Clickbait might not be destroying our general Intelligence

Epistemic sta­tus This post is a “plau­si­ble con­jec­ture” al­ter­na­tive to Eliezer’s take on click­bait. My ar­ro­gant epis­ti­mol­ogy thinks it’s sub­stan­tially closer to the truth than Eliezer’s ver­sion.


In the 1950′s there were a similar num­ber (or­der of mag­ni­tude) of hu­mans, and they did a similar amount of so­cial­iz­ing as to­day. This would sug­gest that a similar quan­tity of memetic op­ti­miza­tion was go­ing on then. Any uni­ver­sally pop­u­lar meme would have risen to pop­u­lar­ity. The differ­ence be­tween then and to­day was that then peo­ple usu­ally only com­mu­ni­cated with peo­ple ge­o­graph­i­cally lo­cal to them. With a small num­ber of news­pa­pers be­ing widely read. Whereas to­day, peo­ple com­mu­ni­cate with like minded in­di­vi­d­u­als around the world reg­u­larly.

Sup­pose that differ­ent hu­mans have differ­ent se­lec­tion crite­ria when de­cid­ing to share a meme. In 1950, a meme had to ap­peal to a broad sec­tion of so­ciety to be spread. Not 100%, more like 25%, the small num­ber of differ­ent news­pa­pers, and small num­bers of lo­cal peo­ple to com­mu­ni­cate with still let memes spe­cial­ize along a few so­cioe­co­nomic lines. Eg Catholic memes vs Protes­tant memes in North­ern Ire­land, or Liberal vs Con­ser­va­tive memes in many places. Each news­pa­per had a side, and many of your neigh­bors were on your side.

Nowa­days, memes can spe­cial­ize to fo­cus onto tiny sub­sets of the pop­u­la­tion. Given that a tiny frac­tion of the pop­u­la­tion think in a par­tic­u­lar and un­usual way, they can gather to­gether on­line, and share memes op­ti­mized ex­clu­sively to them. This pro­duces many in­ter­net sub­cul­tures. Within each sub­cul­ture, the se­lec­tion pres­sure isn’t that strong, there aren’t that many model railway buffs or code golfers or … But the memes are op­ti­mized to a par­tic­u­lar way of think­ing, not the com­bi­na­tion of sev­eral.

In some cir­cum­stances, Schel­ling points cre­ate an av­er­ag­ing effect in the 1950 case. Sup­pose an is­sue, say cats vs dogs, is suffi­ciently minor that news­pa­pers are not split into a pro cat pa­per, and a pro dog pa­per. Then if a news­pa­per says any­thing ex­ces­sively pro cat, the dog fans will call them out on it, and pos­si­bly stop read­ing the pa­per. Like­wise if the pa­per is pro dog. So the pa­per finds a Schel­ling point, where nei­ther side are up­set enough to cause a real fuss. Note that this pro­cess is only truth seek­ing to the ex­tent that cat sup­port­ers will let valid pro dog ar­gu­ments slide and call out in­valid ones.

The main effect of filter bub­bles is in­creas­ing the var­i­ance in the meme pool. The biol­o­gists can stop ar­gu­ing with cre­ation­ists, and get down to sort­ing out the de­tails of kin se­lec­tion or what­ever. The cre­ation­ists can stop hav­ing to pedal cre­ation­ism to the un­con­vinced and can get to­gether to work out the differ­ence be­tween micro-evolu­tion and macro-evolu­tion.

Op­ti­miz­ing for a com­bi­na­tion of A and B can pro­duce more of both than choos­ing which to op­ti­mize for at ran­dom. If we have memes that are (A=10, B=0) and (A=9, B=9) and (A=0, B=10), then the mid­dle one is likely to spread in a world with­out filter bub­bles, but the ex­tremes could spread in bub­bles op­ti­miz­ing only A and B re­spec­tively. If A=san­ity, then the av­er­age san­ity could fall due to the op­ti­miza­tion for san­ity be­ing fo­cused into one place, and diminish­ing marginal re­turns on san­ity for op­ti­miza­tion.

Other effects on the qual­ity of dis­course could in­clude stupid peo­ple hav­ing an eas­ier time voic­ing their opinions. Eliezer says that the qual­ity of in­ter­net dis­cus­sion has de­graded from 2002 to 2017. (I wasn’t old enough to use the in­ter­net in 2002, so can’t con­firm or deny this.) Ac­cord­ing to these sources, un­der 10% of the world was on­line in 2002, face­book and twit­ter hadn’t started yet. In short, get­ting on the in­ter­net re­quired more tech­ni­cal com­pe­tence, the hard­ware was more ex­pen­sive, and there was less to do there. The typ­i­cal in­ter­net user was mod­er­ately well ed­u­cated and smart. The typ­i­cal news­pa­per jour­nal­ist was also mod­er­ately well ed­u­cated. Any re­duc­tion in qual­ity would seem to be from un­in­formed peo­ple be­ing fi­nally able to tell the world why the earth is flat.

If you think that the world needs a few highly sane peo­ple, not many slightly sane peo­ple, then an ag­gre­ga­tion into a few groups of san­ity is benefi­cial.

Un­der the hy­poth­e­sis that click­bait is de­stroy­ing in­tel­li­gence, the ex­is­tance of less wrong, and places like it, is highly sur­pris­ing, un­der a seg­re­ga­tion hy­poth­e­sis, its ex­pected that the most ra­tio­nal peo­ple clump to­gether.