Debiasing as Non-Self-Destruction

Nick Bostrom asks:

One sign that sci­ence is not all bo­gus is that it en­ables us to do things, like go the moon. What prac­ti­cal things does de­bi­ass­ing en­able us to do, other than re­frain­ing from buy­ing lot­tery tick­ets?

It seems to me that how to be smart varies widely be­tween pro­fes­sions. A hedge-fund trader, a re­search biol­o­gist, and a cor­po­rate CEO must learn differ­ent skill sets in or­der to be ac­tively ex­cel­lent—an ap­pren­tice­ship in one would not serve for the other.

Yet such con­cepts as “be will­ing to ad­mit you lost”, or “policy de­bates should not ap­pear one-sided”, or “plan to over­come your flaws in­stead of just con­fess­ing them”, seem like they could ap­ply to many pro­fes­sions. And all this ad­vice is not so much about how to be ex­traor­di­nar­ily clever, as, rather, how to not be stupid. Each pro­fes­sion has its own way to be clever, but their ways of not be­ing stupid have much more in com­mon. And while vic­tors may pre­fer to at­tribute vic­tory to their own virtue, my small knowl­edge of his­tory sug­gests that far more bat­tles have been lost by stu­pidity than won by ge­nius.

De­bi­as­ing is mostly not about how to be ex­traor­di­nar­ily clever, but about how to not be stupid. Its great suc­cesses are dis­asters that do not ma­te­ri­al­ize, defeats that never hap­pen, mis­takes that no one sees be­cause they are not made. Often you can’t even be sure that some­thing would have gone wrong if you had not tried to de­bias your­self. You don’t always see the bul­let that doesn’t hit you.

The great vic­to­ries of de­bi­as­ing are ex­actly the lot­tery tick­ets we didn’t buy—the hopes and dreams we kept in the real world, in­stead of di­vert­ing them into in­finites­i­mal prob­a­bil­ities. The triumphs of de­bi­as­ing are cults not joined; op­ti­mistic as­sump­tions re­jected dur­ing plan­ning; time not wasted on blind alleys. It is the art of non-self-de­struc­tion.

Ad­mit­tedly, none of this is spec­tac­u­lar enough to make the evening news. It’s not a moon land­ing—though the moon land­ing did surely re­quire thou­sands of things to not go wrong.

So how can we know that our de­bi­as­ing efforts are gen­uinely use­ful? Well, this is the worst sort of anec­do­tal ev­i­dence—but peo­ple do some­times ig­nore my ad­vice, and then, some­times, catas­tro­phe en­sues of just the sort I told them to ex­pect. That is a very weak kind of con­fir­ma­tion, and I would like to see con­trol­led stud­ies… but most of the stud­ies I’ve read con­sist of tak­ing a few un­der­grad­u­ates who are in it for the course credit, merely tel­ling them about the bias, and then wait­ing to see if they im­prove. What we need is lon­gi­tu­di­nal stud­ies of life out­comes, and I can think of few peo­ple I would name as can­di­dates for the ex­per­i­men­tal group.

The fact is, most peo­ple who take a half­hearted pot­shot at de­bi­as­ing them­selves do not get huge amounts of mileage out of it. This is one of those things you have to work at for quite a while be­fore you get good at it, es­pe­cially since there’s cur­rently no source of sys­tem­atic train­ing, or even a de­cent man­ual. If for many years you prac­tice the tech­niques and sub­mit your­self to strict con­straints, it may be that you will glimpse the cen­ter. But un­til then, mis­takes avoided are of­ten just re­placed by other mis­takes. It takes time for your mind to be­come sig­nifi­cantly quieter. In­deed, a lit­tle knowl­edge of cog­ni­tive bias of­ten does more harm than good.

As for pub­lic proof, I can see at least three ways that it could come about. First, there might be founded an Order of Bayescraft for peo­ple who are se­ri­ous about it, and the grad­u­ates of these do­jos might prove sys­tem­at­i­cally more suc­cess­ful even af­ter con­trol­ling for mea­sures of fluid in­tel­li­gence. Se­cond, you could wait for some in­di­vi­d­ual or group, work­ing on an im­por­tant do­main-spe­cific prob­lem but also known for their com­mit­ment to de­bi­as­ing, to pro­duce a spec­tac­u­larly huge pub­lic suc­cess. Third, there might be found tech­niques that can be taught eas­ily and that have read­ily mea­sure­able re­sults; and then sim­ple con­trol­led ex­per­i­ments could serve as pub­lic proof, at least for peo­ple who at­tend to Science.