!

This set of es­ti­mates is ob­vi­ously con­sis­tent with an ap­pear­ance of stag­na­tion: each year small ad­vances build up, but no big break­throughs ap­pear - un­til they do.


“Al­most in the same way as ear­lier physi­cists are said to have found sud­denly that they had too lit­tle math­e­mat­i­cal un­der­stand­ing to be able to mas­ter physics; we may say that young peo­ple to­day are sud­denly in the po­si­tion that or­di­nary com­mon sense no longer suffices to meet the strange de­mands life makes. Every­thing has be­come so in­tri­cate that for its mas­tery an ex­cep­tional de­gree of un­der­stand­ing is re­quired. For it is not enough any longer to be able to play the game well; but the ques­tion is again and again: what sort of game is to be played now any­way?”

Wittgen­stein’s Cul­ture and Value, MS 118 20r: 27.8.1937

The idea of - per­ma­nent and a - used to be dis­missed con­temp­tu­ously as the . (There are mod­els where tech­nol­ogy does pro­duce per­ma­nent un­em­ploy­ment, and quite plau­si­ble ones too; see Au­tor et al 2003 and Au­tor & Hamil­ton26 and Krug­man’s com­men­tary point­ing to re­cent data show­ing the ‘hol­low­ing out’ and ‘deskilling’ pre­dicted by the Au­tor mod­el, which is also con­sis­tent with the long-term de­cline in teenage em­ploy­ment due to im­mi­gra­tion. Mar­tin Ford has some graphs ex­plain­ing the com­ple­men­ta­tion-sub­sti­tu­tion mod­el.) But ever since the In­ter­net bub­ble burst, it’s been look­ing more and more like­ly, with scads of ev­i­dence for it since the hous­ing bub­ble like the oth­er­wise pe­cu­liar changes in the value of col­lege de­grees27. (This is closely re­lated to my grounds for be­liev­ing in a dis­tant Sin­gu­lar­i­ty.) When I look around, it seems to me that we have been suffer­ing tremen­dous un­em­ploy­ment for a long time. When Alex Tabar­rok writes “If the Lud­dite fal­lacy were true we would all be out of work be­cause pro­duc­tiv­ity has been in­creas­ing for two cen­turies”, I think, is­n’t that cor­rect? If you’re not a stu­dent, you’re re­tired; if you’re not re­tired, you’re dis­abled28; if you’re not dis­abled, per­haps you are in­sti­tu­tion­al­ized; if you’re not that, maybe you’re on wel­fare, or just un­em­ployed.

Com­pare now to most of hu­man his­to­ry, or just the 1300s:

  • every kid in spe­cial ed would be out work­ing on the farm; there would, if only from re­duced 29 be fewer dis­abled than now (fed­eral alone sup­ports 8 mil­lion Amer­i­cans)

  • every­one in col­lege would be out work­ing (be­cause the num­ber of stu­dents was a round­ing er­ror and they did­n’t spend very long in higher ed­u­ca­tion to be­gin with)

    In­deed, ed­u­ca­tion and health­care are a huge chunk of the US econ­omy - and both have se­ri­ous ques­tions about how much good, ex­act­ly, they do and whether they are grotesquely in­effi­cient or just in­effi­cient.

  • re­tirees did­n’t ex­ist out­side the tiny no­bil­ity

  • ‘guard la­bor’ - peo­ple em­ployed solely to con­trol and en­sure pro­duc­tiv­ity of the oth­ers has in­creased sub­stan­tially (Bowles & Jayadev 2006 claim US guard la­bor has gone from 6% of the 1890 la­bor force to 26% in 2002; this is not due to man­u­fac­tur­ing de­clines30); ex­am­ples of guard labor:

    • stand­ing mil­i­taries were un­usual (although effec­tive when needed31); the US main­tains the in the world - ~1.5m (~0.5% of the pop­u­la­tion), which em­ploys mil­lions more with its $700 bil­lion bud­get32 and is a key source of pork33 and make-work
    • pris­ons were mostly for tem­po­rary in­car­cer­a­tion pend­ing trial or pun­ish­ment34; the US has ~2.3m (n­early 1% of the pop­u­la­tion!), and per­haps an­other 4.9m on parole/pro­ba­tion. (See also the re­la­tion­ship of psy­chi­atric im­pris­on­ment with crim­i­nal im­pris­on­men­t.) That’s im­pres­sive enough, but as with the mil­i­tary, con­sider how many peo­ple are tied down solely be­cause of the need to main­tain and sup­ply the prison sys­tem - prison war­dens, builders, po­lice etc.
  • peo­ple worked hard; the and 5-day work­week were ma­jor hard-fought changes (a plank of the !). Switch­ing from a 16-hour to an 8-hour day means we are half-re­tired al­ready and need many more work­ers than oth­er­wise.

In con­trast, Amer­i­cans now spend most of their lives not work­ing.

The un­em­ploy­ment rate looks good - 9% is surely a refu­ta­tion of the Lud­dite fal­la­cy! - un­til you look into the meat fac­tory and see that that is the best rate, for col­lege grad­u­ates ac­tively look­ing for jobs, and not the over­all pop­u­la­tion one in­clud­ing those who have given up. Econ­o­mist Alan Krueger writes of the ra­tio (which cov­ers only 15-64 year old­s):

Telling­ly, the has hardly budged since reach­ing a low of 58.2% in De­cem­ber 2009. Last month it stood at just 58.4%. Even in the ex­pan­sion from 2002 to 2007 the share of the pop­u­la­tion em­ployed never reached the peak of 64.7% it at­tained be­fore the March-No­vem­ber 2001 re­ces­sion.

Let’s break it down by age group us­ing :

La­bor force par­tic­i­pa­tion rate 1993–2013, by age groups: 25–54yo, 20–24yo, >55yo

cor­rectly points out that the em­ploy­men­t:pop­u­la­tion ra­tio it­self does­n’t in­trin­si­cally tell us about whether things are go­ing well or poorly - one could imag­ine a happy and highly au­to­mated coun­try with a where only 20% of the pop­u­la­tion works or an agri­cul­tural coun­try where every­one works and is des­per­ately poor. What mat­ters more is wealth in­equal­ity com­bined with em­ploy­ment ra­tio: how many peo­ple are ei­ther rich enough that not hav­ing a job is not a dis­as­ter or at least can get a job?

What do you sup­pose the em­ploy­men­t-to-pop­u­la­tion rate was in 1300 in the poorer 99% of the world pop­u­la­tion (re­mem­ber­ing how home­mak­ing and rais­ing chil­dren is effec­tively a ful­l-time job)? I’d bet it was a lot higher than the world record in 2005, Ice­land’s 84%. And Ice­land is a very brainy place. What are the merely av­er­age with IQs of 100-110 sup­posed to do? (Heck, what is the half of Amer­ica with IQs in that re­gion or be­low sup­posed to do? Learn C++ and sta­tis­tics so they can work on Wall Street?) If you want to see the fu­ture, look at our youth; where are sum­mer jobs these days? Gre­gory Clark com­ments sar­don­ically (although he was likely not think­ing of whole brain em­u­la­tion) in :

Thus, while in prein­dus­trial agrar­ian so­ci­eties half or more of the na­tional in­come typ­i­cally went to the own­ers of land and cap­i­tal, in mod­ern in­dus­tri­al­ized so­ci­eties their share is nor­mally less than a quar­ter. Tech­no­log­i­cal ad­vance might have been ex­pected to dra­mat­i­cally re­duce un­skilled wages. After all, there was a class of work­ers in the prein­dus­trial econ­omy who, offer­ing only brute strength, were quickly swept aside by ma­chin­ery. By 1914 most horses had dis­ap­peared from the British econ­o­my, swept aside by steam and in­ter­nal com­bus­tion en­gi­nes, even though a mil­lion had been at work in the early nine­teenth cen­tu­ry. When their value in pro­duc­tion fell be­low their main­te­nance costs they were con­demned to the knack­er’s yard.

Tech­nol­ogy may in­crease to­tal wealth un­der many mod­els, but there’s a key loop­hole in the idea of “Pare­to-im­prov­ing” gains—they don’t ever have to make some peo­ple bet­ter off. And a Pare­to-im­prove­ment is a good re­sult! Many mod­els don’t guar­an­tee even that - it’s per­fectly pos­si­ble to be­come worse off (see the horses above and the fate of hu­mans in ‘crack of a fu­ture dawn’ sce­nar­i­o). Such doc­tri­nairism is not use­ful:

“Like ex­perts in many fields who give pol­icy ad­vice, the au­thors show a pref­er­ence for first-best, text­book ap­proaches to the prob­lems in their field, while leav­ing other messy ob­jec­tives ac­knowl­edged but as­signed to oth­ers. In this way, they are much like those pub­lic fi­nance econ­o­mists who op­pose tax ex­pen­di­tures on prin­ci­ple, be­cause they pre­fer di­rect ex­pen­di­ture pro­grams, but do not re­ally an­a­lyze the var­i­ous diffi­cul­ties with such pro­grams; or like trade econ­o­mists who know that the losers from trade surges need to be pro­tected but re­gard this as not a prob­lem for trade pol­i­cy.” –, “Com­ments on ‘The Con­tra­dic­tion in Chi­na’s Grad­u­al­ist Bank­ing Re­forms’”, Brook­ings Pa­pers on Eco­nomic Ac­tiv­ity 2006, 2, 149-162

This is closely re­lated to what I’ve dubbed the “‘Lud­dite fal­lacy’ fal­lacy” (a­long the lines of the Pas­cal’s Wa­ger Fal­lacy Fal­lacy): tech­nol­o­gists who are ex­tremely in­tel­li­gent and have worked most of their life only with fel­low po­ten­tial con­fi­dently say that “if there is struc­tural un­em­ploy­ment (and I’m be­ing gen­er­ous in grant­ing you Lud­dites even this con­tention), well, bet­ter ed­u­ca­tion and train­ing will fix that!” It’s a lit­tle hard to ap­pre­ci­ate what a stu­pen­dous mix­ture of avail­abil­ity bi­as, in­fi­nite op­ti­mism, and plain de­nial of in­tel­li­gence differ­ences this all is. Marc An­dreessen offers an ex­am­ple in 2011:

Sec­ond­ly, many peo­ple in the U.S. and around the world lack the ed­u­ca­tion and skills re­quired to par­tic­i­pate in the great new com­pa­nies com­ing out of the soft­ware rev­o­lu­tion. This is a tragedy since every com­pany I work with is ab­solutely starved for tal­ent. Qual­i­fied soft­ware en­gi­neers, man­agers, mar­keters and sales­peo­ple in Sil­i­con Val­ley can rack up dozens of high­-pay­ing, high­-up­side job offers any time they want, while na­tional un­em­ploy­ment and un­der­em­ploy­ment is sky high. This prob­lem is even worse than it looks be­cause many work­ers in ex­ist­ing in­dus­tries will be stranded on the wrong side of soft­ware-based dis­rup­tion and may never be able to work in their fields again. There’s no way through this prob­lem other than ed­u­ca­tion, and we have a long way to go.

I see. So all we have to do with all the peo­ple with <120 IQs, who strug­gled with al­ge­bra and never made it to cal­cu­lus (when they had the self­-dis­ci­pline to learn it at al­l), is just to train them into world-class soft­ware en­gi­neers and man­agers who can sat­isfy Sil­i­con Val­ley stan­dards; and we have to do this for the first time in hu­man his­to­ry. Gosh, is that all? Why did­n’t you say so be­fore - we’ll get on that right away! Or an anony­mous “data sci­en­tist” recorded in the NYT: “He found my con­cerns to be amus­ing. Peo­ple can get work cre­at­ing SEO-optimized niche blogs, he said. Or they can learn to code.” Thomas Fried­man:

Every mid­dle-class job to­day is be­ing pulled up, out or down faster than ever. That is, it ei­ther re­quires more skill or can be done by more peo­ple around the world or is be­ing buried - made ob­so­lete - faster than ever. Which is why the goal of ed­u­ca­tion to­day, ar­gues Wag­n­er, should not be to make every child “col­lege ready” but “in­no­va­tion ready” - ready to add value to what­ever they do…­more than ever, our kids will have to “in­vent” a job. (For­tu­nate­ly, in to­day’s world, that’s eas­ier and cheaper than ever be­fore.) Sure, the lucky ones will find their first job, but, given the pace of change to­day, even they will have to rein­vent, re-engi­neer and reimag­ine that job much more often than their par­ents if they want to ad­vance in it… What does that mean for teach­ers and prin­ci­pals? [Tony Wag­ner:] “All stu­dents should have dig­i­tal port­fo­lios to show ev­i­dence of mas­tery of skills like crit­i­cal think­ing and com­mu­ni­ca­tion, which they build up right through K-12 and post-sec­ondary. Se­lec­tive use of high­-qual­ity tests, like the Col­lege and Work Readi­ness As­sess­ment, is im­por­tant. Fi­nal­ly, teach­ers should be judged on ev­i­dence of im­prove­ment in stu­dents’ work through the year - in­stead of a score on a bub­ble test in May. We need lab schools where stu­dents earn a high school diploma by com­plet­ing a se­ries of skil­l-based ‘merit badges’ in things like en­tre­pre­neur­ship. And schools of ed­u­ca­tion where all new teach­ers have ‘res­i­den­cies’ with mas­ter teach­ers and per­for­mance stan­dards - not con­tent stan­dards - must be­come the new nor­mal through­out the sys­tem.”

These sen­ti­ments or goals are so breath­tak­ingly delu­sional (have these peo­ple ever met the av­er­age Amer­i­can? or tried to re­call their mid­dle school al­ge­bra? or thought about how many of their class­mates ac­tu­ally learned any­thing?) that I find my­self won­der­ing (de­spite my per­sonal in­junc­tions against re­sort­ing to ad hominems) that “surely no one could be­lieve such im­pos­si­ble things, ei­ther be­fore or after break­fast; surely an award-win­ning New York Times colum­nist or a fa­mous Har­vard ed­u­ca­tional the­o­rist, surely these peo­ple can­not se­ri­ously be­lieve the claims they are sup­pos­edly mak­ing, and there is some more rea­son­able ex­pla­na­tion - like they have been bribed by spe­cial in­ter­ests, or are ex­pound­ing pro­pa­ganda de­signed to safe­guard their lu­cra­tive profits from pop­ulist re­dis­tri­b­u­tion, or are pulling a prank in very bad taste, or (like Pres­i­dent Rea­gan) are trag­i­cally in the grips of a de­bil­i­tat­ing brain dis­ease?” But the sen­ti­ments are so con­sis­tent and peo­ple who’ve met pro­po­nents of the train­ing panacea say they are gen­uine about it (eg Scott Alexan­der thought the re­train­ing peo­ple were just In­ter­net straw­men un­til he met them), that it must be what they think.

But mov­ing on past An­dreessen and Fried­man. If it re­ally is pos­si­ble for peo­ple to rise to the de­mands of the New Econ­o­my, why is it not hap­pen­ing? For ex­am­ple (em­pha­sis added)

As doc­u­mented in Turner (2004), Bound and Turner (2007, 2011), and , while the num­ber of stu­dents at­tend­ing col­lege has in­creased over the past three decades in the U.S., col­lege grad­u­a­tion rates (i.e., the frac­tion of col­lege en­rollees that grad­u­ate) and col­lege at­tain­ment rates (i.e., the frac­tion of the pop­u­la­tion with a col­lege de­gree) have hardly changed since 1970 and the time it takes col­lege stu­dents to com­plete a bac­calau­re­ate (BA) de­gree has in­creased (Bound, Loven­heim and Turn­er, 2010b). The dis­par­i­ties be­tween the trends in col­lege at­ten­dance and com­ple­tion or time-to-com­ple­tion of col­lege de­grees is all the more stark given that the earn­ings pre­mium for a col­lege de­gree rel­a­tive to a high school de­gree nearly dou­bled over this same pe­riod (Goldin and Katz, 2008).

  • Bound, John and Sarah Turner (2011). “Dropouts and Diplo­mas: The Di­ver­gence in Col­le­giate Out­comes.” in Hand­book of the Eco­nom­ics of Ed­u­ca­tion, Vol. 4, E. Hanushek, S. Machin and L. Woess­mann (ed­s.) El­se­vier B.V., 573-613
  • Gold­in, Clau­dia and Lawrence Katz (2008). The Race be­tween Ed­u­ca­tion and Tech­nol­ogy. Cam­bridge: Har­vard Uni­ver­sity Press

Or “Study of Men’s Falling In­come Cites Sin­gle Par­ents”:

The fall of men in the work­place is widely re­garded by econ­o­mists as one of the na­tion’s most im­por­tant and puz­zling trends. While men, on av­er­age, still earn more than wom­en, the gap be­tween them has nar­rowed con­sid­er­ably, par­tic­u­larly among more re­cent en­trants to the la­bor force. For all Amer­i­cans, it has be­come much harder to make a liv­ing with­out a col­lege de­gree, for in­ter­twined rea­sons in­clud­ing for­eign com­pe­ti­tion, ad­vance­ments in tech­nol­ogy and the de­cline of unions. Over the same pe­ri­od, the earn­ings of col­lege grad­u­ates have in­creased. Women have re­sponded ex­actly as econ­o­mists would have pre­dict­ed, by go­ing to col­lege in record num­bers. Men, mys­te­ri­ous­ly, have not. Among peo­ple who were 35 years old in 2010, for ex­am­ple, women were 17% more likely to have at­tended col­lege, and 23% more likely to hold an un­der­grad­u­ate de­gree. “I think the great­est, most as­ton­ish­ing fact that I am aware of in so­cial sci­ence right now is that women have been able to hear the la­bor mar­ket scream­ing out ‘You need more ed­u­ca­tion’ and have been able to re­spond to that, and men have not,” said Michael Green­stone, an M.I.T. eco­nom­ics pro­fes­sor who was not in­volved in Pro­fes­sor Au­tor’s work. “And it’s very, very scary for econ­o­mists be­cause peo­ple should be re­spond­ing to price sig­nals. And men are not. It’s a fact in need of an ex­pla­na­tion.”

It’s al­ways a lit­tle strange to read an econ­o­mist re­mark that po­ten­tial re­turns to ed­u­ca­tion have been ris­ing and so more peo­ple should get an ed­u­ca­tion, but this same econ­o­mist some­how not re­al­ize that the con­tin­ued pres­ence of this free lunch in­di­cates it is not free at all. Look at how the trend of in­creas­ing ed­u­ca­tion has stalled out:

“Ed­u­ca­tion at­tain­ment climbed dra­mat­i­cally in the 20th cen­tu­ry, but its growth has flat­tened re­cently (source: Cen­sus)”

Ap­par­ently mar­kets work and peo­ple re­spond to in­cen­tives—ex­cept when it comes to ed­u­ca­tion, and there peo­ple sim­ply aren’t pick­ing up those $100 bills lay­ing on the ground and have been not pick­ing them up for decades for some rea­son35, as the share of in­come ac­cru­ing to ‘la­bor’ falls both in the USA and world­wide I see. (In Eng­land, there’s ev­i­dence that col­lege grad­u­ates were still be­ing suc­cess­fully ab­sorbed in the ’90s and ear­lier, al­though ap­par­ently there weren’t rel­a­tively many dur­ing those pe­ri­ods36.) What do bad stu­dents know that good econ­o­mists don’t?

(in­ven­tor of the much-cited - by op­ti­mists and an­ti-neo-lud­dites - com­par­a­tive ad­van­tage con­cept) changed his mind about whether tech­no­log­i­cal un­em­ploy­ment was pos­si­ble, but he thought it was pos­si­ble only un­der cer­tain con­di­tions; Sachs & Kot­likoff 2013 gives a mul­ti­-gen­er­a­tional model of suffer­ing. Most econ­o­mists, though, con­tinue to dis­miss this line of thought, say­ing that tech­no­log­i­cal changes and struc­tural un­em­ploy­ment are real but things will work them­selves out some­how. Robin Han­son, for ex­am­ple, seems to think that and he’s a far bet­ter econ­o­mist than me and has thought a great deal about AI and the eco­nomic im­pli­ca­tions. Their op­po­si­tion to Neo-Lud­dism is about the only rea­son I re­main un­cer­tain, be­cause oth­er­wise, the data for the eco­nomic trou­bles start­ing in 2007, and es­pe­cially the un­em­ploy­ment data, seem to match nice­ly. From a Fed­eral Re­serve brief (prin­ci­pally ar­gu­ing that the data is bet­ter matched by a model in which the longer a worker re­mains un­em­ployed, the longer they are likely to re­main un­em­ployed):

For most of the post-World War II era, un­em­ploy­ment has been a rel­a­tively short­-lived ex­pe­ri­ence for the av­er­age work­er. Be­tween 1960 and 2010, the av­er­age du­ra­tion of un­em­ploy­ment was about 14 weeks. The du­ra­tion al­ways rose dur­ing re­ces­sions, but rel­a­tively quick upticks in hir­ing after re­ces­sions kept the long-term un­em­ploy­ment rate fairly low. Even dur­ing the two “job­less re­cov­er­ies” that fol­lowed the 1990-91 and 2001 re­ces­sions, the peak shares of long-term un­em­ploy­ment were 21% and 23%, re­spec­tive­ly. But the 2007-09 re­ces­sion rep­re­sents a marked de­par­ture from pre­vi­ous ex­pe­ri­ence: the av­er­age du­ra­tion has in­creased to 40 weeks, and the share of long-term un­em­ploy­ment re­mains high more than two years after the offi­cial end of the re­ces­sion.37 Never be­fore in the post­war pe­riod have the un­em­ployed been un­em­ployed for so long.

The Econ­o­mist asked in 2011:

But here is the ques­tion: if the pace of tech­no­log­i­cal progress is ac­cel­er­at­ing faster than ever, as all the ev­i­dence in­di­cates it is, why has un­em­ploy­ment re­mained so stub­bornly high - de­spite the re­bound in busi­ness profits to record lev­els? Two-and-a-half years after the Great Re­ces­sion offi­cially end­ed, un­em­ploy­ment has re­mained above 9% in Amer­i­ca. That is only one per­cent­age point bet­ter than the coun­try’s job­less­ness three years ago at the depths of the re­ces­sion. The mod­est 80,000 jobs added to the econ­omy in Oc­to­ber were not enough to keep up with pop­u­la­tion growth, let alone re-em­ploy any of the 12.3m Amer­i­cans made re­dun­dant be­tween 2007 and 2009. Even if job cre­ation were mirac­u­lously nearly to triple to the monthly av­er­age of 208,000 that is was in 2005, it would still take a dozen years to close the yawn­ing em­ploy­ment gap caused by the re­cent re­ces­sion, says Laura D’An­drea Tyson, an econ­o­mist at Uni­ver­sity of Cal­i­for­nia, Berke­ley, who was chair­man of the Coun­cil of Eco­nomic Ad­vis­ers dur­ing the Clin­ton ad­min­is­tra­tion.

And lays out the cen­tral ar­gu­ment for neo-Lud­dism, why “this time is differ­ent”:

Thanks to trac­tors, com­bine har­vesters, crop-pick­ing ma­chines and other forms of mech­a­ni­sa­tion, agri­cul­ture now ac­counts for lit­tle more than 2% of the work­ing pop­u­la­tion. Dis­placed agri­cul­tural work­ers then, though, could mi­grate from fields to fac­to­ries and earn higher wages in the process. What is in store for the Dil­berts of to­day? Me­dia the­o­rist (Pro­gram or Be Pro­grammed and Life Inc) would ar­gue “noth­ing in par­tic­u­lar.” Put blunt­ly, few new white-col­lar jobs, as peo­ple know them, are go­ing to be cre­ated to re­place those now be­ing lost-de­spite the hopes many place in tech­nol­o­gy, in­no­va­tion and bet­ter ed­u­ca­tion.

The ar­gu­ment against the Lud­dite Fal­lacy rests on two as­sump­tions: one is that ma­chines are tools used by work­ers to in­crease their pro­duc­tiv­i­ty; the other is that the ma­jor­ity of work­ers are ca­pa­ble of be­com­ing ma­chine op­er­a­tors. What hap­pens when these as­sump­tions cease to ap­ply - when ma­chines are smart enough to be­come work­ers? In other words, when cap­i­tal be­comes labour. At that point, the Lud­dite Fal­lacy looks rather less fal­la­cious…In his analy­sis [Lights in the Tun­nel], Mr [Mart­in] Ford noted how tech­nol­ogy and in­no­va­tion im­prove pro­duc­tiv­ity ex­po­nen­tial­ly, while hu­man con­sump­tion in­creases in a more lin­ear fash­ion. In his view, Lud­dism was, in­deed, a fal­lacy when pro­duc­tiv­ity im­prove­ments were still on the rel­a­tively flat, or slowly ris­ing, part of the ex­po­nen­tial curve. But after two cen­turies of tech­no­log­i­cal im­prove­ments, pro­duc­tiv­ity has “turned the cor­ner” and is now mov­ing rapidly up the more ver­ti­cal part of the ex­po­nen­tial curve. One im­pli­ca­tion is that pro­duc­tiv­ity gains are now out­strip­ping con­sump­tion by a large mar­gin.

The Amer­i­can odd­i­ties be­gan be­fore the cur­rent re­ces­sion:

Un­em­ploy­ment in­creased dur­ing the 2001 re­ces­sion, but it sub­se­quently fell al­most to its pre­vi­ous low (from point A to B and then back to C). In con­trast, job open­ings plum­met­ed-much more sharply than un­em­ploy­ment rose-and then failed to re­cov­er. In pre­vi­ous re­cov­er­ies, open­ings even­tu­ally out­num­bered job seek­ers (where a ris­ing blue line crosses a falling green line), but dur­ing the last re­cov­ery a la­bor short­age never emerged. The ane­mic re­cov­ery was fol­lowed in 2007 by an in­crease in un­em­ploy­ment to lev­els not seen since the early 1980s (the rise after point C). How­ev­er, job open­ings fell only a lit­tle-and then re­cov­ered. The re­ces­sion did not re­duce hir­ing; it just dumped a lot more peo­ple into an al­ready weak la­bor mar­ket.38

And then there is the well-known ex­am­ple of Japan. Yet over­all, both Japan­ese, Amer­i­can, and global wealth con­tinue to grow. The hope­ful sce­nario is that all we are suffer­ing is tem­po­rary pains, which will even­tu­ally be grown out of, as fore­cast in his 1930 es­say “Op­ti­mism in a Ter­ri­ble Econ­omy”:

At the same time tech­ni­cal im­prove­ments in man­u­fac­ture and trans­port have been pro­ceed­ing at a greater rate in the last ten years than ever be­fore in his­to­ry. In the United States fac­tory out­put per head was 40 per cent greater in 1925 than in 1919. In Eu­rope we are held back by tem­po­rary ob­sta­cles, but even so it is safe to say that tech­ni­cal effi­ciency is in­creas­ing by more than 1 per cent per an­num com­pound…­For the mo­ment the very ra­pid­ity of these changes is hurt­ing us and bring­ing diffi­cult prob­lems to solve. Those coun­tries are suffer­ing rel­a­tively which are not in the van­guard of progress. We are be­ing afflicted with a new dis­ease of which some read­ers may not yet have heard the name, but of which they will hear a great deal in the years to come - name­ly, tech­no­log­i­cal un­em­ploy­ment. This means un­em­ploy­ment due to our dis­cov­ery of means of economis­ing the use of labour out­run­ning the pace at which we can find new uses for labour. But this is only a tem­po­rary phase of mal­ad­just­ment. All this means in the long run that mankind is solv­ing its eco­nomic prob­lem. I would pre­dict that the stan­dard of life in pro­gres­sive coun­tries one hun­dred years hence will be be­tween four and eight times as high as it is to-day. There would be noth­ing sur­pris­ing in this even in the light of our present knowl­edge. It would not be fool­ish to con­tem­plate the pos­si­bil­ity of afar greater progress still.


Of course, as plau­si­ble as this all looks, that does­n’t mean much. Any­one can cher­ryp­ick a bunch of quotes and ci­ta­tions. When mak­ing pre­dic­tions, there are a few heuris­tics or prin­ci­ples I try to ap­ply, and it might be worth ap­ply­ing a few here.

The spec­i­fi­ca­tion seems fairly clear: the Neo-Lud­dite claim, in its sim­plest form pre­dicts that ever fewer peo­ple will be able to find em­ploy­ment in undis­torted free mar­kets. We can see other as­pects as ei­ther tan­gents (will peo­ple be able to con­sume due to a Ba­sic In­come or via cap­i­tal own­er­ship?) or sub­sets (the Au­tor the­sis of po­lar­iza­tion would nat­u­rally lead to an over­all in­crease in un­em­ploy­men­t). The due date is not clear, but we can see the Neo-Lud­dite the­sis as closely linked to ar­ti­fi­cial in­tel­li­gences, and 2050 would be as good a due date as any inas­much as I ex­pect to be alive then & AI will have ma­tured sub­stan­tially (if we date se­ri­ous AI to 1960 then 2012 is a bit past halfway) & many pre­dic­tions like Ray Kurzweil’s will have been ver­i­fied or fal­si­fied.

The prob­a­bil­ity part of a pre­dic­tion is the hard part. Go­ing in or­der (the lat­ter heuris­tics aren’t help­ful):

  1. “What does the pre­dic­tion about the fu­ture world im­ply about the present world?”

    What would we ex­pect in a world in which the Neo-Lud­dite the­sis were true?

    • first and fore­most, we would ex­pect both soft­ware & hard­ware to con­tinue im­prov­ing. Both are true: Moore’s law con­tin­ues de­spite the break­down in chip fre­quen­cies, and AI re­search forges on with things like deep neural net­works be­ing de­ployed at scale by com­pa­nies such as Google. If we did not see im­prove­ment, that would be ex­tremely dam­ag­ing to the the­sis. How­ev­er, this is a pretty bor­ing retro­d­ic­tion to make: tech­nol­ogy has im­proved for so many cen­turies now that it would be sur­pris­ing if the im­prove­ments had sud­denly stopped, and if it had, why would any­one be tak­ing this the­sis se­ri­ous­ly? It’s not like any­one wor­ries over the im­pli­ca­tions of a philoso­pher’s stone for forex.

    • More mean­ing­ful­ly: cap­i­tal & la­bor in­creas­ingly cease to be com­ple­ments, and be­come sub­sti­tutes. We would ex­pect grad­u­ally ris­ing dis­em­ploy­ment as al­go­rithms & soft­ware & hard­ware were re­fined and com­pa­nies learned when em­ploy­ees could be re­placed by tech­no­log­i­cal sub­sti­tutes, with oc­ca­sional jumps as idio­syn­cratic break­throughs were made for par­tic­u­lar tasks. We would ex­pect re­turns on cap­i­tal to in­crease, and we would ex­pect that em­ploy­ees with un-sub­sti­tutable skills or prop­er­ties would in­crease wealth. This seems sort of true: STEM-related salaries in par­tic­u­lar fields seem to be steady and tech com­pa­nies con­tinue to com­plain that good soft­ware en­gi­neers are hard to find (and Con­gress should au­tho­rize ever more H-1B visas) with con­se­quences such as sky­rock­et­ing San Fran­cisco real es­tate as tech com­pa­nies flock there to find the rare tal­ent they re­quire, which is the sort of “su­per­star effect” we would ex­pect if hu­man be­ings with cer­tain prop­er­ties were in­trin­si­cally rare & valu­able and the re­main­der just so much use­less dross that hold back a busi­ness or worse. This is par­tic­u­larly strik­ing when we note that it has never been cheaper or eas­ier to be­come a soft­ware en­gi­neer as ad­e­quate com­puter hard­ware is dirt cheap & all nec­es­sary soft­ware is avail­able for free on­line & in­struc­tional ma­te­ri­als like­wise, and it’s un­clear how bar­ri­ers like cer­ti­fi­ca­tion could mat­ter when pro­gram­mers are pro­duc­ing ob­jec­tive prod­ucts - ei­ther a web­site is awe­some and works, or it does­n’t.

      On the other hand, I also read of boom­ing poor economies like China or Africa where wages are ris­ing in gen­eral and un­em­ploy­ment seems to be less of a con­cern. This might fit the Au­tor model of po­lar­iza­tion if we fig­ure that those boom­ing economies are pric­ing hu­man la­bor so cheap that it out­com­petes soft­ware/ro­bot­s/etc, in which case we would ex­pect to see these coun­tries hit a “wall” where only a part of their pop­u­la­tions can pass the ‘val­ley of death’ to reach the happy part of the po­lar­ized econ­omy but the rest of the pop­u­la­tion is now strug­gling to be cheap enough to com­pete with the cap­i­tal-al­ter­na­tives. I’m not sure I see this. Yes, there are a lot of ro­botic fac­to­ries be­ing set up in China now, but does that re­ally mean any­thing im­por­tant on Chi­na’s scale? What’s a few mil­lion ro­bots in a coun­try of 1.3 bil­lion peo­ple? If China does wind up falling into what looks like a , that would be con­sis­tent with the Au­tor mod­el, I think, and strengthen this retro­d­ic­tion.

    • As tech­nol­ogy is mo­bile and can eas­ily be sold or ex­port­ed, we would ex­pect to see this gen­eral trend in many wealthy West­ern coun­tries. This is a se­ri­ous weak point of my knowl­edge thus far: I sim­ply don’t know what it re­ally looks like in eg Japan or Eng­land or Ger­many. Are they see­ing sim­i­lar things to my fac­toids about the USA?

  2. “Base rates” here is es­sen­tially ap­ply­ing the Out­side View

    The main prob­lem here is that it’s very diffi­cult to re­but the Out­side View: the Lud­dite the­sis has, it seems, failed many times in the past; why ex­pect this time to be any differ­ent? The his­tor­i­cal horse ex­am­ple is amus­ing, cer­tain­ly, but there could be many fac­tors sep­a­rat­ing horses from or­di­nary peo­ple. To this, I don’t have any good re­ply. Even if the the­sis is “right” from the per­spec­tive of 1000 years from now, there is good rea­son to be chary of ex­pect­ing it to hap­pen in my life­time. Com­put­ers them­selves fur­nish a great many ex­am­ples of peo­ple who, with vi­sion and deep in­sight not shared by the peo­ple who ridiculed them as tech­no-u­topi­ans, cor­rectly fore­saw things like the per­sonal com­puter or the In­ter­net or on­line sales - and started their com­pa­nies too ear­ly. The best I can say is that soft­ware/AI seems com­pletely & qual­i­ta­tively differ­ent from ear­lier tech­nolo­gies like rail­roads or as­sem­bly lines, in that they are per­form­ing deeply hu­man men­tal func­tions that ear­lier tech­nolo­gies did not come any­where near: the reg­u­la­tor of a steam en­gine is solv­ing a prob­lem so much sim­pler than an au­tonomous car solves that it’s hard to even see them as be­ing even the­o­ret­i­cally re­lated in ex­ert­ing con­trol on processes by feed­back process­es. The dimmest hu­man could pro­duc­tively use con­tem­po­rary tech­nolo­gies, where to­day we strug­gle to find sub­si­dized jobs for the men­tally hand­i­capped where they are even just not a net loss.

From these mus­ings, I think we can ex­tract a few warn­ing signs which would in­di­cate the Neo-Lud­dite the­sis break­ing down:

  • global eco­nomic growth stop­ping
  • AI re­search progress stop­ping
  • Moore’s law in terms of FLOPS/$ break­ing down
  • de­creased wealth in­equal­ity (eg. Gini) in the First World
  • in­creases in pop­u­la­tion work­ing

Daniel Kah­ne­man has an in­ter­est­ing think­ing tech­nique he calls the “pre-mortem”, where you ask your­self: “as­sume it’s the fu­ture, and my con­fi­dent pre­dic­tions have com­pletely failed to come true. What went wrong?” Look­ing back, if the Neo-Lud­dite the­sis fails, I think the most likely ex­pla­na­tion for what I’ve seen in the USA would be some­thing re­lated to glob­al­iza­tion & China in par­tic­u­lar: the po­lar­iza­tion, in­creased dis­em­ploy­ment, in­creas­ing need for tech­ni­cal train­ing etc, all seem ex­plain­able by those jobs head­ing over­seas, ex­ac­er­bated by other fac­tors such as do­mes­tic pol­i­tics (Bush’s tax cuts on the rich?) and maybe things like the struc­tural un­em­ploy­ment re­lat­ing to ex­ist­ing work­ers hav­ing diffi­culty switch­ing sec­tors or jobs but new work­ers be­ing able to adapt. If this is so, then I think we would ex­pect the trends to grad­u­ally ame­lio­rate them­selves: older work­ers will die off & re­tire, new work­ers will re­place them, new niches and jobs will open up as the econ­omy adapts, Chi­na’s ex­po­nen­tial growth will re­sult in catchup be­ing com­pleted within 2 or 3 decades, and so on.


may be even more in­flam­ma­tory than sup­port­ing nicotine, but it’s an im­por­tant en­try on any hon­est list. I never doubted that IQ was in part hered­i­tary (Stephen Jay Gould aside, this is too ob­vi­ous - what, every­thing from drug re­sponses to skin and eye color would be her­i­ta­ble ex­cept the most im­por­tant things which would have a huge effect on re­pro­duc­tive fit­ness?), but all the ex­perts seemed to say that di­luted over en­tire pop­u­la­tions, any ten­dency would be non-ex­is­tent. Well, OK, I could be­lieve that; vis­i­ble traits con­sis­tent over en­tire pop­u­la­tions like skin color might differ sys­tem­at­i­cally be­cause of sex­ual se­lec­tion or some­thing, but why not leave IQ fol­low­ing the ex­act same bell curve in each pop­u­la­tion? There was no spe­cific thing here that made me start to won­der, more a grad­ual un­der­min­ing (Gould’s work like be­ing com­pletely dis­hon­est is one ex­am­ple - with en­e­mies like that…) as I con­tin­ued to read stud­ies and won­der why Asian model mi­nori­ties did so well, and a lack of re­ally con­vinc­ing coun­ter-ev­i­dence like one would ex­pect the last two decades to have pro­duced - given the pol­i­tics in­volved - if the idea were false. And one can al­ways ask one­self: sup­pose that in­tel­li­gence was mean­ing­ful, and did have a large ge­netic com­po­nent, and the likely ge­netic rank­ing East Asians > Cau­casian > Africans; in what way would the world, or the last mil­len­nium (eg the growth of the Asian tigers vs Africa, or the differ­ent ex­pe­ri­ences of dis­crim­i­nat­ed-a­gainst mi­nori­ties in the USA), look differ­ent than it does now?


It’s worth not­ing that the IQ wars are a rab­bit hole you can eas­ily dive down. The lit­er­a­ture is vast, spans all sorts of groups, all sorts of de­signs, from test va­lidi­ties to sam­pling to sta­tis­ti­cal re­gres­sion vs causal in­fer­ence to forms of bi­as; every point is hotly de­bat­ed, the ways in which stud­ies can be validly cri­tiqued are an ed­u­ca­tion in how to read pa­pers and look for how they are weak or make jumps or some of the data just looks wrong, and you’ll learn every tech­ni­cal re­quire­ment and premise and method­olog­i­cal lim­i­ta­tion be­cause the op­po­nents of that par­tic­u­lar re­sult will be sure to bring them up if it’ll at all help their case.

In this re­spect, it’s a lot like the feuds in bib­li­cal crit­i­cism over is­sues like , or the long philo­soph­i­cal de­bate over the . There too is an in­cred­i­ble amount of ma­te­r­ial to cov­er, by some re­ally smart peo­ple (what did geeks do be­fore sci­ence and moder­ni­ty? well, for the most part, they seem to have done the­ol­o­gy; con­sider how much time and effort Isaac New­ton re­port­edly and , or the sheer brain­power that must’ve been spent over the cen­turies in rab­bini­cal stud­ies). You could learn a lot about the an­cient world or the in­cred­i­bly com­plex chain of trans­mis­sion of the Bible’s con­stituents in their end­less va­ri­eties and how they are put to­gether into a sin­gle canon­i­cal mod­ern text, or the other count­less is­sues of . An aw­ful lot, in­deed. One could, and peo­ple as smart or smarter than you have, lose one’s life in ex­plor­ing lit­tle back­-al­leys and de­tails.

If, like most peo­ple, you’ve only read a few pa­pers or books on it, your opin­ion (what­ever that is) is worth­less and you prob­a­bly don’t even re­al­ize how worth­less your opin­ion is, how far you are from ac­tu­ally grasp­ing the sub­tleties in­volved and hav­ing a com­mand of all the stud­ies and crit­i­cisms of said stud­ies. I ex­empt my­self from this only inas­much as I have re­al­ized how lit­tle I still know after all my read­ing. No mat­ter how tempt­ing it is to think that you may be able to fi­nally put to­gether the com­pelling refu­ta­tion of God’s ex­is­tence or to demon­strate that Je­sus’s di­vin­ity was a late ad­di­tion to his gospel, you won’t make a dent in the de­bate. In other words, these can be­come forms of nerd snip­ing and in­tel­lec­tual crack. “If only I com­pile a few more stud­ies, make a few more points—then my case will be­come clear and con­vinc­ing, and peo­ple on the In­ter­net will stop be­ing wrong!”

But hav­ing said that, and ad­mir­ing things like Planti­nga’s free will de­fense, and the sub­tle log­i­cal is­sues in for­mu­lat­ing it and the lack of any re­ally con­crete ev­i­dence for or against Je­sus’s ex­is­tence, do I take the ba­sic ques­tion of God se­ri­ous­ly? No. The the­ists’ rear­guard at­tempts and ever more in­ge­nious ex­pla­na­tions and in­di­rect path­ways of rea­sons and touted mir­a­cles fun­da­men­tally do not add up to an ex­ist­ing whole. The uni­verse does not look any­thing like a om­ni-benev­o­len­t/pow­er­ful/­scient god was in­volved, a great deal of de­ter­mined effort has failed to pro­vide any con­vinc­ing proof, there not be­ing a god is con­sis­tent with all the ob­served processes and an­i­mal king­dom and nat­ural events and ma­te­r­ial world we see, and so on. The per­sis­tence of the de­bate re­flects more what mo­ti­vated cog­ni­tion can ac­com­plish and the weak­ness of ex­ist­ing epis­te­mol­ogy and de­bate. Un­for­tu­nate­ly, this could be equally well-said by some­one on the other side of the de­bate, and in any case, I can­not com­mu­ni­cate my gestalt im­pres­sion of the field to any­one else. I don’t ex­pect any­one to be the least bit swayed by what I’ve writ­ten here.

So why be in­ter­ested in the top­ics at all? If you can­not con­vince any­one, if you can­not learn the field to a rea­son­able depth, and you can­not even com­mu­ni­cate well what con­vinced you, why bother? In the spirit of , I say: it’s not clear at all. So you should know in ad­vance whether you want to take the red pill and see how far down the rab­bit hole you go be­fore you fi­nally give up, or you take the blue pill and be an on­looker as you set­tle for a high­-level overview of the more in­ter­est­ing pa­pers and is­sues and ac­cept that you will only have that and a gen­eral in­de­fen­si­ble as­sess­ment of the state of play.

My own be­lief is that as in­ter­est­ing as it is, you should take the blue pill and not adopt any strong po­si­tion but per­haps (if it does­n’t take too much time) point out any par­tic­u­larly naive or egre­gious holes in ar­gu­ment, by peo­ple who are sim­ply wrong or don’t re­al­ize how lit­tle they know or how slanted a view they have re­ceived from the ma­te­r­ial they’ve read. It’s sad to not reach agree­ment with other peo­ple, dan­ger­ous to ig­nore crit­ics, tempt­ing to en­gage trolls - but life is too short to keep tread­ing the same ground.

The rea­son for IQ is this: yes, Mur­ray failed to or­ga­nize a de­fin­i­tive ge­netic study. It has­n’t hap­pened yet even though it’s more im­por­tant than most of the triv­i­al­i­ties that get stud­ied in pop­u­la­tion ge­net­ics (like his­tor­i­cal move­ments of ran­dom group­s). I don’t need to ex­plain why this would be the case even if peo­ple on the en­vi­ron­men­tal­ist side of the IQ wars were con­fi­dent they were right. But the mas­sive fall in genome se­quenc­ing costs (pro­jected to be <$1000 by ~2014) means that large hu­man datasets will be pro­duced, and the ge­net­ics di­rectly ex­am­ined, elim­i­nat­ing en­tire ar­eas of ob­jec­tions to the pre­vi­ous hered­ity stud­ies. And at some point, some re­searcher will man­age the study - some group in­side or out­side the USA will fund it, at some point a large enough ge­netic data­base will be cross-ref­er­enced against IQ tests and ex­ist­ing racial mark­ers. We al­ready see some of this in re­search: (fol­lowup: ) found 3 SNPs sim­ply by pool­ing ex­ist­ing data­bases of ge­net­ics data & cor­re­lat­ing against school­ing. I don’t know when the de­fin­i­tive pa­per will come out, if it’ll be this year, or by 2020, al­though I would be sur­prised if there was still noth­ing by 2030; but it will hap­pen and it will hap­pen rel­a­tively soon (for a de­bate go­ing on for the past cen­tury or more). Genome se­quenc­ing is sim­ply go­ing to be too cheap for it to not hap­pen. By 2030 or 2040, I ex­pect the is­sue will be de­fin­i­tively set­tled in the same way ear­lier de­bates about the va­lid­ity of IQ tests were even­tu­ally set­tled (even if the pub­lic has­n’t yet got­ten the word, the ex­perts all con­cede that IQ tests are valid, re­li­able, not bi­ased, and mean­ing­ful pre­dic­tors of a wide va­ri­ety of re­al-world vari­ables).

Value of Information

What is the di­rect value of learn­ing about IQ? Speak­ing of it in terms of money may not be the best ap­proach, so in­stead we can split the ques­tion up into a few differ­ent sub­-ques­tions:

  1. how much do your efforts lead to ad­di­tional in­for­ma­tion?

    In this case, not much. I would have to be very ar­ro­gant to think I can go through a large frac­tion of the lit­er­a­ture and eval­u­ate it bet­ter than the ex­ist­ing au­thor­i­ties like Nis­bett or Flynn or Jensen. I have no ad­van­tages over them.

  2. would this in­for­ma­tion-gath­er­ing be ex­pen­sive?

    Yes. A sin­gle pa­per can take an hour to read well, and a tech­ni­cal book weeks. There are hun­dreds of pa­pers and dozens of books to learn. The math­e­mat­ics and sta­tis­tics are non­triv­ial, and sooner or lat­er, one will have to learn them in or­der to eval­u­ate the se­ri­ous­ness of crit­i­cisms for one­self. The time spent will not have been throw-away recre­ational time, ei­ther, like slum­ming on the couch watch­ing TV, but will be one’s high­est-qual­ity time, which could have been spent learn­ing other diffi­cult ma­te­ri­al, work­ing, mean­ing­fully in­ter­act­ing with other peo­ple, and so on. Given the de­cline with age of fluid in­tel­li­gence, one may be wast­ing a non-triv­ial frac­tion of one’s life­time learn­ing.

  3. will new in­for­ma­tion come in the ab­sence of your efforts?

    Yes. My in­ter­est does not ma­te­ri­ally affect when the fi­nal ge­netic stud­ies will be con­duct­ed.

  4. what de­ci­sions or be­liefs would the ad­di­tional in­for­ma­tion change?

    Sup­pose the en­vi­ron­men­tal­ists were 100% right and the be­tween-race ge­net­ics were a neg­li­gi­bly small fac­tor. Re­gard­less, the topic of IQ and its cor­re­lates and what it pre­dicts does not live and die based on there be­ing a ge­netic fac­tor to av­er­age IQ differ­ences be­tween groups; if the ad­mix­ture and ge­net­ics stud­ies turn in a solid es­ti­mate of 0, IQ will still pre­dict life­time in­come, still pre­dict crime rates, still pre­dict ed­u­ca­tional scores, and so on.

    In con­trast, some of the other top­ics have very con­crete im­me­di­ate im­pli­ca­tions. Switch­ing from oc­cultism/the­ism to athe­ism im­plies many changed be­liefs & choic­es; a near vs far Sin­gu­lar­ity has con­sid­er­able con­se­quences for re­tire­ment plan­ning, if noth­ing else; while Neo-Lud­dism has im­pli­ca­tions for both ca­reer choice and re­tire­ment plan­ning; at­ti­tudes to­wards fic­tion and nico­tine also cash out in ob­vi­ous ways. Of the top­ics here, per­haps only Com­mu­nism and the Amer­i­can Rev­o­lu­tion are as ster­ile in prac­ti­cal ap­pli­ca­tion.

If ge­netic differ­ences and in­equal­ity ex­ists, they should be en­gi­neered away.

So, I try not to spend too much time think­ing about this is­sue: the re­sults will come in re­gard­less of my opin­ion, and un­like other is­sues here, does not ma­te­ri­ally affect my world­view or sug­gest ac­tion. Given this, there’s no rea­son to in­vest your life in the top­ic! It has no prac­ti­cal ram­i­fi­ca­tions for you, dis­cussing the is­sue can only lead to neg­a­tive con­se­quences - and on the in­tel­lec­tual lev­el, no mat­ter how much you read, you’ll al­ways have nag­ging doubts, so you won’t get any sat­is­fac­tion. You might as well just wait pa­tiently for the in­evitable fi­nal an­swer.


Miller on neo-Luddism

From chap­ter 13 of Sin­gu­lar­ity Ris­ing, James Miller 2012:

“There’s this stu­pid myth out there that AI has failed, but AI is every­where around you every sec­ond of the day. Peo­ple just don’t no­tice it. You’ve got AI sys­tems in cars, tun­ing the pa­ra­me­ters of the fuel in­jec­tion sys­tems. When you land in an air­plane, your gate gets cho­sen by an AI sched­ul­ing sys­tem. Every time you use a piece of Mi­crosoft soft­ware, you’ve got an AI sys­tem try­ing to fig­ure out what you’re do­ing, like writ­ing a let­ter, and it does a pretty damned good job. Every time you see a movie with com­put­er-gen­er­ated char­ac­ters, they’re all lit­tle AI char­ac­ters be­hav­ing as a group. Every time you play a video game, you’re play­ing against an AI sys­tem.” –Rod­ney Brooks, Di­rec­tor, MIT Com­puter Sci­ence and AI Lab­o­ra­tory 288

…In the next few decades, all of my read­ers might have their mar­ket value dec­i­mated by in­tel­li­gent ma­chines. Should you be afraid? Fear of job-de­stroy­ing tech­nol­ogy is noth­ing new. Dur­ing the eigh­teenth cen­tu­ry, cloth­ing man­u­fac­tur­ers in Eng­land re­placed some of their hu­man la­bor­ers with ma­chines. In re­spon­se, a gang sup­pos­edly led by one Ned Ludd smashed a few ma­chines owned by a sock mak­er. Ever since then, peo­ple op­pos­ing tech­nol­ogy have been called Lud­dites. Lud­dites are cor­rect in think­ing that ma­chines can cause work­ers to lose their jobs. But over­all, in the past, job-de­stroy­ing ma­chine pro­duc­tion has over­all greatly ben­e­fited work­ers. “De­stroy­ing jobs” sounds bad - like some­thing that should harm an econ­o­my. But the ben­e­fits of job de­struc­tion be­come ap­par­ent when you re­al­ize that an econ­o­my’s most valu­able re­source is hu­man brains. If a busi­ness­man fig­ured out how to make a prod­uct us­ing less en­ergy or fewer ma­te­ri­als, we would ap­plaud him be­cause the sav­ings could be used to pro­duce ad­di­tional goods. The same holds true when we fig­ure out how to make some­thing us­ing less la­bor. If you used to need 1,000 work­ers to run your sock fac­tory but you can now pro­duce the same num­ber of socks by em­ploy­ing only 900 work­ers, then you prob­a­bly would (and per­haps even should) fire the other 100. Al­though in the short run these work­ers will lack jobs, in the long run they will likely find new em­ploy­ment and ex­pand the econ­o­my.

The oblit­er­a­tion of most agri­cul­tural jobs has been a huge source of eco­nomic growth for Amer­i­ca. In 1900, farm­ers made up 38% of the Amer­i­cans work­force, whereas now they con­sti­tute less than 2% of it.289 Most of the dis­placed agri­cul­tural la­bor­ers found work in cities. Yet de­spite the mas­sive de­crease in farm­ing jobs, the United States has steadily pro­duced more and more food since 1900. Agri­cul­tural tech­nol­ogy gave the Amer­i­can peo­ple a “free lunch,” in which we got more food with less effort, mak­ing obe­sity a greater threat to Amer­i­can health than calo­rie de­pri­va­tion. Tech­nol­ogy raises wages by in­creas­ing worker pro­duc­tiv­i­ty. In a free-mar­ket econ­o­my, the value of the goods an em­ployee pro­duces for his em­ployer roughly de­ter­mines his wage. A farmer with a trac­tor pro­duces more food than one with just a hoe. Con­se­quent­ly, mod­ern farm­ers earn higher wages than they would if they lived in a world de­prived of mod­ern agri­cul­tural tech­nol­o­gy. In rich na­tions, wages have risen steadily over the last two hun­dred years be­cause tech­nol­ogy keeps in­creas­ing worker pro­duc­tiv­i­ty. But will this trend con­tin­ue? Past tech­nolo­gies never com­pletely elim­i­nated the need for hu­mans, so fired sock work­ers usu­ally found other em­ploy­ment. But a suffi­ciently ad­vanced AI pos­sess­ing a ro­bot body might out­per­form peo­ple at every sin­gle task.

…If a Kurzweil­ian merger does­n’t oc­cur, sen­tient AIs might com­pete di­rectly with peo­ple in the la­bor mar­ket. Let’s now ex­plore what hap­pens to hu­man wages if these AIs be­come bet­ter than hu­mans at every task. Adam Smith, the great eigh­teen­th-cen­tury econ­o­mist, ex­plained that every­one ben­e­fits from trade if each par­tic­i­pant makes what he is best at. So, for ex­am­ple, if I’m bet­ter at mak­ing boots than you are, but you have more skill at mak­ing can­dles, then we would both be­come richer if I pro­duced your boots and you made my can­dles. But what if you’re more skilled at mak­ing both boots and can­dles? What if, com­pared to you, I’m worse at do­ing every­thing? Adam Smith never an­swered this ques­tion, but nine­teen­th-cen­tury econ­o­mist David Ri­cardo did. This ques­tion is highly rel­e­vant to our fu­ture, as an AI might be able to pro­duce every good and ser­vice at a lower cost than any hu­man could, and if we turn out to have no eco­nomic value to the ad­vanced ar­ti­fi­cial in­tel­li­gences, then they might (at best) ig­nore us, de­priv­ing hu­man­ity of any ben­e­fits of their su­per­hu­man skills.

Most peo­ple in­tu­itively be­lieve that mu­tu­ally ben­e­fi­cial trades take place only when each per­son has an area of ab­solute ex­cel­lence. But Ri­car­do’s the­ory of com­par­a­tive ad­van­tage shows that trade can make every­one bet­ter off re­gard­less of a per­son’s ab­solute skill be­cause every­one has an area of com­par­a­tive ad­van­tage. I’ll il­lus­trate Ri­car­do’s nine­teen­th-cen­tury the­ory with a twen­ty-first-cen­tury ex­am­ple in­volv­ing donuts and an­ti-grav­ity fly­ing cars. Let’s as­sume that hu­mans can’t make fly­ing cars, but an AI can; and al­though peo­ple can make donuts, an AI can make them much faster than we can. Let’s pre­tend that at least one AI likes donuts, where donuts rep­re­sent any­thing a hu­man can make that an AI would want. Here’s how a hu­man and AI could both ben­e­fit from trade: a hu­man could offer to give an AI many donuts in re­turn for a fly­ing car. The trade could clearly ben­e­fit the hu­man. If it gets enough donuts, the AI also ben­e­fits from the trade. To see how this could work, imag­ine that (ab­sent trade) it takes an AI one sec­ond to make a donut. The AI could build a fly­ing car in one minute.

  • Time needed for an AI to make a donut: one sec­ond
  • Time needed for an AI to make a fly­ing: car one minute

A hu­man then offers the fol­low­ing deal to the AI: Build me a fly­ing car and I will give you one hun­dred donuts. It will take you one minute to make me a fly­ing car. In re­turn for this fly­ing car you get some­thing that would cost you 100 sec­onds to make. Con­se­quent­ly, our trade saves you 40 sec­onds. As the AI’s pow­ers grew, peo­ple could still gain from trad­ing with it. If, say, it took the AI only one nanosec­ond to make a donut and 60 nanosec­onds to make a fly­ing car, then it would still be­come bet­ter off by trad­ing 100 donuts for 1 fly­ing car. 292 In gen­er­al, as an AI be­comes more in­tel­li­gent, trad­ing with hu­mans will save it less time, but what the AI can do with this saved time goes up, es­pe­cially since a smarter AI would prob­a­bly gain the ca­pac­ity to cre­ate en­tirely new cat­e­gories of prod­ucts. An AI might trade 100 donuts for a fly­ing car, but an “AI+” would trade this num­ber of donuts for a worm­hole gen­er­a­tor. Mod­ern econ­o­mists use Ri­car­do’s the­ory of com­par­a­tive ad­van­tage to show how rich and poor coun­tries can ben­e­fit by trad­ing with each oth­er. Un­der­stand­ing Ri­car­do’s the­ory causes al­most all econ­o­mists to fa­vor free trade. If we sub­sti­tute “hu­man­ity” for “poor coun­tries” and AI for “rich coun­tries,” then Ri­cardo gives us some hope for be­liev­ing that even self­-in­ter­ested ad­vanced ar­ti­fi­cial in­tel­li­gences would want to take ac­tions that be­stow tremen­dous eco­nomic ben­e­fits on mankind.


In the pre­vi­ous sce­nar­io, I im­plic­itly as­sumed that pro­duc­ing donuts does­n’t re­quire the use of some “fac­tor of pro­duc­tion.” A fac­tor of pro­duc­tion is an es­sen­tial non­hu­man el­e­ment needed to cre­ate a good. Fac­tors of pro­duc­tion for donuts in­clude land, ma­chi­nes, and raw ma­te­ri­als, and with­out these fac­tors, a per­son (no mat­ter how smart and hard­work­ing) can’t make donuts. In­stead of us­ing the in­tim­i­dat­ing and bor­ing term “fac­tor of pro­duc­tion,” I’m go­ing to say that to make a good or pro­duce a ser­vice you need the right “magic pro­duc­tion wand,” with the wand be­ing the ap­pro­pri­ate set of fac­tors of pro­duc­tion. For ex­am­ple, a donut maker needs a donut wand.

If a rel­a­tively small num­ber of wands ex­isted and no more could be cre­at­ed, then all of the wands would go to AIs. Let’s say donuts sell for $1 each and an AI could use a donut wand to pro­duce one mil­lion donuts, whereas a hu­man us­ing the same wand could make only a thou­sand donuts. A hu­man would never be will­ing to pay more than $1,000 for the wand, whereas an AI would earn a huge profit if it bought a donut wand for, say, $10,000. Even if a hu­man ini­tially owned a donut wand, he would soon sell it to an AI. Hu­man wand own­ers in this sit­u­a­tion would ben­e­fit from AIs be­cause AIs would greatly raise the mar­ket value of wands. Hu­man work­ers who had never had a wand would be­come im­pov­er­ished be­cause they could­n’t pro­duce any­thing.

The Ro­man Re­pub­lic’s con­quests in the first cen­tury BC effec­tively stripped many Ro­man cit­i­zens of their pro­duc­tion wands. In the early Re­pub­lic, poor cit­i­zens had ac­cess to wands, as they were often hired to farm the land of the no­bil­i­ty. But after the Re­pub­lic’s con­quests brought in a huge num­ber of slaves, the no­ble­men had their slaves use al­most all of the avail­able land wands. Cheap slave la­bor en­riched the landown­ing no­bil­ity by re­duc­ing their pro­duc­tion costs. But abun­dant slave la­bor im­pov­er­ished non-landown­ing Ro­mans by de­priv­ing them of wands. Cheap slave la­bor con­tributed to the fall of the Ro­man Re­pub­lic. As Ro­man in­equal­ity in­creased, com­mon sol­diers came to rely on their gen­er­als for fi­nan­cial sup­port. The troops put loy­alty to their gen­er­als ahead of loy­alty to the Ro­man state. Gen­er­als such as Sulla and Julius Cae­sar took ad­van­tage of their in­creased in­flu­ence over their troops to pro­pel them­selves to ab­solute po­lit­i­cal pow­er. Cae­sar sought to re­duce the so­cial in­sta­bil­ity caused by slaves by giv­ing im­pov­er­ished free Ro­man cit­i­zens new lands from the ter­ri­to­ries Rome had re­cently con­quered. Cae­sar es­sen­tially cre­ated many new wands and gave them to his sub­jects.

Al­though AIs will use wands, they will also likely help cre­ate them. For ex­am­ple, us­ing nan­otech­nol­o­gy, they might be able to build dikes to re­claim land from the ocean. Or per­haps they’ll fig­ure out how to ter­raform Mars, mak­ing Mar­t­ian land cheap enough for nearly any hu­man to afford. AIs could also fig­ure out bet­ter ways to ex­tract raw ma­te­ri­als from the earth or in­vent new ways to use raw ma­te­ri­als, re­sult­ing in each prod­uct need­ing fewer wands. The fu­ture of hu­man wages might come down to a race be­tween the num­ber of AIs and the quan­tity of wands. Econ­o­mist and for­mer ar­ti­fi­cial-in­tel­li­gence pro­gram­mer Robin Han­son has cre­ated a highly coun­ter­in­tu­itive the­ory of why (in the long run) AIs will de­stroy nearly all hu­man jobs: they will end up us­ing all of the pro­duc­tion wands (“Eco­nomic Growth Given Ma­chine In­tel­li­gence”).

…What I’ve writ­ten so far about the eco­nom­ics of em­u­la­tions prob­a­bly seems cor­rect to most read­ers. After all, if we can make copies of ex­tra­or­di­nar­ily bright and pro­duc­tive peo­ple and em­ploy mul­ti­ple copies of them in sci­ence and in­dus­try, then we should all get rich­er. The re­sults would be sim­i­lar to what would hap­pen if a se­lect few nurs­ery schools be­came so fan­tas­ti­cally good that each year they turned ten thou­sand tod­dlers into von Neu­man­n-level ge­niuses who then im­me­di­ately en­tered the work­force.

Robin Han­son, how­ev­er, is­n’t will­ing to rely on mere in­tu­ition when an­a­lyz­ing the eco­nom­ics of em­u­la­tions. Robin re­al­izes that if, after we have em­u­la­tions, the price of com­put­ing power con­tin­ues to fall at an ex­po­nen­tial rate, then em­u­la­tions will soon be­come ex­tra­or­di­nar­ily cheap. If you com­bine ex­tremely in­ex­pen­sive em­u­la­tions with a bit of eco­nomic the­o­ry, you get a seem­ingly crazy re­sult, some­thing that you might think is too ab­surd to ever hap­pen. But Robin, ever the bul­let-eater, re­fuses to turn away from his con­clu­sion. Robin thinks that in the long run, em­u­la­tions will drive wages down to al­most ze­ro, push­ing most of the peo­ple who are un­for­tu­nate enough to rely on their wages into star­va­tion-be­cause em­u­la­tions will kick us back into a “Malthu­sian trap.” Ar­guably, hu­man­i­ty’s great­est ac­com­plish­ment was es­cap­ing the Malthu­sian trap. Thomas Malthus, a nine­teen­th-cen­tury econ­o­mist, be­lieved that star­va­tion would ul­ti­mately strike every coun­try in the en­tire world. Malthus wrote that if a pop­u­la­tion is not fac­ing star­va­tion, peo­ple in that pop­u­la­tion will have many chil­dren who grow up, get mar­ried, and have even more chil­dren. A coun­try with an abun­dance of food, Malthus wrote, is one with an in­creas­ing pop­u­la­tion. Un­for­tu­nate­ly, in Malthus’s time, as the size of a coun­try’s pop­u­la­tion went up, it be­came more diffi­cult to feed every­one in the coun­try. Even­tu­al­ly, when the pop­u­la­tion got large enough, many starved. Only when lots of peo­ple were dy­ing of star­va­tion would the coun­try’s pop­u­la­tion sta­bi­lize. Con­se­quent­ly, Malthus be­lieved that all coun­tries were trapped in one of two sit­u­a­tions:

  1. Many peo­ple are starv­ing.
  2. The pop­u­la­tion is grow­ing, and so many will even­tu­ally starve.

…Pre­tend that some­one em­u­lates Robin and places the soft­ware in the pub­lic do­main. Any­one can now freely copy e-Robin, al­though it still costs some­thing to buy enough com­put­ing power to run him on, say a hun­dred thou­sand dol­lars a year. A profit-max­i­miz­ing busi­ness would em­ploy an e-Robin if the e-Robin brought the busi­ness more than $100,000 a year in rev­enue. After Moore’s law pushes the an­nual hard­ware costs of an e-Robin down to a mere $1, then a com­pany would hire e-Robins as long as each brought the busi­ness more than $1 per an­num. What hap­pens to the salary of bio-Robin if you can hire an e-Robin for only a dol­lar? David Ri­cardo im­plic­itly knew the an­swer to that ques­tion. Ri­cardo wrote that if it costs 5,000 pounds to rent a ma­chine, and this ma­chine could do the work of 100 men, the to­tal wages paid to 100 men will never be greater than 5,000 pounds be­cause if the to­tal wages were high­er, man­u­fac­tur­ers would fire the work­ers and rent the ma­chine.296 Ap­ply­ing Ri­car­do’s the­ory to an econ­omy with em­u­la­tions tells us that, if an em­u­la­tion can do what­ever you can do, your wage will never be higher than what it costs to em­ploy the em­u­la­tion. The ques­tion now is whether, if it’s ex­tremely cheap to run an e-Robin, these e-Robins would still earn high salaries and there­fore al­low the orig­i­nal Robin to bring home a de­cent pay­check. Un­for­tu­nate­ly, the an­swer is no be­cause if an e-Robin were earn­ing much more than what it costs to run an e-Robin, then it would be profitable for busi­nesses to cre­ate many more of them. Com­pa­nies will keep mak­ing copies of their em­u­la­tions un­til they no longer make a profit by pro­duc­ing the next copy. A gen­eral rule of eco­nom­ics is that the more you have of some­thing, the smaller its val­ue. For ex­am­ple, even though wa­ter is in­her­ently much more use­ful than di­a­monds be­cause there is so much more wa­ter than di­a­monds, the price of wa­ter is much low­er. If any­one can freely copy e-Robin, then the free mar­ket would drive the wage of an e-Robin down to what it costs to run one.

…Even if the em­u­la­tions push wages to al­most ze­ro, lots of bio-hu­mans would be much richer than they would be in a world with­out em­u­la­tions. Though an­cient Rome was in a Malthu­sian trap, its landown­ing no­bil­ity was rich. When you have lots of peo­ple and lit­tle land, the land is ex­tremely valu­able be­cause it’s cheap to hire peo­ple to work the land and there is great de­mand for the food the land pro­duces. Sim­i­lar­ly, if there are a huge num­ber of em­u­la­tions and rel­a­tively few pro­duc­tion wands, then the wands be­come ex­tra­or­di­nar­ily valu­able. True, the em­u­la­tions will in­crease the num­ber of pro­duc­tion wands. But be­cause it’s so cheap to copy soft­ware, if the price of hard­ware is low enough, there will al­ways be a lot more la­bor than wands. Con­se­quent­ly, bio-peo­ple who own wands will be­come fan­tas­ti­cally rich. Even though you would lose your job, the value of your stock port­fo­lio might jump a thou­sand­fold…S­ince bio-hu­mans could earn al­most noth­ing by work­ing, our pros­per­ity would de­pend on our own­ing prop­erty or re­ceiv­ing wel­fare pay­ments. If bio-hu­mans be­came mas­ters of an em­u­la­tion-filled Malthu­sian world, keep­ing most of the wealth for our­selves, then we would live like a landed aris­toc­racy that re­ceives in­come from tax­ing oth­ers and rent­ing out our agri­cul­tural lands to poor peas­ants. Eliezer Yud­kowsky doubts this pos­si­bil­i­ty:

“The prospect of bi­o­log­i­cal hu­mans sit­ting on top of a pop­u­la­tion of [em­u­la­tions] that are smarter, much faster, and far more nu­mer­ous than bios while hav­ing all the stan­dard hu­man dri­ves, and the bios treat­ing the [em­u­la­tions] as stan­dard eco­nomic [val­ue] to be milked and traded around, and the [em­u­la­tions sit­ting] still for this for more than a week of bio time - this does not seem his­tor­i­cally re­al­is­tic.”301

Carl Shul­man, one of the most knowl­edge­able peo­ple I’ve spo­ken to about Sin­gu­lar­ity is­sues, goes even fur­ther than Yud­kowsky. He writes that since ob­so­les­cence would fre­quently kill en­tire cat­e­gories of em­u­la­tions, bio-hu­mans could main­tain to­tal con­trol of the gov­ern­ment and econ­omy only if the em­u­la­tions reg­u­larly sub­mit­ted “to geno­cide, even though the over­whelm­ing ma­jor­ity of the pop­u­la­tion ex­pects the same thing to hap­pen to it soon.”302

…Robin thinks that if we be­haved in­tel­li­gently and main­tained good re­la­tions with the em­u­la­tions, bio-hu­mans could safely take up to around 5% of the world’s eco­nomic out­put with­out hav­ing the em­u­la­tions seek to de­stroy us. By ap­pro­pri­at­ing 5% rather than the pre­pon­der­ance of the world’s in­come, we would en­sure that the em­u­la­tions would have less to gain from killing us and tak­ing our stuff. But as power flows from mon­ey, hav­ing less in­come would make us less able to de­fend our­selves from any em­u­la­tions that did wish to strip us of our wealth. Robin is op­ti­mistic about our abil­ity to keep this 5%. He cor­rectly notes that many times in his­tory wealthy but weak groups have man­aged to keep their prop­erty for long pe­ri­ods of time. For ex­am­ple, many Amer­i­cans over the age of sev­enty are rich even though they no longer con­tribute to eco­nomic pro­duc­tion. These Amer­i­cans, if stand­ing with­out al­lies, would not have the slight­est chance of pre­vail­ing in a fight in which Amer­i­cans in their twen­ties joined to­gether to steal the prop­erty of se­niors. Yet it’s al­most in­con­ceiv­able that this would hap­pen. Sim­i­lar­ly, in many so­ci­eties through­out hu­man his­to­ry, rich se­nior cit­i­zens have en­joyed se­cure prop­erty rights even though they would quickly lose their wealth if enough younger men col­luded to take it from them. Even se­nior cit­i­zens whom de­men­tia has made much less in­tel­li­gent than most of their coun­try­men are still usu­ally able to re­tain their prop­er­ty. Robin men­tioned to me that tourists from rich coun­tries are gen­er­ally se­cure when they travel to poor na­tions even when the tourists are clearly un­de­fended wealthy out­siders. A wealthy white Amer­i­can wear­ing ex­pen­sive West­ern clothes could prob­a­bly walk safely through most African vil­lages even if the vil­lagers knew that the Amer­i­can earns more in a day than a vil­lager does in a year.

…As Robin points out, through­out hu­man his­tory most re­volts broke out when con­di­tions had im­proved for the poor­est in so­ci­ety.305 And great rev­o­lu­tions have al­most in­vari­ably been led by the rich. George Wash­ing­ton and Thomas Jeffer­son were wealthy landown­ers; Lenin’s fa­ther, born a serf, had risen through gov­ern­ment ser­vice to the rank of a no­ble­man and mar­ried a woman of wealth, and Trot­sky’s fa­ther was an il­lit­er­ate but pros­per­ous land­lord; Julius Cae­sar was the first- or sec­ond-wealth­i­est in­di­vid­ual alive at the time he over­threw the Ro­man Re­pub­lic; and the mutiny on the Bounty was led by an offi­cer. Per­haps bio-hu­mans would have more to fear from the small num­ber of wealthy em­u­la­tions than from the em­u­la­tions fac­ing star­va­tion.

  1. This line stayed with me after watch­ing —one does not care, in­deed.↩︎

  2. “…Once we have taken on a defi­nite form, we do not lose it un­til death.” –Chap­ter 2 of the ; trans­la­tion in Vi­tal­i­ty, En­er­gy, Spir­it: A Taoist Source­book (1991), ISBN 978-0877735199↩︎

  3. One of the few good bits of Kathryn Schulz’s 2011 book Be­ing Wrong (part 1) is where she does a more read­able ver­sion of Wittgen­stein’s ob­ser­va­tion (PI Pt II, p. 162), “One can mis­trust one’s own sens­es, but not one’s own be­lief. If there were a verb mean­ing”to be­lieve false­ly," it would not have any [mean­ing­ful] first per­son, present in­dica­tive." Her ver­sion goes:

    But be­fore we can plunge into the ex­pe­ri­ence of be­ing wrong, we must pause to make an im­por­tant if some­what per­verse point: there is no ex­pe­ri­ence of be­ing wrong.

    There is an ex­pe­ri­ence of re­al­iz­ing that we are wrong, of course. In fact, there is a stun­ning di­ver­sity of such ex­pe­ri­ences. As we’ll see in the pages to come, rec­og­niz­ing our mis­takes can be shock­ing, con­fus­ing, fun­ny, em­bar­rass­ing, trau­mat­ic, plea­sur­able, il­lu­mi­nat­ing, and life-al­ter­ing, some­times for ill and some­times for good. But by de­fi­n­i­tion, there can’t be any par­tic­u­lar feel­ing as­so­ci­ated with sim­ply be­ing wrong. In­deed, the whole rea­son it’s pos­si­ble to be wrong is that, while it is hap­pen­ing, you are obliv­i­ous to it. When you are sim­ply go­ing about your busi­ness in a state you will later de­cide was delu­sion­al, you have no idea of it what­so­ev­er. You are like the coy­ote in the car­toons, after he has gone off the cliff but be­fore he has looked down. Lit­er­ally in his case and fig­u­ra­tively in yours, you are al­ready in trou­ble when you feel like you’re still on solid ground. So I should re­vise my­self: it does feel like some­thing to be wrong. It feels like be­ing right.

  4. “You’re only as young as the last time you changed your mind.” – (quoted in Office Yo­ga: Sim­ple Stretches for Busy Peo­ple (2000) by Dar­rin Zeer, p. 52)↩︎

  5. “Every­one thinks they’ve won the Mag­i­cal Be­lief Lot­tery. Every­one thinks they more or less have a han­dle on things, that they, as op­posed to the bil­lions who dis­agree with them, have some­how lucked into the one true be­lief sys­tem.” –, Neu­ropath↩︎

  6. From “Bayesian Meth­ods: Gen­eral Back­ground”:

    As soon as we look at the na­ture of in­fer­ence at this many-moves-a­head level of per­cep­tion, our at­ti­tude to­ward prob­a­bil­ity the­ory and the proper way to use it in sci­ence be­comes al­most di­a­met­ri­cally op­po­site to that ex­pounded in most cur­rent text­books. We need have no fear of mak­ing shaky cal­cu­la­tions on in­ad­e­quate knowl­edge; for if our pre­dic­tions are in­deed wrong, then we shall have an op­por­tu­nity to im­prove that knowl­edge, an op­por­tu­nity that would have been lost had we been too timid to make the cal­cu­la­tions.

    In­stead of fear­ing wrong pre­dic­tions, we look ea­gerly for them; it is only when pre­dic­tions based on our present knowl­edge fail that prob­a­bil­ity the­ory leads us to fun­da­men­tal new knowl­edge.

    From Wittgen­stein’s Cul­ture and Value, MS 117 168 c: 17.2.1940:

    You can’t be re­luc­tant to give up your lie & still tell the truth.

  7. As quoted in page 207 of Genna Sosonko’s Russ­ian Sil­hou­ettes of :

    He did not dis­solve and he did not change. On the last pages of the book he is still the same Misha Botvin­nik, pupil of the 157th School of United Work­ers in Leningrad and Kom­so­mol mem­ber. He had not changed at all for sev­enty years, and, lis­ten­ing to his sin­cere and pas­sion­ate mono­logue, one in­vol­un­tar­ily thinks of Con­fu­cius: “Only the most clever and the most stu­pid can­not change.”

  8. Socrates, Pla­to’s , 458a (Zeyl trans­la­tion)↩︎

  9. I have a sim­i­lar ab­sence of story for my gen­er­ally be­liefs, since I was born hear­ing-im­paired and grew up us­ing hear­ing aids. That tech­nol­ogy could im­prove my nat­ural con­di­tion, that the flesh was im­per­fect, or that sci­en­tific & tech­no­log­i­cal ad­vance­ment was a good thing were not things I ever had to be ar­gued into be­liev­ing: I re­ceived those lessons daily when I took off my hear­ing aids, the world went silent, and I could no longer un­der­stand any­one around me. Or for that mat­ter, when my hear­ing aids were pe­ri­od­i­cally re­placed & up­grad­ed. And how could I re­gard cy­borgs as bizarre or hor­ri­ble things when I my­self verged on the cy­borg?↩︎

  10. One such ar­gu­ment is that mir­a­cles don’t work be­cause we are too skep­ti­cal or don’t be­lieve faith­fully enough (this ar­gu­ment is also used in para­psy­chol­o­gy, oddly enough). This is ab­surd, since much of the point of mir­a­cles in the Bible (and es­pe­cially by saints & mis­sion­ar­ies) was to con­vert in­fi­dels or skep­tics or wa­ver­ing be­liev­ers; and es­pe­cially rings hol­low when peo­ple like Pat Robert­son at­tempt to ex­plain why mir­a­cles are gen­er­ally re­ported from poor, su­per­sti­tious, and un­e­d­u­cated con­tem­po­rary ar­eas like Africa:

    Cause peo­ple over­seas did­n’t go to Ivy League schools! [chuck­les] Well, we’re so so­phis­ti­cat­ed. We think we’ve got every­thing fig­ured out. We know about evo­lu­tion, we know about Dar­win, we know about all these things that say God is­n’t re­al. We know about all this stuff and if we’ve been in many schools, the more ad­vanced schools, we have been in­un­dated with skep­ti­cism and sec­u­lar­ism. And over­seas they’re sim­ple, hum­ble, you tell them God loves them and they say “okay he loves me.” And you tell them God will do mir­a­cles and they say “okay, we be­lieve you.” And that’s what God’s look­ing for. That’s why they have mir­a­cles.

  11. What­ever the truth may be, I stand staunchly by this point: a kid sees so much ev­i­dence and be­lief in God that he ought to ra­tio­nally be­lieve. Men­cius Mold­bug:

    Most peo­ple are the­ists not be­cause they were ‘rea­soned into’ be­liev­ing in God, but be­cause they ap­plied Oc­cam’s ra­zor at too early an age. Their sim­plest ex­pla­na­tion for the rea­son that their par­ents, not to men­tion every­one else in the world, be­lieved in God, was that God ac­tu­ally ex­ist­ed. The same could be said for, say, Aus­tralia. Den­net­t’s ap­proach, which of course is prob­a­bly in­effec­tive in al­most all cas­es, is to ex­plain why, if God does­n’t ex­ist, every­one knows who He is. How did this whole God thing hap­pen? Why is it not weird that peo­ple be­lieved in Him for 2000 years, but ac­tu­ally they were wrong?

  12. , Korb 2003:

    In some cir­cles (or cir­cum­stances) it is pop­u­larly be­lieved that there are witch­es; in oth­ers, it is be­lieved that hair­less aliens walk the plan­et. If a bald ap­peal to the pop­u­lar­ity of a be­lief were enough to es­tab­lish its ac­cept­abil­i­ty, then rea­son­able be­liefs and the ar­gu­ments for them would ebb and flow with the tides of fash­ion. Nev­er­the­less, there seems to be some merit to the ap­peal to pop­u­lar be­lief. John­son (1996) points out a di­rect rel­e­vance be­tween pop­u­lar be­lief and (propo­si­tions con­cern­ing) the out­come of de­mo­c­ra­tic elec­tions! But even when there is no di­rect (or in­di­rect) causal chain lead­ing from pop­u­lar be­lief to the truth of a propo­si­tion, there may well be a com­mon cause that re­lates the two, mak­ing the pop­u­lar­ity of a be­lief a (pos­si­bly mi­nor) rea­son to shift one’s be­lief in a con­clu­sion. Pre­sum­ably, a world in which no one be­lieves in witches would sup­port a mod­er­ately smaller ra­tio­nal de­gree of be­lief in them than one in which many do, at least prior to the de­vel­op­ment of sci­ence…Pop­u­lar­ity of a be­lief may well in gen­eral be as­so­ci­ated with the truth of what is be­lieved; so, lack­ing any clear sci­en­tific judg­ment (say, dur­ing the Dark Ages), com­mon be­lief in the effi­cacy of witch­craft may well ra­tio­nally lift our own be­lief, if only slight­ly. Nev­er­the­less, given an im­proved un­der­stand­ing of nat­ural phe­nom­ena and the fal­li­bil­ity of hu­man be­lief for­ma­tion (per­haps some time in the fu­ture!), the pop­u­lar be­lief is no longer rel­e­vant for de­cid­ing whether witches ex­ist or not: sci­ence ac­counts for both the be­lief in witches and their un­re­al­i­ty.

  13. From a the­ol­ogy blog, “Trust in tes­ti­mony and mir­a­cles”:

    …Har­ris found that chil­dren do not fall into ei­ther pat­tern. Pace the Humean ac­count, he found that young chil­dren are read­ily in­clined to be­lieve ex­tra­or­di­nary claims, such as that there are in­vis­i­ble or­gan­isms on your hands that can make you ill and that you need to wash off, and that there is a man who vis­its you each 24th De­cem­ber to bring presents and candy if you are nice (see e.g., , Child De­vel­op­ment, 77, 505-524). But chil­dren are not blindly cred­u­lous ei­ther, as Reid sup­posed. In a se­ries of ex­per­i­ments, Har­ris could show that even chil­dren of 24 months pay at­ten­tion to the re­li­a­bil­ity of the tes­ti­fi­er. When they see two peo­ple, one of which sys­tem­at­i­cally mis­names known ob­jects (e.g., say­ing “that’s a bear”, while pre­sent­ing a bot­tle), tod­dlers are less likely to trust later ut­ter­ances by these un­re­li­able speak­ers (when they name un­fa­mil­iar ob­ject­s), and more likely to trust peo­ple who sys­tem­at­i­cally gave ob­jects their cor­rect names (see e.g., Paul L. Har­ris and Kath­leen H. Cor­riveau Phil. Trans. R. Soc. B 2011 366, 1179-1187.) Ex­per­i­ments by Mills and Keil show that 6-year-olds al­ready take into ac­count a tes­ti­fier’s self­-in­ter­est: they are more likely to be­lieve some­one who says he lost a race than some­one who says he won it ( Psy­cho­log­i­cal Sci­ence 2005 16: 385).

  14. I some­times won­der if this had any­thing to do with my later phi­los­o­phy train­ing; athe­ists make up some­thing like 70% of re­spon­dents to the Philpa­pers sur­vey, and a crit­i­cal ‘re­flec­tive’ style both cor­re­lates with and causes ; an­other in­ter­est­ing cor­re­la­tion is that peo­ple on the autism spec­trum (which I have often been told I must surely be on) seem to be heav­ily ag­nos­tic or athe­is­tic.

    For dis­cus­sion of these points, see:

    1. “Athe­ism & the autism spec­trum”
    2. “Cog­ni­tive style tends to pre­dict re­li­gious con­vic­tion”
    3. “On the eti­ol­ogy of re­li­gious be­lief”
  15. One of my fa­vorite books on mod­ern re­li­gion is Luhrman­n’s 2012 When God Talks Back. It’s not that it’s fan­tas­ti­cally writ­ten or re­searched, al­though I do like books where the au­thor has done re­search them­selves on the topic and cite a rea­son­able num­ber of claims. I like it be­cause, as the last chap­ter says, it pro­vides a large part of the an­swer to the “non­be­liev­er’s ques­tion”: in an age of zero mir­a­cles be­yond the ris­i­ble (“my tu­mor went away after I prayed!”), with no gods thun­der­ing to crowds, and with the best philo­soph­i­cal ar­gu­ments con­tent­ing them­selves with the log­i­cal pos­si­bil­ity of the god of the philoso­phers, how could any­one sin­cerely be­lieve in su­per­nat­ural be­ings and why is­n’t a sort of prac­ti­cal ag­nos­ti­cism (“yeah, I don’t re­ally be­lieve, but church is where all my so­cial ac­tiv­i­ties are”) uni­ver­sal? Why are there so many fer­vent be­liev­ers and some re­li­gions spread­ing rapidly while hold­ing fairly con­stant in highly de­vel­oped in­dus­tri­al­ized coun­tries? The book an­swers that this ab­sence is partly il­lu­so­ry: they do hear God’s voice, through a va­ri­ety of au­to-sug­ges­tive med­i­ta­tive prac­tices which col­lec­tively con­sti­tute the that athe­ists are ac­cused of sadly lack­ing, which com­bined with the other fac­tors (the in­tu­itive­ness of su­per­nat­ural be­ings pace the re­search in kids and evo-psych rea­son­ing, the sup­pres­sion of an­a­lytic thought, the so­cial ben­e­fits, etc) main­tains re­li­gion at its his­tor­i­cal pop­u­lar­ity in de­vel­oped coun­tries and spurs new growth in de­vel­op­ing coun­tries. (Africa is grow­ing fan­tas­ti­cally for both Is­lam and Chris­tian­i­ty; and other de­vel­op­ing coun­tries ex­pe­ri­ence their own ver­sions of Japan’s “rush hour of the gods”.)

    Ex­cerpts from Luhrmann 2012: pref­ace / 1 / 2 / 3 / 4 / 5 / 6 / 7 / 8 / 9 / 10↩︎

  16. pg 120, When Lon­don was Cap­i­tal of Amer­ica, Julie Flavell 2011:

    The British gov­ern­ment hoped that a west sealed off from en­croach­ments by whites, and where traders had to op­er­ate un­der the watch­ful eye of a British army de­tach­ment, would bring about good re­la­tions with the In­di­ans. To the great dis­con­tent of spec­u­la­tors, in 1761 it was an­nounced that all ap­pli­ca­tions for land grants now had to go to Lon­don; no colo­nial gov­ern­ment could ap­prove them. The Procla­ma­tion of 1763 banned west­ward set­tle­ment al­to­gether and in­stead en­cour­aged colonists who wanted new lands to set­tle to the north in Que­bec, and to the south in Flori­da. Within just a few years British min­is­ters would be re­treat­ing from the Procla­ma­tion and grant­ing west­ern lands.

  17. From Wikipedia: ‘It re­mains the dead­liest war in Amer­i­can his­to­ry, re­sult­ing in the deaths of 620,000 sol­diers and an un­de­ter­mined num­ber of civil­ian ca­su­al­ties. Ac­cord­ing to John Hud­dle­ston, “10% of all North­ern males 20-45 years of age died, as did 30% of all South­ern white males aged 18-40.”’↩︎

  18. An In­quiry into the Na­ture and Causes of the Wealth of Na­tions , “Book IV: On Sys­tems of Po­lit­i­cal Econ­omy”:

    When the Act of Nav­i­ga­tion was made, though Eng­land and Hol­land were not ac­tu­ally at war, the most vi­o­lent an­i­mos­ity sub­sisted be­tween the two na­tion­s…They are as wise, how­ev­er, as if they had all been dic­tated by the most de­lib­er­ate wis­dom. Na­tional an­i­mos­ity at that par­tic­u­lar time aimed at the very same ob­ject which the most de­lib­er­ate wis­dom would have rec­om­mend­ed, the diminu­tion of the naval power of Hol­land, the only naval power which could en­dan­ger the se­cu­rity of Eng­land.

    The Act of Nav­i­ga­tion is not favourable to for­eign com­merce, or to the growth of that op­u­lence which can arise from it. The in­ter­est of a na­tion in its com­mer­cial re­la­tions to for­eign na­tions is, like that of a mer­chant with re­gard to the differ­ent peo­ple with whom he deals, to buy as cheap and to sell as dear as pos­si­ble. But it will be most likely to buy cheap, when by the most per­fect free­dom of trade it en­cour­ages all na­tions to bring to it the goods which it has oc­ca­sion to pur­chase; and, for the same rea­son, it will be most likely to sell dear, when its mar­kets are thus filled with the great­est num­ber of buy­ers. The Act of Nav­i­ga­tion, it is true, lays no bur­den upon for­eign ships that come to ex­port the pro­duce of British in­dus­try. Even the an­cient aliens’ du­ty, which used to be paid upon all goods ex­ported as well as im­port­ed, has, by sev­eral sub­se­quent acts, been taken off from the greater part of the ar­ti­cles of ex­por­ta­tion. But if for­eign­ers, ei­ther by pro­hi­bi­tions or high du­ties, are hin­dered from com­ing to sell, they can­not al­ways afford to come to buy; be­cause com­ing with­out a car­go, they must lose the freight from their own coun­try to Great Britain. By di­min­ish­ing the num­ber of sell­ers, there­fore, we nec­es­sar­ily di­min­ish that of buy­ers, and are thus likely not only to buy for­eign goods dear­er, but to sell our own cheap­er, than if there was a more per­fect free­dom of trade. As de­fence, how­ever it is of much more im­por­tance than op­u­lence, the Act of Nav­i­ga­tion is, per­haps, the wis­est of all the com­mer­cial reg­u­la­tions of Eng­land.

  19. Ni­et­zsche writes this sum­mary of tra­di­tional philoso­phers to mock them, but is­n’t there a great deal of truth in it? “How could any­thing orig­i­nate out of its op­po­site? Truth out of er­ror or the pure and sun­like gaze of the sage out of lust? Such ori­gins are im­pos­si­ble; who­ever dreams of them is a fool.”↩︎

  20. Friedrich Hayek, The Con­sti­tu­tion of Lib­erty (1960)↩︎

  21. The rhetoric in the 1990s and early 2000s is amaz­ing to read in ret­ro­spect; some of the claims were about as wrong as it is pos­si­ble to be. For ex­am­ple, the CEO of Mil­len­nium Phar­ma­ceu­ti­cals - not at all a small or fly­-by-night phar­ma­corp - said in 2000 it had high hopes for 6 drugs in hu­man tri­als and claimed in 2002 that thanks to ge­netic re­search it would have 1-2 drugs en­ter­ing tri­als every year within 3 years, for 6-12 new drugs by 2011. As of Oc­to­ber 2011, it has ex­actly 1 ap­proved drug.↩︎

  22. The sub­se­quently cited re­view cov­ers this; al­most all of the fa­mous in­crease in longevity by decades is due to the young:

    Ta­ble 1 shows the av­er­age num­ber of years of life re­main­ing from 1900 to 2007 from var­i­ous ages, com­bin­ing both sexes and eth­nic groups. From birth, life ex­pectancy in­creased from 49.2 years (pre­vi­ously es­ti­mated at 47.3 years in these same sources) in 1900 to 77.9 in 2007, a gain of life ex­pectancy of nearly 29 years and a prodi­gious ac­com­plish­ment. The in­crease was largely due to de­clines in peri­na­tal mor­tal­ity and re­duc­tion in in­fec­tious dis­eases which affected mainly younger per­sons. Over this pe­ri­od, de­vel­oped na­tions moved from an era of acute in­fec­tious dis­ease to one dom­i­nated by chronic ill­ness. As a re­sult, life ex­ten­sion from age 65 was in­creased only 6 years over the en­tire 20th cen­tu­ry; from age 75 gains were only 4.2 years, from age 85 only 2.3 years and from age 100 a sin­gle year. From age 65 over the most re­cent 20 years, the gain has been about a year [16].

    Much con­fu­sion in longevity pre­dic­tions comes from us­ing pro­jec­tions of life ex­pectancy at birth to es­ti­mate fu­ture pop­u­la­tion longevity [18]. For ex­am­ple, “If the pace of in­crease in life ex­pectancy (from birth) for de­vel­oped coun­tries over the past two cen­turies con­tin­ues through the 21st cen­tu­ry, most ba­bies born since 2000 will cel­e­brate their 100th birth­days” . Note from the 100-year line of Ta­ble 1 that life ex­pectan­cies for cen­te­nar­i­ans would be pro­jected to rise only one year in the 21st cen­tu­ry, as in the 20th. Such at­ten­tion-grab­bing state­ments fol­low from pro­ject­ing from birth rather than age 65, thus in­clud­ing in­fant and early life events to project “se­nior” ag­ing, us­ing data from women rather than both gen­ders com­bined, cher­ry-pick­ing the best data for each year, ne­glect­ing to com­pute effects of in­-mi­gra­tion and out­-mi­gra­tion, and oth­ers.

    Re­mark­ably, some groups show a de­crease in longevi­ty; a cen­te­nar­ian in 1980 has an av­er­age re­main­ing lifes­pan of 2.7 years, but in 2000, that has fallen to 2.6. There was an even larger re­ver­sal in 1940 (2.1) to 1960 (1.9). Younger groups show larger gains (eg. 85-year-olds had 6.0 years in 1980 and 6.3 in 2000), ev­i­dence for .↩︎

  23. If ex­po­nen­tials & sig­moids re­ally do ex­plain Ama­ra’s ob­ser­va­tion, that im­plies that there ought to be some sort of “re­verse” Amara effect: where an ob­server is at the top of the sig­moid and naively ex­trap­o­lates that the long run will look very differ­ent from now - and it turns out that the long run looks iden­ti­cal to right now. Iden­ti­fy­ing re­verse Amara effects is eas­ier than reg­u­lar Amara effects be­cause one has the ben­e­fit of hind­sight. For ex­am­ple, nu­clear en­er­gy: it was ini­tially a puz­zling physics anom­aly and a re­search prob­lem at most - ura­nium was im­por­tant to the world econ­omy be­cause it was used in things like the which helped en­able World War I. Even in the 1930s, it was more in­ter­est­ing than use­ful. Then sud­denly nu­clear re­ac­tors were demon­strated and atomic bombs trans­form the world in the 1940s, lead­ing to wide­spread fu­tur­ism pre­dic­tions of ubiq­ui­tous nu­clear en­ergy “too cheap to me­ter”, but fur­ther use of nu­clear tech­nolo­gies sud­denly breaks down; with the hon­or­able ex­cep­tion of nu­clear med­i­cine, the world in the 2010s looks pretty much iden­ti­cal to the 1950s. Given this, one would not be very sur­prised if in 2112, nu­clear tech­nolo­gies were noth­ing but re­fine­ments of 2012 nu­clear tech­nolo­gies.↩︎

  24. Monte Carlo trees are very sim­i­lar to the tech­niques used in one of the com­putable im­ple­men­ta­tions of AIXI, in­ci­den­tal­ly: (back­ground).↩︎

  25. This is known as the ‘over­hang’ ar­gu­ment. The de­vel­op­ment and canon­i­cal form of it is un­clear; it may sim­ply be Sin­guli­tar­ian folk­lore-knowl­edge. Eliezer Yud­kowsky, from the 2008 “Hard Take­off”:

    Or con­sider the no­tion of sud­den re­source bo­nan­zas. Sup­pose there’s a semi­-so­phis­ti­cated Ar­ti­fi­cial Gen­eral In­tel­li­gence run­ning on a clus­ter of a thou­sand CPUs. The AI has not hit a wall - it’s still im­prov­ing it­self - but its self­-im­prove­ment is go­ing so slowly that, the AI cal­cu­lates, it will take an­other fifty years for it to en­gi­neer / im­ple­ment / re­fine just the changes it cur­rently has in mind. Even if this AI would go FOOM even­tu­al­ly, its cur­rent progress is so slow as to con­sti­tute be­ing flat­lined…

    So the AI turns its at­ten­tion to ex­am­in­ing cer­tain blobs of bi­nary code - code com­pos­ing op­er­at­ing sys­tems, or routers, or DNS ser­vices - and then takes over all the poorly de­fended com­put­ers on the In­ter­net. This may not re­quire what hu­mans would re­gard as ge­nius, just the abil­ity to ex­am­ine lots of ma­chine code and do rel­a­tively low-grade rea­son­ing on mil­lions of bytes of it. (I have a say­ing/hy­poth­e­sis that a hu­man try­ing to write code is like some­one with­out a vi­sual cor­tex try­ing to paint a pic­ture - we can do it even­tu­al­ly, but we have to go pixel by pixel be­cause we lack a sen­sory modal­ity for that medi­um; it’s not our na­tive en­vi­ron­men­t.) The Fu­ture may also have more le­gal ways to ob­tain large amounts of com­put­ing power quick­ly.

    …A sub­tler sort of hard­ware over­hang, I sus­pect, is rep­re­sented by mod­ern CPUs have a 2GHz se­r­ial speed, in con­trast to neu­rons that spike 100 times per sec­ond on a good day. The “hun­dred-step rule” in com­pu­ta­tional neu­ro­science is a rule of thumb that any pos­tu­lated neural al­go­rithm which runs in re­al­time has to per­form its job in less than 100 se­r­ial steps one after the oth­er. We do not un­der­stand how to effi­ciently use the com­puter hard­ware we have now, to do in­tel­li­gent think­ing. But the much-vaunted “mas­sive par­al­lelism” of the hu­man brain, is, I sus­pect, mostly cache lookups to make up for the sheer awk­ward­ness of the brain’s se­r­ial slow­ness - if your com­puter ran at 200Hz, you’d have to re­sort to all sorts of ab­surdly mas­sive par­al­lelism to get any­thing done in re­al­time. I sus­pect that, if cor­rectly de­signed, a mid­size com­puter clus­ter would be able to get high­-grade think­ing done at a se­r­ial speed much faster than hu­man, even if the to­tal par­al­lel com­put­ing power was less.

    So that’s an­other kind of over­hang: be­cause our com­put­ing hard­ware has run so far ahead of AI the­ory, we have in­cred­i­bly fast com­put­ers we don’t know how to use for think­ing; get­ting AI right could pro­duce a huge, dis­con­tin­u­ous jolt, as the speed of high­-grade thought on this planet sud­denly dropped into com­puter time.

    A still sub­tler kind of over­hang would be rep­re­sented by hu­man fail­ure to use our gath­ered ex­per­i­men­tal data effi­ciently. [A bet­ter link on in­effi­cient hu­man in­duc­tion might be Phil Goet­z’s “In­for­ma­tion The­ory and FOOM.]

    An­ders Sand­berg & Carl Shul­man gave a 2010 talk on it; from the blog post:

    We give an ar­gu­ment for why - if the AI sin­gu­lar­ity hap­pens - an early sin­gu­lar­ity is likely to be slower and more pre­dictable than a late-oc­cur­ring one….

    If you are on the hard­ware side, how much hard­ware do you be­lieve will be avail­able when the first hu­man level AI oc­curs? You should ex­pect the first AI to be pretty close to the lim­its of what re­searchers can afford: a project run­ning on the fu­ture coun­ter­part to Se­quoia or the Google servers. There will not be much ex­tra com­put­ing power avail­able to run more copies. An in­tel­li­gence ex­plo­sion will be bounded by the growth of more hard­ware.

    If you are on the soft­ware side, you should ex­pect that hard­ware has con­tin­ued to in­crease after pass­ing “hu­man equiv­a­lence”. When the AI is fi­nally con­structed after all the hu­man and con­cep­tual bot­tle­necks have passed, hard­ware will be much bet­ter than needed to just run a hu­man-level AI. You have a “hard­ware over­hang” al­low­ing you to run many copies (or fast or big ver­sions) im­me­di­ately after­wards. A rapid and sharp in­tel­li­gence ex­plo­sion is pos­si­ble.

    This leads to our con­clu­sion: if you are an op­ti­mist about soft­ware, you should ex­pect an early sin­gu­lar­ity that in­volves an in­tel­li­gence ex­plo­sion that at the start grows “just” as Moore’s law (or its suc­ces­sor). If you are a pes­simist about soft­ware, you should ex­pect a late sin­gu­lar­ity that is very sharp. It looks like it is hard to co­her­ently ar­gue for a late but smooth sin­gu­lar­i­ty.

    …Note that sharp, un­pre­dictable sin­gu­lar­i­ties are dan­ger­ous. If the break­through is sim­ply a mat­ter of the right in­sights and ex­per­i­ments to fi­nally co­here (after end­less dis­ap­point­ing per­for­mance over a long time) and then will lead to an in­tel­li­gence ex­plo­sion nearly in­stant­ly, then most so­ci­eties will be un­pre­pared, there will be lit­tle time to make the AIs docile, there are strong first-mover ad­van­tages and in­cen­tives to com­pro­mise on safe­ty. A recipe for some nasty dy­nam­ics.

    Jaan Tallinn in 2011:

    It’s im­por­tant to note that with every year the AI al­go­rithm re­mains un­solved, the hard­ware marches to the beat of Moore’s Law - cre­at­ing a mas­sive hard­ware over­hang. The first AI is likely to find it­self run­ning on a com­puter that’s sev­eral or­ders of mag­ni­tude faster than needed for hu­man level in­tel­li­gence. Not to men­tion that it will find an In­ter­net worth of com­put­ers to take over and re­tool for its pur­pose.

  26. Or Robin Han­son’s pa­per, “Eco­nomic Growth Given Ma­chine In­tel­li­gence”:

    Ma­chines com­ple­ment hu­man la­bor when they be­come more pro­duc­tive at the jobs they per­form, but ma­chines also sub­sti­tute for hu­man la­bor by tak­ing over hu­man jobs. At first, ex­pen­sive hard­ware and soft­ware does only the few jobs where com­put­ers have the strongest ad­van­tage over hu­mans. Even­tu­al­ly, com­put­ers do most jobs. At first, com­ple­men­tary effects dom­i­nate, and hu­man wages rise with com­puter pro­duc­tiv­i­ty. But even­tu­ally sub­sti­tu­tion can dom­i­nate, mak­ing wages fall as fast as com­puter prices now do. An in­tel­li­gence pop­u­la­tion ex­plo­sion makes per-in­tel­li­gence con­sump­tion fall this fast, while eco­nomic growth rates rise by an or­der of mag­ni­tude or more.

  27. In­tu­itive­ly, one would guess that the value of ed­u­ca­tion and changes in it value would fol­low some sort of lin­ear or ex­po­nen­tial - more is bet­ter, less is worse. If the value of a high school diploma in­creas­es, an un­der­grad­u­ate ought to in­crease more, and post­grad­u­ate de­grees in­crease even more, right? A ‘hol­low­ing-out’ mod­el, on the other hand, would seem to pre­dict that there would be a sort of U-curve where the mediocre ed­u­ca­tion is not worth what it costs and one would be bet­ter off not both­er­ing with get­ting more ed­u­ca­tion or stick­ing it out and get­ting a ‘real’ de­gree. With that in mind, it is in­ter­est­ing to look at the Cen­sus data:

    In fact, new Cen­sus Bu­reau data show that if you di­vide the pop­u­la­tion by ed­u­ca­tion, on av­er­age wages have risen only for those with grad­u­ate de­grees over the past 10 years. (On av­er­age, of course, means that some have done bet­ter and some have done worse.) Here (thanks to econ­o­mist Matthew Slaugh­ter of Dart­mouth Col­lege’s Tuck School of Busi­ness) are changes in U.S. work­ers wages as re­ported in the lat­est Cen­sus Bu­reau re­port, ad­justed for in­fla­tion us­ing the CPI-U-RS mea­sure rec­om­mended by the Bu­reau of La­bor Sta­tis­tics:

    “Change be­tween 2000 and 2010 in in­fla­tion-ad­justed av­er­age earn­ings by ed­u­ca­tional at­tain­ment”
  28. Charles Mur­ray re­port­edly cites sta­tis­tics in Com­ing Apart: The State of White Amer­ica 1960-2010 that the dis­abil­ity rate for men - work­ing class - was 2% in 1960; with more than half a cen­tury of med­ical pro­gress, the rate has not fallen but risen to 10%.↩︎

  29. And there have al­ways been ru­mors that the moral haz­ard is sub­stan­tial; eg. the psy­chi­a­trist Steve Balt, “How To Re­tire At Age 27” and com­men­tary.↩︎

  30. From pg 13/340 of Bowles & Jayadev 2006:

    Other differ­ences in tech­nol­ogy (or differ­ent dis­tri­b­u­tions of la­bor across sec­tors of the econ­o­my) may ac­count for some of the differ­ences. How­ev­er, the data on su­per­vi­sion in­ten­sity by man­u­fac­tur­ing sec­tor in five sub­-Sa­ha­ran African coun­tries shown in Ta­ble 4 sug­gest large coun­try effects in­de­pen­dent of the com­po­si­tion of out­put. Su­per­vi­sory in­ten­si­ties in Zam­bi­a’s ‘wood and fur­ni­ture’ and ‘food pro­cess­ing’ in­dus­tries, are twice and five times Ghana’s re­spec­tive­ly. A coun­try-and-in­dus­try fixed effects re­gres­sion in­di­cates that Zam­bi­a’s su­per­vi­sion in­ten­sity con­di­tioned on in­dus­trial struc­ture is two and a half times Ghana’s. Of course these differ­ences could re­flect within sec­tor vari­a­tion among coun­tries in out­put com­po­si­tion or tech­nolo­gies, but there is no way to de­ter­mine how much (if any) of the es­ti­mated coun­try effects are due to this. We also ex­plored if su­per­vi­sion in­ten­sity was re­lated to more ad­vanced tech­nolo­gies gener­i­cal­ly. How­ev­er, in the ad­vanced econ­omy dataset (shown in Ta­ble 2) the value added of knowl­edge in­ten­sive sec­tors as a share of gross value added was sub­stan­tially un­cor­re­lated with the su­per­vi­sory ra­tio (r = 0.14).

    While the data are in­ad­e­quate to pro­vide a com­pelling test of the hy­poth­e­sis, we thus find lit­tle ev­i­dence that the in­crease in guard la­bor in the U.S. or the differ­ences across the coun­tries is due to differ­ences in out­put com­po­si­tion and tech­nol­o­gy. A more likely ex­pla­na­tion is what we term ‘en­force­ment spe­cial­iza­tion’. Eco­nomic de­vel­op­ment pro­ceeds through a process of spe­cial­iza­tion and in­creas­ing di­vi­sion of labor; the work of per­pet­u­at­ing a so­ci­ety’s in­sti­tu­tions is no ex­cep­tion to this tru­is­m….Our data in­di­cate that the United States de­votes well over twice as large a frac­tion of its la­bor force to guard la­bor as does Switzer­land. This may oc­cur in part be­cause peer mon­i­tor­ing and in­for­mal sanc­tion­ing play a larger role in Switzer­land, as well as the fact that or­di­nary Swiss cit­i­zens have mil­i­tary de­fense ca­pac­i­ties and du­ties and are not counted in our data as sol­diers.

  31. A key ad­van­tage of the , ac­cord­ing to , was that it had an effi­cient tax sys­tem which en­abled it to sup­port a stand­ing mil­i­tary, which was able to be trained in horse-archery all the way up to steppe-no­mad stan­dards - a task which took years for the trainees who could man­age it at all. (In con­trast, the US mil­i­tary is happy to send many sol­diers into com­bat with only a few months of train­ing.)↩︎

  32. If you think that’s the whole mil­i­tary-in­dus­tri­al-in­tel­li­gence bud­get, you are quite naive.↩︎

  33. Wit­ness the mas­sive fights over the and un­usual mea­sures re­quired; the Con­gress­men aren’t stu­pid, they un­der­stand how valu­able the mil­i­tary-in­dus­trial wel­fare is for their com­mu­ni­ties.↩︎

  34. Im­pris­on­ment as a per­ma­nent pun­ish­ment was prior to the In­dus­trial Rev­o­lu­tion, and what pris­ons there were often were pri­mar­ily a mine or other fa­cil­ity of that kind; it is very ex­pen­sive to im­prison and only im­prison some­one, which is why tech­niques like fines (eg. North­ern Eu­rope), tor­ture (Chi­na), ex­ile (Greece) or (Eng­land & Aus­trali­a), or ex­e­cu­tion (ev­ery­one) were the usual meth­ods.↩︎

  35. For ex­am­ple, MIT econ­o­mist “ on In­equal­ity” has all the pieces but some­how es­cape the ob­vi­ous con­clu­sion:

    Let’s go through your books. Your first choice is The Race be­tween Ed­u­ca­tion and Tech­nol­ogy, pub­lished by Har­vard Uni­ver­sity Press. You men­tioned in an ear­lier email to me that it is “a must-read for any­one in­ter­ested in in­equal­ity”. Tell me more.

    This is a re­ally won­der­ful book. It gives a mas­ter­ful out­line of the stan­dard eco­nomic mod­el, where earn­ings are pro­por­tional to con­tri­bu­tion, or to pro­duc­tiv­i­ty. It high­lights in a very clear man­ner what de­ter­mines the pro­duc­tiv­i­ties of differ­ent in­di­vid­u­als and differ­ent groups. It takes its cue from a phrase that the fa­mous Dutch econ­o­mist, Jan Tin­ber­gen coined. The key idea is that tech­no­log­i­cal changes often in­crease the de­mand for more skilled work­ers, so in or­der to keep in­equal­ity in check you need to have a steady in­crease in the sup­ply of skilled work­ers in the econ­o­my. He called this “the race be­tween ed­u­ca­tion and tech­nol­ogy”. If the race is won by tech­nol­o­gy, in­equal­ity tends to in­crease, if the race is won by ed­u­ca­tion, in­equal­ity tends to de­crease.

    The au­thors, Clau­dia Goldin and Larry Katz, show that this is ac­tu­ally a pretty good model in terms of ex­plain­ing the last 100 years or so of US his­to­ry. They give an ex­cel­lent his­tor­i­cal ac­count of how the US ed­u­ca­tion sys­tem was formed and why it was very pro­gres­sive, lead­ing to a very large in­crease in the sup­ply of ed­u­cated work­ers, in the first half of the cen­tu­ry. This cre­ated greater equal­ity in the US than in many other parts of the world.

    They also point to three things that have changed that pic­ture over the last 30 to 40 years. One is that tech­nol­ogy has be­come even more bi­ased to­wards more skilled, higher earn­ing work­ers than be­fore. So, all else be­ing equal, that will tend to in­crease in­equal­i­ty. Sec­ond­ly, we’ve been go­ing through a phase of glob­al­i­sa­tion. Things such as trad­ing with China - where low-skill labour is much cheaper - are putting pres­sure on low wages. Third, and pos­si­bly most im­por­tant, is that the US ed­u­ca­tion sys­tem has been fail­ing ter­ri­bly at some lev­el. We haven’t been able to in­crease the share of our youth that com­pletes col­lege or high school. It’s re­ally re­mark­able, and most peo­ple would­n’t ac­tu­ally guess this, but in the US, the co­horts that had the high­est high­-school grad­u­a­tion rates were the ones that were grad­u­at­ing in the mid­dle of the 1960s. Our high­-school grad­u­a­tion rate has ac­tu­ally been de­clin­ing since then. If you look at col­lege, it’s the same thing. This is hugely im­por­tant, and it’s re­ally quite shock­ing. It has a ma­jor effect on in­equal­i­ty, be­cause it is mak­ing skills much more scarce then they should be.

    Do Goldin and Katz go into the rea­sons why ed­u­ca­tion is fail­ing in the US?

    They do dis­cuss it, but no­body knows. It’s not a mono­causal, sim­ple sto­ry. It’s not that we’re spend­ing less. In fact, we are spend­ing more. It’s cer­tainly not that col­lege is not val­ued, it’s val­ued a lot. The col­lege pre­mium - what col­lege grad­u­ates earn rel­a­tive to high­-school grad­u­ates - has been in­creas­ing rapid­ly. It’s not that the US is not in­vest­ing enough in low-in­come schools. There has been a lot of in­vest­ment in low-in­come schools. Not just free lunch­es, but lots of grants and other forms of spend­ing from both states and the fed­eral gov­ern­ment.

    The fail­ure of ed­u­ca­tion to in­crease may be masked by the dy­ing of the un­e­d­u­cated el­derly, but that is an effect that can only last so long. And then we will see some­thing like that looks more like this, a log graph which may be­gin pe­ter­ing out soon (and which looks like a di­min­ish­ing re­turns graph - every time unit sees less and less in­crease squeezed out as ad­di­tional efforts or larger re­turns are ap­plied to the pop­u­lace):

    Log-Rel­a­tive Sup­ply of Col­lege/non-Col­lege Labor, 1963-2008

    Or econ­o­mist , in a pod­cast, who iden­ti­fies the prob­lem and blames it on a de­crease in teacher qual­i­ty!

    You ar­gue that the Amer­i­can ed­u­ca­tion sys­tem, both K-12 and at the col­lege lev­els, has got some se­ri­ous prob­lems. Let’s talk about it. What’s wrong with it? And of course, as a re­sult, ed­u­ca­tion is a key part of in­no­va­tion and pro­duc­tiv­i­ty. If you don’t have a well-e­d­u­cated pop­u­lace you are not go­ing to have a very good econ­o­my. What’s wrong with our ed­u­ca­tion sys­tem?

    Let’s talk about K-12. Here’s two re­mark­able facts, which have just blown me away. Right now, in the United States, peo­ple 55-64 years old, they are more likely to have had a high school ed­u­ca­tion than 25-34 year olds. Just a lit­tle bit, but they are more like­ly. So, you look every­where in the world and what do you see? You see younger peo­ple hav­ing more ed­u­ca­tion than older peo­ple. Not true in the United States. That is a shock­ing claim. In­cred­i­ble. And the rea­son is that the drop-out rate has in­creased? Ex­act­ly. So, the high school dropout rate has in­creased. Now, 25% of males in the United States drop out of high school. And that’s in­creased since the 1960s, even as the prospects for a high school dropout have got­ten much worse. We’ve seen an in­crease, 21st cen­tury - 25% of males not grad­u­at­ing high school. That’s mind-bog­gling. Why? One of the un­der­ly­ing facts re­lat­ing to ed­u­ca­tion, which is [?], which is that the more ed­u­ca­tion you get on av­er­age - and I’m go­ing to talk about why on av­er­age can be very mis­lead­ing - high school grad­u­ates do bet­ter than high school dropouts; peo­ple with some col­lege do bet­ter than high school grad­u­ates; peo­ple grad­u­at­ing from col­lege do bet­ter than peo­ple with some col­lege; peo­ple with grad­u­ate de­grees do bet­ter than col­lege grads. And the differ­ences are large. Par­tic­u­larly if you com­pare a col­lege grad­u­ate to a high school dropout, there is an enor­mous differ­ence.

    So, nor­mally we would say: Well, this prob­lem kind of solves it­self. There’s a nat­ural in­cen­tive to stay in school, and I would­n’t worry about it. Why should we be wor­ry­ing about it? It does­n’t seem to be work­ing. Why is­n’t it work­ing and what could be done? I think there’s a few prob­lems. One is the qual­ity of teach­ers I think has ac­tu­ally gone down. So I think that’s a prob­lem. This is a case of every sil­ver lin­ing has a cloud, or some­thing like that, in that in 1970s about half of col­lege-e­d­u­cated women be­came teach­ers. This is at a time when there’s maybe 4% are get­ting an MBA, less than 10% are go­ing to med­ical school, go­ing to law school. These smart wom­en, they are be­com­ing teach­ers. Well, as we’ve opened up, by 1980 you’ve got 30% or so of the in­com­ing class of MBAs, doc­tors, lawyers, are women. Which is great. Their com­par­a­tive ad­van­tage, mov­ing into these fields, pro­duc­tiv­i­ty, and so forth. And yet that is meant that on av­er­age, the qual­ity of teach­ers, the qual­ity pool we are draw­ing from, has gone down in terms of their SAT lev­els and so forth. So, I think we need to fix that.

    Also rel­e­vant: “Lib­eral Arts Grads Win Long-Term”.↩︎

  36. “Over-E­d­u­ca­tion and the Skills of UK Grad­u­ates” (Cheva­lier & Lind­ley 2006):

    Be­fore the Eight­ies, Britain had one of the low­est par­tic­i­pa­tion rates in higher ed­u­ca­tion across OECD coun­tries. Con­se­quent­ly, in­creas­ing par­tic­i­pa­tion in higher ed­u­ca­tion be­came the mantra of British gov­ern­ments. The pro­por­tion of school leavers reach­ing higher ed­u­ca­tion be­gan to slowly in­crease dur­ing the early Eight­ies, un­til it sud­denly in­creased rapidly to­wards the end of the decade. As il­lus­trated in Fig­ure 1, the pro­por­tion of a co­hort par­tic­i­pat­ing in higher ed­u­ca­tion dou­bled over a five year pe­ri­od, from 15% in 1988 to 30% by 1992…we analyse the early labour mar­ket ex­pe­ri­ence of the 1995 co­hort, since these peo­ple grad­u­ated at the peak of the higher ed­u­ca­tion ex­pan­sion pe­ri­od. We find a re­duc­tion in the pro­por­tion of matched grad­u­ates, com­pared to the 1990 co­hort. This sug­gests that the labour mar­ket could not fully ac­com­mo­date the in­creased in­flow of new grad­u­ates, al­though this did not lead to an in­creased wage penalty as­so­ci­ated with over-e­d­u­ca­tion. Hence, the post-ex­pan­sion co­hort had the ap­pro­pri­ate skills to suc­ceed in the labour mar­ket. Sec­ond­ly, we are the first to in­ves­ti­gate whether the over-e­d­u­ca­tion wage penalty re­mains even after con­trol­ling for ob­serv­able grad­u­ate skills, skill mis­match, as well as un­ob­serv­able char­ac­ter­is­tics. We find some ev­i­dence that gen­uinely over-e­d­u­cated in­di­vid­u­als lack ‘grad­u­ate skills’; mostly man­age­ment and lead­er­ship skills. Ad­di­tion­al­ly, the lon­gi­tu­di­nal el­e­ment of the dataset is used to cre­ate a mea­sure of time-in­vari­ant labour mar­ket un­ob­serv­able char­ac­ter­is­tics which are also found to be an im­por­tant de­ter­mi­nant of the prob­a­bil­ity to be over-e­d­u­cat­ed. Over-e­d­u­ca­tion im­pacts neg­a­tively on the wages of grad­u­ates, over and above skill lev­els (ob­served or not) which sug­gests that the penalty can­not be solely ex­plained by a lack of skills but also re­flects some job idio­syn­cratic char­ac­ter­is­tics. It also in­creases un­em­ploy­ment by up to three months but does not lead to an in­crease in job search, as the num­bers of job held since grad­u­a­tion is not affected by the cur­rent over-e­d­u­ca­tion sta­tus.

    …Most of the UK lit­er­a­ture has re­lied on self­-assess­ment of over-e­d­u­ca­tion, and typ­i­cally finds that 30% of grad­u­ates are overe­d­u­cated4. Battu et al. (2000) pro­vide one of the most com­pre­hen­sive stud­ies of over-e­d­u­ca­tion. The av­er­age pro­por­tion of over-e­d­u­cated in­di­vid­u­als across the 36 es­ti­mates of their analy­sis was around one-quar­ter, with es­ti­mates rang­ing be­tween one-four­teenth and as high as two-thirds. For the UK, Battu et al (2000) con­cluded that over-e­d­u­ca­tion has not in­creased in the early Nineties.

    This re­sult is sup­ported by Groot and Maassen van den Brink (2000) whose meta-analy­sis of 25 stud­ies found no ten­dency for a world-wide in­crease in the in­ci­dence of over-e­d­u­ca­tion de­spite the gen­eral im­prove­ment in the level of ed­u­ca­tion, al­though they do sug­gest it has be­come in­creas­ingly con­cen­trated among lower abil­ity work­ers, sug­gest­ing the over-e­d­u­ca­tion is not solely due to mis­match of work­ers and jobs. Free­man’s pi­o­neer­ing work on over-e­d­u­ca­tion (1976) sug­gests that over-e­d­u­ca­tion is a tem­po­rary phe­nom­e­non due to fric­tion in the labour mar­ket, al­though UK ev­i­dence is con­trary to this as­sump­tion. Dolton and Vi­g­noles (2000) found that 38% of 1980 UK grad­u­ates were over-e­d­u­cated in their first job and that 30% re­mained in that state six years lat­er. Over a longer pe­riod there is also ev­i­dence that over-e­d­u­ca­tion is a per­ma­nent fea­ture of some grad­u­ates’ ca­reer (Dolton and Silles, 2003). For grad­u­ates the wage penalty as­so­ci­ated with over-e­d­u­ca­tion ranges be­tween 11 and 30%, how­ev­er, con­trary to Free­man’s view over-e­d­u­ca­tion has not led to a de­crease in the UK re­turn to ed­u­ca­tion in gen­eral (Mach­in, 1999 and Dear­den et al., 2002) even if re­cent ev­i­dence by Walker and Zhu (2005) re­port lower re­turns for the most re­cent co­hort of grad­u­ates.

    The gen­eral con­sen­sus is that after con­trol­ling for differ­ences in so­cio-e­co­nomic and in­sti­tu­tional fac­tors, over-e­d­u­ca­tion is a con­se­quence of un­ob­serv­able el­e­ments such as het­ero­ge­neous abil­ity and skills. There is ev­i­dence to sup­port this from stud­ies by Büchel and Poll­man­n-Schult (2001), Bauer (2002), Cheva­lier (2003) and Frenette (2004). Most over-e­d­u­cated work­ers are effi­ciently matched into ap­pro­pri­ate jobs and after ac­count­ing for the un­ob­served het­ero­gene­ity, the wage penalty for over-e­d­u­ca­tion is re­duced. How­ev­er, a re­main­ing group of work­ers ap­pear over-skilled for their jobs and suffer from sub­stan­tial wage penal­ties.

  37. The Bu­reau of La­bor Sta­tis­tics re­vised its data-col­lec­tion method for un­em­ploy­ment du­ra­tion in Jan­u­ary 2011. Based on the pre­vi­ous method, the av­er­age un­em­ploy­ment du­ra­tion would be about 37 weeks rather than 40 weeks. For more in­for­ma­tion, see “Changes to data col­lected on un­em­ploy­ment du­ra­tion”.↩︎

  38. “A Decade of Slack La­bor Mar­kets”, Scott Win­ship, Fel­low; other good quotes:

    From 1951 through 2007, there were never more than three un­em­ployed work­ers for each job open­ing, and it was rare for that fig­ure even to hit two-to-one. In con­trast, there have been more than three job­seek­ers per open­ing in every sin­gle month since Sep­tem­ber 2008. The ra­tio peaked some­where be­tween five-to-one and sev­en-to-one in mid-2009. It has since de­clined but we have far to go be­fore we re­turn to “nor­mal” lev­els.

    The bleak out­look for job­seek­ers has three im­me­di­ate sources. The sharp de­te­ri­o­ra­tion be­gin­ning in early 2007 is the most dra­matic fea­ture of the above chart (the rise in job scarcity after point C in the chart, the steep­ness of which de­pends on the data source used). But two less ob­vi­ous fac­tors pre­dated the re­ces­sion. The first is the steep­ness of the rise in job scarcity dur­ing the pre­vi­ous re­ces­sion in 2001 (from point A to point B), which ri­valed that dur­ing the deep down­turn of the early 1980s. The sec­ond is the fail­ure be­tween 2003 and 2007 of jobs per job­seeker to re­cover from the 2001 re­ces­sion (the fail­ure of point C to fall back to point A).