Brain structure and the halo effect


When peo­ple on LW want to ex­plain a bias, they of­ten turn to Evolu­tion­ary psy­chol­ogy. For ex­am­ple, Luke­prog writes

Hu­man rea­son­ing is sub­ject to a long list of bi­ases. Why did we evolve such faulty think­ing pro­cesses? Aren’t false be­liefs bad for sur­vival and re­pro­duc­tion?

I think that ″evolved faulty think­ing pro­cesses″ is the wrong way to look at it and I will ar­gue that some bi­ases are the con­se­quence of struc­tural prop­er­ties of the brain, which ‘can­not’ be af­fected by evolu­tion.

Brain struc­ture and the halo effect

I want to in­tro­duce a sim­ple model, which re­lates the halo effect to a struc­tural prop­erty of the brain. My hope is that this ap­proach will be use­ful to un­der­stand the halo effect more sys­tem­at­i­cally and shows that think­ing in evolu­tion­ary terms is not always the best way to think about cer­tain bi­ases.

One cru­cial prop­erty of the brain is that it has to map a (es­sen­tially in­finite) high-di­men­sional re­al­ity onto a finite low-di­men­sional in­ter­nal rep­re­sen­ta­tion. (If you know some Lin­ear Alge­bra, you can think of this as a pro­jec­tion from a high-di­men­sional space into a low-di­men­sional space.) This is done more or less au­to­mat­i­cally by the limi­ta­tion of our senses and brain’s struc­ture as a neu­ral net­work.

Neural network (Wikipedia)

An im­me­di­ate con­se­quence of this ob­ser­va­tion is that there will be many states of the world, which are mapped to an al­most iden­ti­cal in­ner rep­re­sen­ta­tion. In terms of com­pu­ta­tional effi­ciency it makes sense to use over­lap­ping set of neu­rons with similar ac­ti­va­tion level to rep­re­sent similar con­cepts. (This is also a con­se­quence of how the brain ac­tu­ally builds rep­re­sen­ta­tions from sense in­puts.)

Now com­pare this to the fol­low­ing pas­sage from here.

The halo effect is that per­cep­tions of all pos­i­tive traits are cor­re­lated. Pro­files rated higher on scales of at­trac­tive­ness, are also rated higher on scales of tal­ent, kind­ness, hon­esty, and in­tel­li­gence.

This shouldn’t be a sur­prise, since ‘pos­i­tive’ (‘feels good’) seems to be one of the evolu­tion­ary hard-wired con­cepts. Other con­cepts that we ac­quire dur­ing our life and as­so­ci­ate with pos­i­tive emo­tions, like kind­ness and hon­esty are mapped to ‘nearby’ neu­ral struc­tures. When one of those men­tal struc­tures is ac­ti­vated, the ‘closed ones’ will be ac­ti­vated to a cer­tain de­gree as well.

Since we differ­en­ti­ate con­cepts more when we are learn­ing about a sub­ject, the above rea­son­ing should im­ply that chil­dren and peo­ple with less ed­u­ca­tion in a cer­tain area should be more in­fluenced by this (gen­er­al­ized) halo effect in that area.


Since evolu­tion can only mod­ify the ex­ist­ing brain struc­ture but can­not get away from the neu­ral net­work ‘de­sign’, the halo effect is a nec­es­sary by-product of hu­man think­ing. But the de­gree of ‘throw­ing things in one pot’ will de­pend on how much we learn about those things and in­crease our rep­re­sen­ta­tion di­men­sion­al­ity.

My hope is that we can re­lief evolu­tion from the bur­den of hav­ing to ex­plain so many things and fo­cus more on struc­tural ex­pla­na­tions, which provide a work­ing model for pos­si­ble ap­pli­ca­tions and a bet­ter un­der­stand­ing.

PS: I am always grate­ful for feed­back!