The Costs of Reliability

Link post

A ques­tion that used to puz­zle me is “Why can peo­ple be so much bet­ter at do­ing a thing for fun, or to help their friends and fam­ily, than they are at do­ing the ex­act same thing as a job?”

I’ve seen it in my­self and I’ve seen it in oth­ers. Peo­ple can be hugely more pro­duc­tive, cre­ative, in­tel­li­gent, and effi­cient on just-for-fun stuff than they are at work.

Maybe it’s some­thing around co­er­cion? But it hap­pens to peo­ple even when they choose their work and have no di­rect su­per­vi­sor, as when a pro­lific hob­by­ist writer sud­denly gets writer’s block as soon as he goes pro.

I think it has a very mun­dane ex­pla­na­tion; it’s always more ex­pen­sive to have to meet a spe­cific com­mit­ment than merely to do some­thing valuable.

If I feel like writ­ing some­times and not other times, then if writ­ing is my hobby I’ll write when I feel like it, and my out­put per hour of writ­ing will be fairly high. Even within “writ­ing”, if my in­ter­ests vary, and I write about what­ever I feel like, I can take full ad­van­tage of ev­ery writ­ing hour. By con­trast, if I’ve com­mit­ted to write a spe­cific piece by a spe­cific dead­line, I have to do it whether or not I’m in the mood for it, and that means I’ll prob­a­bly be less effi­cient, spend more time dither­ing, and I’ll de­mand more ex­ter­nal com­pen­sa­tion in ex­change for my in­con­ve­nience.

The stuff I write for fun may be valuable! And if you sim­ply di­vide the value I pro­duce by my hours of la­bor or the amount I need to be paid, I’m hugely more effi­cient in my free time than in my paid time! But I can’t just triv­ially “up my effi­ciency” in my paid time; re­li­a­bil­ity it­self has a cost.

The costs of re­li­a­bil­ity are of­ten in­visi­ble, but they can be very im­por­tant. The cost (in time and in office sup­plies and soft­ware tools) of track­ing and doc­u­ment­ing your work so that you can de­liver it on time. The cost (in la­bor and equip­ment) of qual­ity as­surance test­ing. The op­por­tu­nity cost of cre­at­ing sim­pler and less am­bi­tious things so that you can de­liver them on time and free of defects.

Reli­a­bil­ity be­comes more im­por­tant with scale. Large or­ga­ni­za­tions have more rules and pro­ce­dures than small ones, and this is ra­tio­nal. Ac­cord­ingly, they pay more costs in re­li­a­bil­ity.

One rea­son is that the at­tack sur­face for er­rors grows with the num­ber of in­di­vi­d­u­als in­volved. For in­stance, large or­ga­ni­za­tions of­ten have rules against down­load­ing soft­ware onto com­pany com­put­ers with­out per­mis­sion. The chance that any one per­son down­loads mal­i­cious soft­ware that se­ri­ously harms the com­pany is small, but the chance that at least one per­son does rises with the num­ber of em­ploy­ees.

Another rea­son is that co­or­di­na­tion be­comes more im­por­tant with more peo­ple. If a pro­ject de­pends on many peo­ple co­op­er­at­ing, then you as an in­di­vi­d­ual aren’t sim­ply try­ing to do the best thing, but rather the best thing that is also un­der­stand­able and pre­dictable and ca­pa­ble of achiev­ing buy-in from oth­ers.

Fi­nally, large in­sti­tu­tions are more tempt­ing to at­tack­ers than small ones, since they have more value to cap­ture. For in­stance, large com­pa­nies are more likely to be tar­geted by law­suits or pub­lic out­cry than pri­vate in­di­vi­d­u­als, so it’s strate­gi­cally cor­rect for them to spend more on defen­sive mea­sures like le­gal com­pli­ance pro­ce­dures or pro­fes­sional PR.

All of these types of defen­sive or pre­ven­ta­tive ac­tivity re­duce effi­ciency — you can do less in a given timeframe with a given bud­get. Large in­sti­tu­tions, even when do­ing ev­ery­thing right, ac­quire in­effi­cien­cies they didn’t have when small, be­cause they have higher re­li­a­bil­ity re­quire­ments.

Of course, there are also economies of scale that in­crease effi­ciency. There are fixed ex­penses that only large in­sti­tu­tions can af­ford, that make marginal pro­duc­tion cheaper. There are ways to ag­gre­gate many risky com­po­nents so that the whole is more ro­bust than any one part, e.g. in dis­tributed com­pu­ta­tion, com­pressed sens­ing, or sim­ply av­er­ag­ing. Op­ti­mal firm size is a bal­ance.

This frame­work tells us when we ought to find it pos­si­ble to get bet­ter-than-de­fault effi­ciency eas­ily, i.e. with­out any clever tricks, just by ac­cept­ing differ­ent trade­offs than oth­ers do. For ex­am­ple:

1.) Peo­ple given an open-ended man­date to do what they like can be far more effi­cient than peo­ple work­ing to spec…at the cost of un­pre­dictable out­put with no guaran­tees of get­ting what you need when you need it. (See: aca­demic re­search.)

2.) Things that come with fewer guaran­tees of re­li­able perfor­mance can be cheaper in the av­er­age use case…at the cost of com­pletely let­ting you down when they oc­ca­sion­ally fail. (See: pro­to­type or beta-ver­sion tech­nol­ogy.)

3.) Ac­tivi­ties within highly co­op­er­a­tive so­cial en­vi­ron­ments can be more effi­cient…at the cost of not scal­ing to more ad­ver­sar­ial en­vi­ron­ments where you have to spend more re­sources on defend­ing against at­tacks. (See: Eter­nal Septem­ber)

4.) Hav­ing an “op­por­tunis­tic” policy of tak­ing what­ever op­por­tu­ni­ties come along (for in­stance, hang­ing out in a pub­lic place and chat­ting with whomever comes along and seems in­ter­est­ing, vs. schedul­ing ap­point­ments) al­lows you to make use of time that oth­ers have to spend do­ing “noth­ing” … at the cost of never be­ing able to com­mit to ac­tivi­ties that need time blocked out in ad­vance.

5.) Shar­ing high-fixed-cost items (like cars) can be more effi­cient than own­ing…at the cost of not hav­ing a guaran­tee that they’ll always be available when you need them.

In gen­eral, you can get greater effi­ciency for things you don’t ab­solutely need than for things you do; if some­thing is merely nice-to-have, you can han­dle it if it oc­ca­sion­ally fails, and your av­er­age cost-benefit ra­tio can be very good in­deed. But this doesn’t mean you can eas­ily copy the effi­ciency of lux­u­ries in the pro­duc­tion of ne­ces­si­ties.

(This sug­gests that “toys” are a good place to look for in­no­va­tion. Frivolous, op­tional goods are where we should ex­pect it to be most af­ford­able to ex­per­i­ment, all else be­ing equal; and we should ex­pect tech­nolo­gies that first suc­ceed in “toy” do­mains to ex­pand to “im­por­tant, life-and-death” do­mains later.)