From Personal to Prison Gangs: Enforcing Prosocial Behavior

This post origi­nally ap­peared here; I’ve up­dated it slightly and posted it here as a fol­low-up to this post.

David Fried­man has a fas­ci­nat­ing book on al­ter­na­tive le­gal sys­tems. One chap­ter fo­cuses on prison law—not the nom­i­nal rules, but the rules en­forced by pris­on­ers them­selves.

The un­offi­cial le­gal sys­tem of Cal­ifor­nia pris­on­ers is par­tic­u­larly in­ter­est­ing be­cause it un­der­went a phase change in the 1960’s.

Prior to the 1960’s, pris­on­ers ran on a de­cen­tral­ized code of con­duct—var­i­ous un­writ­ten rules roughly amount­ing to “mind your own busi­ness and don’t cheat any­one”. Pri­son­ers who kept to the code were af­forded some re­spect by their fel­low in­mates. Pri­son­ers who vi­o­lated the code were os­tra­cized, mak­ing them fair game for the more preda­tory in­mates. There was no for­mal en­force­ment; the code was es­sen­tially a rep­u­ta­tion sys­tem.

In the 1960’s, that changed. Dur­ing the code era, Cal­ifor­nia’s to­tal prison pop­u­la­tion was only about 5000, with about 1000 in­mates in a typ­i­cal prison. That’s quite a bit more than Dun­bar’s num­ber, but still low enough for a rep­u­ta­tion sys­tem to work through sec­ond-or­der con­nec­tions. By 1970, Cal­ifor­nia’s prison pop­u­la­tion had bal­looned past 25000; to­day it is over 170000. The num­ber of pris­ons also grew, but not nearly as quickly as the pop­u­la­tion, and to­day’s pris­on­ers fre­quently move across pris­ons any­way. In short, a de­cen­tral­ized rep­u­ta­tion sys­tem be­came un­ten­able. There were too many other in­mates to keep track of.

As the rep­u­ta­tion sys­tem col­lapsed, a new le­gal in­sti­tu­tion grew to fill the void: prison gangs. Un­der the gang sys­tem, each in­mate is ex­pected to af­fili­ate with a gang (though most are not for­mal gang mem­bers). The gang will ex­plain the rules, of­ten in writ­ten form, and en­force them on their own af­fili­ates. When con­flict arises be­tween af­fili­ates of differ­ent gangs, the gang lead­ers ne­go­ti­ate set­tle­ment, with gang lead­ers en­forc­ing pun­ish­ments on their own af­fili­ates. (Gang lead­ers are strongly mo­ti­vated to avoid gang-level con­flicts.) Rather than need­ing to track rep­u­ta­tion of ev­ery­one in­di­vi­d­u­ally, in­mates need only pay at­ten­tion to gangs at a group level.
Of course, in­mates need some way to tell who is af­fili­ated with each gang—thus the rise of racial seg­re­ga­tion in prison. Dur­ing the code era, pris­on­ers tended to as­so­ci­ate by race and cul­ture, but there was no overt racial hos­tility and no hard rules against as­so­ci­at­ing across race. But to­day’s prison gangs are highly racially seg­re­gated, mak­ing it easy to rec­og­nize the gang af­fili­a­tion of in­di­vi­d­ual in­mates. They claim ter­ri­tory in pris­ons—show­ers or ball courts—and en­force their claims, re­sult­ing in hard racial seg­re­ga­tion.

The change from a small, low-con­nec­tion prison pop­u­la­tion to a large, high-con­nec­tion pop­u­la­tion was the root cause. That change drove a tran­si­tion from a de­cen­tral­ized, rep­u­ta­tion-based sys­tem to prison gangs. This, in turn, in­volved two fur­ther tran­si­tions. First, a tran­si­tion from de­cen­tral­ized, in­for­mal un­writ­ten rules to for­mal writ­ten rules with cen­tral­ized en­force­ment. Se­cond, a tran­si­tion from in­di­vi­d­ual to group-level iden­tity, in this case man­i­fest­ing as racial seg­re­ga­tion.


This is hardly unique to pris­ons. The pat­tern is uni­ver­sal among hu­man in­sti­tu­tions. In small groups, ev­ery­body knows ev­ery­body. Rules are in­for­mal, iden­tity is in­di­vi­d­ual. But as groups grow:

  • Rules be­come for­mal, writ­ten, and cen­trally enforced

  • Iden­tity be­comes group-based.

Con­sider com­pa­nies. I work at a ten-per­son com­pany. Every­one in the office knows ev­ery­one else by name, and ev­ery­one has some idea of what ev­ery­one else is work­ing on. We have nom­i­nal job ti­tles, but ev­ery­body works on what­ever needs do­ing. Our perfor­mance re­view pro­cess is to oc­ca­sion­ally raise the topic in weekly one-on-one meet­ings.

Go to a thou­sand or ten thou­sand per­son com­pany, and job ti­tles play a much stronger role in who does what. Peo­ple don’t know ev­ery­one, so they iden­tify oth­ers by de­part­ment or role. They un­der­stand what a de­vel­oper or a man­ager does, rather than un­der­stand­ing what John or Allan does. Iden­tity be­comes group-based. At the same time, hi­er­ar­chy and bu­reau­cracy are for­mal­ized.

The key pa­ram­e­ter here is num­ber of in­ter­ac­tions be­tween each pair of peo­ple (you should click that link, it’s re­ally cool). In small groups, each pair of peo­ple has many in­ter­ac­tions, so peo­ple get to know each other. In large groups, there are many one-off in­ter­ac­tions be­tween strangers. Without past in­ter­ac­tions to fall back on, peo­ple need other ways to figure out how to in­ter­act with each other. One solu­tion is for­mal rules, which give guidance on in­ter­ac­tions with any­one. Another solu­tion is group-based iden­tity—if I know how to in­ter­act with lawyers at work in gen­eral, then I don’t need to know each in­di­vi­d­ual lawyer.

In this re­gard, pris­ons and com­pa­nies are just micro­cosms of so­ciety in gen­eral.


At some point over the past cou­ple hun­dred years, so­ciety un­der­went a tran­si­tion similar to that of the Cal­ifor­nia prison sys­tem.

In 1800, peo­ple were mostly farm­ers, liv­ing in small towns. The lo­cal pop­u­la­tion was within an or­der of mag­ni­tude of Dun­bar’s num­ber, and gen­er­ally small enough to rely on rep­u­ta­tion for day-to-day deal­ings.

To­day, that is not the case [cita­tion needed].

Just as in pris­ons and com­pa­nies, we should ex­pect this change to drive two kinds of tran­si­tions:

  • A tran­si­tion from in­for­mal, de­cen­tral­ized rules to for­mal, writ­ten, cen­trally-en­forced rules.

  • A tran­si­tion from in­di­vi­d­ual to group-level iden­tity.

This can ex­plain an awful lot of the ways in which so­ciety has changed over the past cou­ple hun­dred years, as well as how spe­cific so­cial in­sti­tu­tions evolve over time.
To take just a few ex­am­ples…

  • Reg­u­la­tion. As peo­ple have more one-off in­ter­ac­tions, rep­u­ta­tion be­comes less ten­able, and we should ex­pect for­mal reg­u­la­tion to grow. Con­versely, reg­u­la­tions are rou­tinely ig­nored among peo­ple who know each other.

  • Liti­ga­tion. Again, with more one-off in­ter­ac­tions, we should ex­pect peo­ple to rely more on for­mal liti­ga­tion and less on in­for­mal set­tle­ment. Con­versely, peo­ple who in­ter­act fre­quently rarely sue each other—and when they do, it’s ex­pected to mess up the re­la­tion­ship.

  • Pro­fes­sional li­cens­ing. Without rep­u­ta­tion, peo­ple need some way to sig­nal that they are safe to hire. We should ex­pect li­cens­ing to in­crease as pair­wise in­ter­ac­tions de­crease.

  • Cre­den­tial­ism. This is just a gen­er­al­iza­tion of li­cens­ing. As rep­u­ta­tion fails, we should ex­pect peo­ple to rely more heav­ily on for­mal cre­den­tials—“you are your de­gree” and so forth.

  • Stereo­typ­ing. Without past in­ter­ac­tions with a par­tic­u­lar per­son, we should ex­pect peo­ple to gen­er­al­ize based on su­perfi­cially “similar” peo­ple. This could be any­thing from the usual culprits (race, eth­nic­ity, age) to job roles (ac­tu­ar­ies, lawyers) to con­sump­tion sig­nals (iphone, con­verse, fancy suit).

  • Trib­al­ism. From na­tion­al­ism to sports fans to iden­tity poli­tics, an in­creas­ing prevalence of group-level iden­tity means an in­creas­ing prevalence of tribal be­hav­ior. In par­tic­u­lar, I’d ex­pect that so­cial me­dia out­lets with more one-off or low-count in­ter­ac­tions are char­ac­ter­ized by more ex­treme trib­al­ism.

  • Stan­dards for im­per­sonal in­ter­ac­tions. “Pro­fes­sion­al­ism” at work is a good ex­am­ple.

I’ve fo­cused mostly on nega­tive ex­am­ples here, but it’s not all bad—even some of these ex­am­ples have up­sides. When Cal­ifor­nia’s pris­ons moved from an in­for­mal code to prison gangs, the homi­cide rate dropped like a rock; the gangs hate prison lock­downs, so they go to great lengths to pre­vent homi­cides. Of course, gangs have lots of down­sides too. The point which gen­er­al­izes is this: bod­ies with cen­tral­ized power have their own in­cen­tives, and out­comes will be “good” to ex­actly the ex­tent that the in­cen­tives of the cen­tral­ized power al­ign with ev­ery­body else’ in­cen­tives and de­sires.

Con­sider cre­den­tial­ism, for ex­am­ple. It’s not all bad—to the ex­tent that we now hire based on de­gree rather than nepo­tism, it’s prob­a­bly a step up. But on the other hand, col­leges them­selves have less than ideal in­cen­tives. Even set­ting aside col­leges’ in­cen­tives, the whole cre­den­tial sys­tem shoe­horns peo­ple into one-size-fits-all solu­tions; a brilli­ant patent clerk would have a much more difficult time mak­ing a name in physics to­day than a hun­dred years ago.


Of course, all of these ex­am­ples share one crit­i­cal pos­i­tive fea­ture: they scale. That’s the whole rea­son things changed in the first place—we needed sys­tems which could scale up be­yond per­sonal re­la­tion­ships and rep­u­ta­tion.

This brings us to the take­away: what should you do if you want to change these things? Per­haps you want a so­ciety with less cre­den­tial­ism, reg­u­la­tion, stereo­typ­ing, trib­al­ism, etc. Maybe you like some of these things but not oth­ers. Re­gard­less, surely there’s some­thing some­where on that list you’re less than happy about.
The first take­away is that these are not pri­mar­ily poli­ti­cal is­sues. The changes were driven by tech­nol­ogy and eco­nomics, which cre­ated a broader so­cial graph with fewer re­peated in­ter­ac­tions. Poli­ti­cal ac­tion is un­likely to re­verse any of these changes; the equil­ibrium has shifted, and any policy change would be fight­ing grav­ity. Even if em­ploy­ers were out­lawed from mak­ing hiring de­ci­sions based on col­lege de­gree, they’d find some work-around which amounted to the same thing. Even if the en­tire fed­eral reg­ister dis­ap­peared overnight, de-facto in­dus­try reg­u­la­tory bod­ies would pop up. And so forth.

So if we want to e.g. re­duce reg­u­la­tion, we should first fo­cus on the un­der­ly­ing so­cioe­co­nomic prob­lem: fewer in­ter­ac­tions. A world of Ama­zon and Wal­mart, where ev­ery con­sumer faces de­ci­sions be­tween a mil­lion differ­ent prod­ucts, is in­evitably a world where con­sumers do not know pro­duc­ers very well. There’s just too many prod­ucts and com­pa­nies to keep track of the rep­u­ta­tion of each. To re­duce reg­u­la­tion, first fo­cus on solv­ing that prob­lem, scal­ably. Think ama­zon re­views—it’s an im­perfect sys­tem, but it’s far more flex­ible and effi­cient than for­mal reg­u­la­tion, and it scales.

Now for the real prob­lem: on­line re­views are liter­ally the only ex­am­ple I could come up with where tech­nol­ogy offers a way to scale-up rep­u­ta­tion-based sys­tems, and maybe some­day roll back cen­tral­ized con­trol struc­tures or group iden­tities. How can we solve these sorts of prob­lems more gen­er­ally? Please com­ment if you have ideas.