Might humans not be the most intelligent animals?

The idea that hu­mans are the most in­tel­li­gent an­i­mals on Earth ap­pears patently ob­vi­ous to a lot of peo­ple. And to a large ex­tent, I agree with this in­tu­ition. Hu­mans clearly dom­i­nate the world in tech­nolog­i­cal in­no­va­tion, con­trol, com­mu­ni­ca­tion, and co­or­di­na­tion.

How­ever, more re­cently I have been ac­quainted with some ev­i­dence that the propo­si­tion is not ac­tu­ally true, or at the very least is non-ob­vi­ous. The co­nun­drum arises when we dis­t­in­guish raw in­no­va­tive ca­pa­bil­ity, from the abil­ity to effi­ciently pro­cess cul­ture. I’ll ex­plain the ba­sic case here.

Robin Han­son has some­times pointed to the ac­cu­mu­la­tion of cul­ture as the rele­vant sep­a­ra­tion be­tween hu­mans and other an­i­mals. Un­der this view, the rea­son why hu­mans are able to dom­i­nate the world with tech­nol­ogy has less to due with our raw in­no­va­tion abil­ities, and more to do with the fact that we can effi­ciently pro­cess ac­cu­mu­lated cul­tural in­for­ma­tion and build on it.

If the rea­son for our tech­nolog­i­cal dom­i­nance is due to raw in­no­va­tive abil­ity, we might ex­pect a more dis­con­tin­u­ous jump in ca­pa­bil­ities for AI, since there was a sharp change be­tween “un­able to in­no­vate” and “able to in­no­vate” dur­ing evolu­tion­ary his­tory, which might have been due to some key ar­chi­tec­tural tweak. We might there­fore ex­pect that our AIs will ex­pe­rience the same jump af­ter re­ceiv­ing the same tweak.

If the rea­son for our tech­nolog­i­cal dom­i­nance is due to our abil­ity to pro­cess cul­ture, how­ever, then the case for a dis­con­tin­u­ous jump in ca­pa­bil­ities is weaker. This is be­cause our AI sys­tems can already pro­cess cul­ture some­what effi­ciently right now (see GPT-2) and there doesn’t seem like a hard sep­a­ra­tion be­tween “be­ing able to pro­cess cul­ture in­effi­ciently” and “able to pro­cess cul­ture effi­ciently” other than the ini­tial jump from not be­ing able to do it at all, which we have already passed. There­fore, our cur­rent sys­tems are cur­rently bot­tle­necked on some met­ric which is more con­tin­u­ous.

The ev­i­dence for the cul­tural ac­cu­mu­la­tion view comes from a few lines of in­quiry, such as

  • Feral chil­dren lack an in­no­vat­ing cul­ture to learn from, and are cor­re­spond­ingly ‘un­in­tel­li­gent’ from our point of view. This is de­spite the fact that feral chil­dren hold more-or-less the ex­act same raw in­no­va­tive ca­pa­bil­ities as nor­mal hu­mans. For Ge­nie, one feral child, “Doc­tors found it ex­tremely difficult to test or es­ti­mate Ge­nie’s men­tal age or any of her cog­ni­tive abil­ities, but on two at­tempts they found Ge­nie scored at the level of a 13-month-old.”

  • Books like The Se­cret of Our Suc­cess doc­u­ment re­mark­able in­stances of raw in­no­va­tive ca­pa­bil­ity be­ing much less use­ful than abil­ity to use cul­ture. See the SlateS­tarCodex re­view here.

  • Evolu­tion­ar­ily, in­creases in raw in­no­va­tive ca­pa­bil­ity may uni­ver­sally aid re­pro­duc­tive ca­pa­bil­ity for any an­i­mal, whereas in­creases in abil­ity to effi­ciently pro­cess cul­ture would only help in a species that de­vel­oped cul­ture to be­gin with. It’s no suprise there­fore, that af­ter hu­mans de­vel­oped lan­guage, we quickly de­vel­oped the ca­pa­bil­ity to pro­cess cul­tural in­for­ma­tion, whereas other an­i­mals can’t learn much from our civ­i­liza­tion. How­ever, this doesn’t nec­es­sar­ily mean they are less in­no­va­tive in gen­eral.

Un­der the view that our dom­i­nance is due to cul­tural ac­cu­mu­la­tion, we would ex­pect that there are some an­i­mals that are more in­tel­li­gent than hu­mans in the sense of raw in­no­va­tive abil­ity. The rea­son is that it would be sur­pris­ing a pri­ori for us to be the most in­tel­li­gent, un­less we had rea­son to be­lieve so an­throp­i­cally. How­ever, if cul­ture is the rea­son why we dom­i­nate, then the an­thropic ar­gu­ment here is weaker.

We do see some ini­tial signs that hu­mans might not be the most in­tel­li­gent species on Earth in the in­no­va­tive sense. For in­stance,

  • Although hu­mans have the high­est en­cephal­iza­tion quo­tient, we don’t have the most neu­rons, or even the most neu­rons in our fore­brain.

  • Some an­i­mals have eas­ily mea­surable cog­ni­tive ca­pa­bil­ities which sur­pass ours. One ex­am­ple is the chim­panzee, which may have bet­ter work­ing mem­ory. Edit: Now I’m not sure whether this fact is cor­rect. See com­ments be­low for more dis­cus­sion.

If hu­mans are the most in­tel­li­gent in the sense of hav­ing the best raw in­no­va­tive abil­ities, this hy­poth­e­sis should be testable by ad­minis­ter­ing a bat­tery of cog­ni­tive tests to an­i­mals. How­ever, this is made difficult due to a num­ber of fac­tors.

First, most in­tel­li­gence tests that hu­mans take rely on their abil­ity to pro­cess lan­guage, dis­qual­ify­ing other an­i­mals from the start.

Se­cond, hu­mans might be bi­ased to­wards ad­minis­ter­ing tests about things that we are good at, since we might not even be aware of the type of cog­ni­tive abil­ities we score poorly on. This may have the effect of prov­ing that hu­mans are su­pe­rior in in­tel­li­gence, but only on the limited sub­set of tests that we used.

Third, if we use cur­rent hu­man in­tel­li­gence tests (or any­thing like them), the fol­low­ing ar­gu­ment arises. Com­put­ers can already out­perform hu­mans at some tasks that in­tel­li­gence tests mea­sure, such as mem­ory, but this doesn’t mean that com­put­ers already have a bet­ter abil­ity in­no­vate. We would need to test some­thing that we felt con­fi­dent ac­cu­rately in­di­cated in­no­va­tive abil­ity.

Since I’m mostly in­ter­ested in this ques­tion be­cause of its im­pli­ca­tions for AI take­off speeds, I’d want to know what type of things are most use­ful for de­vel­op­ing tech­nol­ogy, and see if we can see the same abil­ities in an­i­mals sans cul­tural ac­cu­mu­la­tion mod­ules.