Thou Art Godshatter

Be­fore the 20th cen­tury, not a sin­gle hu­man be­ing had an ex­plicit con­cept of “in­clu­sive ge­netic fit­ness”, the sole and ab­solute ob­ses­sion of the blind idiot god. We have no in­stinc­tive re­vul­sion of con­doms or oral sex. Our brains, those supreme re­pro­duc­tive or­gans, don’t perform a check for re­pro­duc­tive effi­cacy be­fore grant­ing us sex­ual plea­sure.

Why not? Why aren’t we con­sciously ob­sessed with in­clu­sive ge­netic fit­ness? Why did the Evolu­tion-of-Hu­mans Fairy cre­ate brains that would in­vent con­doms? “It would have been so easy,” thinks the hu­man, who can de­sign new com­plex sys­tems in an af­ter­noon.

The Evolu­tion Fairy, as we all know, is ob­sessed with in­clu­sive ge­netic fit­ness. When she de­cides which genes to pro­mote to uni­ver­sal­ity, she doesn’t seem to take into ac­count any­thing ex­cept the num­ber of copies a gene pro­duces. (How strange!)

But since the maker of in­tel­li­gence is thus ob­sessed, why not cre­ate in­tel­li­gent agents—you can’t call them hu­mans—who would like­wise care purely about in­clu­sive ge­netic fit­ness? Such agents would have sex only as a means of re­pro­duc­tion, and wouldn’t bother with sex that in­volved birth con­trol. They could eat food out of an ex­plic­itly rea­soned be­lief that food was nec­es­sary to re­pro­duce, not be­cause they liked the taste, and so they wouldn’t eat candy if it be­came detri­men­tal to sur­vival or re­pro­duc­tion. Post-menopausal women would babysit grand­chil­dren un­til they be­came sick enough to be a net drain on re­sources, and would then com­mit suicide.

It seems like such an ob­vi­ous de­sign im­prove­ment—from the Evolu­tion Fairy’s per­spec­tive.

Now it’s clear, as was dis­cussed yes­ter­day, that it’s hard to build a pow­er­ful enough con­se­quen­tial­ist. Nat­u­ral se­lec­tion sort-of rea­sons con­se­quen­tially, but only by de­pend­ing on the ac­tual con­se­quences. Hu­man evolu­tion­ary the­o­rists have to do re­ally high-falutin’ ab­stract rea­son­ing in or­der to imag­ine the links be­tween adap­ta­tions and re­pro­duc­tive suc­cess.

But hu­man brains clearly can imag­ine these links in pro­tein. So when the Evolu­tion Fairy made hu­mans, why did It bother with any mo­ti­va­tion ex­cept in­clu­sive ge­netic fit­ness?

It’s been less than two cen­turies since a pro­tein brain first rep­re­sented the con­cept of nat­u­ral se­lec­tion. The mod­ern no­tion of “in­clu­sive ge­netic fit­ness” is even more sub­tle, a highly ab­stract con­cept. What mat­ters is not the num­ber of shared genes. Chim­panzees share 95% of your genes. What mat­ters is shared ge­netic var­i­ance, within a re­pro­duc­ing pop­u­la­tion—your sister is one-half re­lated to you, be­cause any vari­a­tions in your genome, within the hu­man species, are 50% likely to be shared by your sister.

Only in the last cen­tury—ar­guably only in the last fifty years—have evolu­tion­ary biol­o­gists re­ally be­gun to un­der­stand the full range of causes of re­pro­duc­tive suc­cess, things like re­cip­ro­cal al­tru­ism and costly sig­nal­ing. Without all this highly de­tailed knowl­edge, an in­tel­li­gent agent that set out to “max­i­mize in­clu­sive fit­ness” would fall flat on its face.

So why not pre­pro­gram pro­tein brains with the knowl­edge? Why wasn’t a con­cept of “in­clu­sive ge­netic fit­ness” pro­grammed into us, along with a library of ex­plicit strate­gies? Then you could dis­pense with all the re­in­forcers. The or­ganism would be born know­ing that, with high prob­a­bil­ity, fatty foods would lead to fit­ness. If the or­ganism later learned that this was no longer the case, it would stop eat­ing fatty foods. You could re­fac­tor the whole sys­tem. And it wouldn’t in­vent con­doms or cook­ies.

This looks like it should be quite pos­si­ble in prin­ci­ple. I oc­ca­sion­ally run into peo­ple who don’t quite un­der­stand con­se­quen­tial­ism, who say, “But if the or­ganism doesn’t have a sep­a­rate drive to eat, it will starve, and so fail to re­pro­duce.” So long as the or­ganism knows this very fact, and has a util­ity func­tion that val­ues re­pro­duc­tion, it will au­to­mat­i­cally eat. In fact, this is ex­actly the con­se­quen­tial­ist rea­son­ing that nat­u­ral se­lec­tion it­self used to build au­to­matic eaters.

What about cu­ri­os­ity? Wouldn’t a con­se­quen­tial­ist only be cu­ri­ous when it saw some spe­cific rea­son to be cu­ri­ous? And wouldn’t this cause it to miss out on lots of im­por­tant knowl­edge that came with no spe­cific rea­son for in­ves­ti­ga­tion at­tached? Again, a con­se­quen­tial­ist will in­ves­ti­gate given only the knowl­edge of this very same fact. If you con­sider the cu­ri­os­ity drive of a hu­man—which is not undis­crim­i­nat­ing, but re­sponds to par­tic­u­lar fea­tures of prob­lems—then this com­plex adap­ta­tion is purely the re­sult of con­se­quen­tial­ist rea­son­ing by DNA, an im­plicit rep­re­sen­ta­tion of knowl­edge: Ances­tors who en­gaged in this kind of in­quiry left more de­scen­dants.

So in prin­ci­ple, the pure re­pro­duc­tive con­se­quen­tial­ist is pos­si­ble. In prin­ci­ple, all the an­ces­tral his­tory im­plic­itly rep­re­sented in cog­ni­tive adap­ta­tions can be con­verted to ex­plic­itly rep­re­sented knowl­edge, run­ning on a core con­se­quen­tial­ist.

But the blind idiot god isn’t that smart. Evolu­tion is not a hu­man pro­gram­mer who can si­mul­ta­neously re­fac­tor whole code ar­chi­tec­tures. Evolu­tion is not a hu­man pro­gram­mer who can sit down and type out in­struc­tions at sixty words per minute.

For mil­lions of years be­fore ho­minid con­se­quen­tial­ism, there was re­in­force­ment learn­ing. The re­ward sig­nals were events that cor­re­lated re­li­ably to re­pro­duc­tion. You can’t ask a non­ho­minid brain to fore­see that a child eat­ing fatty foods now will live through the win­ter. So the DNA builds a pro­tein brain that gen­er­ates a re­ward sig­nal for eat­ing fatty food. Then it’s up to the or­ganism to learn which prey an­i­mals are tastiest.

DNA con­structs pro­tein brains with re­ward sig­nals that have a long-dis­tance cor­re­la­tion to re­pro­duc­tive fit­ness, but a short-dis­tance cor­re­la­tion to or­ganism be­hav­ior. You don’t have to figure out that eat­ing sug­ary food in the fall will lead to di­gest­ing calories that can be stored fat to help you sur­vive the win­ter so that you mate in spring to pro­duce offspring in sum­mer. An ap­ple sim­ply tastes good, and your brain just has to plot out how to get more ap­ples off the tree.

And so or­ganisms evolve re­wards for eat­ing, and build­ing nests, and scar­ing off com­peti­tors, and helping siblings, and dis­cov­er­ing im­por­tant truths, and form­ing strong al­li­ances, and ar­gu­ing per­sua­sively, and of course hav­ing sex...

When ho­minid brains ca­pa­ble of cross-do­main con­se­quen­tial rea­son­ing be­gan to show up, they rea­soned con­se­quen­tially about how to get the ex­ist­ing re­in­forcers. It was a rel­a­tively sim­ple hack, vastly sim­pler than re­build­ing an “in­clu­sive fit­ness max­i­mizer” from scratch. The pro­tein brains plot­ted how to ac­quire calories and sex, with­out any ex­plicit cog­ni­tive rep­re­sen­ta­tion of “in­clu­sive fit­ness”.

A hu­man en­g­ineer would have said, “Whoa, I’ve just in­vented a con­se­quen­tial­ist! Now I can take all my pre­vi­ous hard-won knowl­edge about which be­hav­iors im­prove fit­ness, and de­clare it ex­plic­itly! I can con­vert all this com­pli­cated re­in­force­ment learn­ing ma­chin­ery into a sim­ple declar­a­tive knowl­edge state­ment that ‘fatty foods and sex usu­ally im­prove your in­clu­sive fit­ness’. Con­se­quen­tial rea­son­ing will au­to­mat­i­cally take care of the rest. Plus, it won’t have the ob­vi­ous failure mode where it in­vents con­doms!”

But then a hu­man en­g­ineer wouldn’t have built the retina back­ward, ei­ther.

The blind idiot god is not a uni­tary pur­pose, but a many-splin­tered at­ten­tion. Foxes evolve to catch rab­bits, rab­bits evolve to evade foxes; there are as many evolu­tions as species. But within each species, the blind idiot god is purely ob­sessed with in­clu­sive ge­netic fit­ness. No qual­ity is val­ued, not even sur­vival, ex­cept in­so­far as it in­creases re­pro­duc­tive fit­ness. There’s no point in an or­ganism with steel skin if it ends up hav­ing 1% less re­pro­duc­tive ca­pac­ity.

Yet when the blind idiot god cre­ated pro­tein com­put­ers, its mono­ma­ni­a­cal fo­cus on in­clu­sive ge­netic fit­ness was not faith­fully trans­mit­ted. Its op­ti­miza­tion crite­rion did not suc­cess­fully quine. We, the hand­i­work of evolu­tion, are as alien to evolu­tion as our Maker is alien to us. One pure util­ity func­tion splin­tered into a thou­sand shards of de­sire.

Why? Above all, be­cause evolu­tion is stupid in an ab­solute sense. But also be­cause the first pro­tein com­put­ers weren’t any­where near as gen­eral as the blind idiot god, and could only uti­lize short-term de­sires.

In the fi­nal anal­y­sis, ask­ing why evolu­tion didn’t build hu­mans to max­i­mize in­clu­sive ge­netic fit­ness, is like ask­ing why evolu­tion didn’t hand hu­mans a ri­bo­some and tell them to de­sign their own bio­chem­istry. Be­cause evolu­tion can’t re­fac­tor code that fast, that’s why. But maybe in a billion years of con­tinued nat­u­ral se­lec­tion that’s ex­actly what would hap­pen, if in­tel­li­gence were fool­ish enough to al­low the idiot god con­tinued reign.

The Mote in God’s Eye by Niven and Pour­nelle de­picts an in­tel­li­gent species that stayed biolog­i­cal a lit­tle too long, slowly be­com­ing truly en­slaved by evolu­tion, grad­u­ally turn­ing into true fit­ness max­i­miz­ers ob­sessed with out­re­pro­duc­ing each other. But thank­fully that’s not what hap­pened. Not here on Earth. At least not yet.

So hu­mans love the taste of sugar and fat, and we love our sons and daugh­ters. We seek so­cial sta­tus, and sex. We sing and dance and play. We learn for the love of learn­ing.

A thou­sand deli­cious tastes, matched to an­cient re­in­forcers that once cor­re­lated with re­pro­duc­tive fit­ness—now sought whether or not they en­hance re­pro­duc­tion. Sex with birth con­trol, choco­late, the mu­sic of long-dead Bach on a CD.

And when we fi­nally learn about evolu­tion, we think to our­selves: “Ob­sess all day about in­clu­sive ge­netic fit­ness? Where’s the fun in that?

The blind idiot god’s sin­gle mono­ma­ni­a­cal goal splin­tered into a thou­sand shards of de­sire. And this is well, I think, though I’m a hu­man who says so. Or else what would we do with the fu­ture? What would we do with the billion galax­ies in the night sky? Fill them with max­i­mally effi­cient repli­ca­tors? Should our de­scen­dants de­liber­ately ob­sess about max­i­miz­ing their in­clu­sive ge­netic fit­ness, re­gard­ing all else only as a means to that end?

Be­ing a thou­sand shards of de­sire isn’t always fun, but at least it’s not bor­ing. Some­where along the line, we evolved tastes for nov­elty, com­plex­ity, el­e­gance, and challenge—tastes that judge the blind idiot god’s mono­ma­ni­a­cal fo­cus, and find it aes­thet­i­cally un­satis­fy­ing.

And yes, we got those very same tastes from the blind idiot’s god­shat­ter. So what?