Neuroscience basics for LessWrongians

The ori­gins of this ar­ti­cle are in my par­tial tran­script of the live June 2011 de­bate be­tween Robin Han­son and Eliezer Yud­kowsky. While I still feel like I don’t en­tirely un­der­stand his ar­gu­ments, a few of his com­ments about neu­ro­science made me strongly go, “no, that’s not right.”

Fur­ther­more, I’ve no­ticed that while LessWrong in gen­eral seems to be very strong on the psy­cholog­i­cal or “black box” side of cog­ni­tive sci­ence, there isn’t as much dis­cus­sion of neu­ro­science here. This is some­what un­der­stand­able. Our cur­rent un­der­stand­ing of neu­ro­science is frus­trat­ingly in­com­plete, and too much jour­nal­ism on neu­ro­science is sen­sa­tion­al­is­tic non­sense. How­ever, I think what we do know is worth know­ing. (And part of what makes much neu­ro­science jour­nal­ism an­noy­ing is that it makes a big deal out of things that are to­tally un­sur­pris­ing, given what we already know.)

My qual­ifi­ca­tions to do this: while my de­grees are in philos­o­phy, for awhile in un­der­grad I was a neu­ro­science ma­jor, and ended up tak­ing quite a bit of neu­ro­science as a re­sult. This means I can as­sure you that most of what I say here is stan­dard neu­ro­science which could be found in an in­tro­duc­tory text­book like Ni­chols, Martin, Wal­lace, & Fuchs’ From Neu­ron to Brain (one of the text books I used as an un­der­grad­u­ate). The only things that might not be to­tally stan­dard are the con­jec­ture I make about how com­plex cur­rently-poorly-un­der­stood ar­eas of the brain are likely to be, and also some of the points I make in crit­i­cism of Eliezer at the end (though I be­lieve these are not a very big jump from cur­rent text­book neu­ro­science.)

One of the main themes of this ar­ti­cle will be spe­cial­iza­tion within the brain. In par­tic­u­lar, we know that the brain is di­vided into spe­cial­ized ar­eas at the macro level, and we un­der­stand some (though not very much) of the micro-level wiring that sup­ports this spe­cial­iza­tion. It seems likely that each re­gion of the brain has its own micro-level wiring to sup­port its spe­cial­ized func­tion, and in some re­gions the wiring is likely to be quite com­plex.

1. Spe­cial­iza­tion of brain regions

One of the best-es­tab­lished facts about the brain is that spe­cific re­gions han­dle spe­cific func­tions. And it isn’t just that in each in­di­vi­d­ual, spe­cific brain re­gions han­dle spe­cific func­tions. It’s also that which re­gions han­dle which func­tions is con­sis­tent across in­di­vi­d­u­als. This is an ex­tremely well-es­tab­lished find­ing, but it’s worth briefly sum­ma­riz­ing some of the ev­i­dence for it.

One kind of ev­i­dence comes from ex­per­i­ments in­volv­ing di­rect elec­tri­cal stim­u­la­tion of the brain. This can­not eth­i­cally be done on hu­mans with­out a sound med­i­cal rea­son, but it is used with epilep­tic pa­tients in or­der to de­ter­mine the source of the prob­lem, which is nec­es­sary in or­der to treat epilepsy sur­gi­cally.

In epilep­tic pa­tients, stim­u­lat­ing cer­tain re­gions of the brain (known as the pri­mary sen­sory ar­eas) causes the pa­tient to re­port sen­sa­tions: sights, sounds, feel­ings, smells, and tastes. Which sen­sa­tions are caused by stim­u­lat­ing which re­gions of the brain is con­sis­tent across pa­tients. This is the source of the “Pen­field ho­muncu­lus,” a map of brain re­gions which, when stim­u­lated, re­sult in touch sen­sa­tions which pa­tients de­scribe as feel­ing like they come from par­tic­u­lar parts of the body. Stim­u­lat­ing one re­gion, for ex­am­ple, might con­sis­tently lead to a pa­tient re­port­ing a feel­ing in his left foot.

Re­gions of the brain as­so­ci­ated with sen­sa­tions are known as sen­sory ar­eas or sen­sory cor­tex. Other re­gions of the brain, when stim­u­lated, lead to in­vol­un­tary mus­cle move­ments. Those ar­eas are known as mo­tor ar­eas or mo­tor cor­tex, and again, which ar­eas cor­re­spond to which mus­cles is con­sis­tent across pa­tients. The con­sis­tency of the map­ping of brain re­gions across pa­tients is im­por­tant, be­cause it’s ev­i­dence of an in­nate struc­ture to the brain.

An even more sig­nifi­cant kind of ev­i­dence comes from stud­ies of pa­tients with brain dam­age. Brain dam­age can pro­duce very spe­cific abil­ity losses, and pa­tients with dam­age to the same ar­eas will typ­i­cally have similar abil­ity losses. For ex­am­ple, the rear most part of the hu­man cere­bral cor­tex is the pri­mary vi­sual cor­tex, and dam­age to it re­sults in a phe­nomenon known as cor­ti­cal blind­ness. That is to say, the pa­tient is blind in spite of hav­ing perfectly good eyes. Their other men­tal abil­ities may be un­af­fected.

That much is not sur­pris­ing, given what we know from stud­ies in­volv­ing elec­tri­cal stim­u­la­tion, but abil­ity losses from brain dam­age can be strangely spe­cific. For ex­am­ple, neu­ro­scien­tists now be­lieve that one func­tion of the tem­po­ral lobe is to rec­og­nize ob­jects and faces. A key line of ev­i­dence for this is that pa­tient with dam­age to cer­tain parts of the tem­po­ral lobe will be un­able to iden­tify those things by sight, even though they may be able to de­scribe the ob­jects in great de­tail. Here is neu­rol­o­gist Oliver Sacks’ de­scrip­tion of an in­ter­ac­tion with one such pa­tient, the titu­lar pa­tient in Sacks’ book The Man Who Mis­took His Wife For a Hat:

‘What is this?’ I asked, hold­ing up a glove.

‘May I ex­am­ine it?’ he asked, and, tak­ing it from me, he pro­ceeded to ex­am­ine it as he had ex­am­ined the ge­o­met­ri­cal shapes.

‘A con­tin­u­ous sur­face,’ he an­nounced at last, ‘in­folded on it­self. It ap­pears to have’—he hes­i­tated—’five out­pouch­ings, if this is the word.’

‘Yes,’ I said cau­tiously. You have given me a de­scrip­tion. Now tell me what it is.’

‘A con­tainer of some sort?’

‘Yes,’ I said, ‘and what would it con­tain?’

‘It would con­tain its con­tents!’ said Dr P., with a laugh. ‘There are many pos­si­bil­ities. It could be a change purse, for ex­am­ple, for coins of five sizes. It could …’

I in­ter­rupted the barmy flow. ‘Does it not look fa­mil­iar? Do you think it might con­tain, might fit, a part of your body?’

No light of recog­ni­tion dawned on his face. (Later, by ac­ci­dent, he got it on, and ex­claimed, ‘My God, it’s a glove!’)

The fact that dam­age to cer­tain parts of the tem­po­ral lobe re­sults in an in­abil­ity to rec­og­nize ob­jects con­tains an ex­tremely im­por­tant les­son. For most of us, rec­og­niz­ing ob­jects re­quires no effort or thought as long as we can see the ob­ject clearly. Be­cause it’s easy for us, it might be tempt­ing to think it’s an in­her­ently easy task, one that shouldn’t re­quire hardly any brain mat­ter to perform. Cer­tainly it never oc­curred to me be­fore I stud­ied neu­ro­science that ob­ject recog­ni­tion might re­quire a spe­cial brain re­gion. But it turns out that tasks that seem easy to us can in fact re­quire such a spe­cial­ized re­gion.

Another ex­am­ple of this fact comes from two brain re­gions in­volved in lan­guage, Broca’s area and Wer­nicke’s area. Da­m­age to each area leads to dis­tinct types of difficul­ties with lan­guage, known as Broca’s apha­sia and Wer­nicke’s apha­sia, re­spec­tively. Both are strange con­di­tions, and my de­scrip­tion of them may not give a full sense of what they are like. Read­ers might con­sider search­ing on­line for videos of in­ter­views Broca’s apha­sia and Wer­nicke’s apha­sia pa­tients to get a bet­ter idea of what the con­di­tions en­tail.

Broca’s apha­sia is a loss of abil­ity to pro­duce lan­guage. In one ex­treme case, one of the origi­nal cases stud­ied by Paul Broca, a pa­tient was only able to say the word “tan.” Other pa­tients may have a less limited vo­cab­u­lary, but still strug­gle to come up with words for what they want to say. And even when they can come up with in­di­vi­d­ual words, they may be un­able to put them into sen­tences. How­ever, pa­tients with Broca’s apha­sia ap­pear to have no difficulty un­der­stand­ing speech, and show aware­ness of their dis­abil­ity.

Wer­nicke’s apha­sia is even stranger. It is of­ten de­scribed as an in­abil­ity to un­der­stand lan­guage while still be­ing able to pro­duce lan­guage. How­ever, while pa­tients with Wer­nicke’s apha­sia may have lit­tle difficulty pro­duc­ing com­plete, gram­mat­i­cally cor­rect sen­tences, the sen­tences tend to be non­sen­si­cal. And Wer­nicke’s pa­tients of­ten act as if they are com­pletely un­aware of their con­di­tion. A former pro­fes­sor of mine once de­scribed a Wer­nicke’s pa­tient as sound­ing “like a poli­ti­cian,” and from watch­ing a video of an in­ter­view with the pa­tient, I agreed: I was im­pressed by his abil­ity to con­fi­dently ut­ter non­sense.

The fact of these two forms of apha­sia sug­gest that Broca’s area and Wer­nicke’s area have two very im­por­tant and dis­tinct roles in our abil­ity to pro­duce and un­der­stand lan­guage. And I find this fact strange to write about. Like ob­ject recog­ni­tion, lan­guage comes nat­u­rally to us. As I write this, my in­tu­itive feel­ing is that the work I am do­ing comes mainly in the ideas, plus mak­ing a few sub­tle stylis­tic de­ci­sions. I know from neu­ro­science that I would be un­able to write this if I had sig­nifi­cant dam­age to ei­ther re­gion. Yet I am to­tally un­con­scious of the work they are do­ing for me.

2. Com­plex, spe­cial­ized wiring within regions

“Wiring” is a hard metaphor to avoid when talk­ing about the brain, but it is also a po­ten­tially mis­lead­ing one. Peo­ple of­ten talk about “elec­tri­cal sig­nals” in the brain, but un­like elec­tri­cal sig­nals in hu­man tech­nol­ogy, which in­volves move­ment of elec­trons be­tween the atom­ics of the con­duc­tor, sig­nals in the hu­man brain in­volve move­ment of ions and small molecules across cell mem­branes and be­tween cells.

Fur­ther­more, the first thing most peo­ple who know a lit­tle bit about neu­ro­science will think of when they hear the word “wiring” is ax­ons and den­drites, the long skinny pro­jec­tions along which sig­nals are trans­mit­ted from neu­ron to neu­ron. But it isn’t just the lay­out of ax­ons and den­drites that mat­ters. Ion chan­nels, and the struc­tures that trans­port neu­ro­trans­mit­ters across cell mem­branes, are also im­por­tant.

Th­ese can vary a lot at the synapse, the place where two neu­rons touch. For ex­am­ple, synapses vary in strength, that is to say, the strength of the one neu­ron’s effect on the other neu­ron. Sy­napses can also be ex­ci­ta­tory (ac­tivity in one leads in­creased ac­tivity in the other) or in­hibitory (ac­tivity in one leads to de­creased ac­tivity in the other). And this is just a cou­ple of the ways synapses can vary; the de­tails can be some­what com­pli­cated, and I’ll give one ex­am­ple of how the de­tails can be com­pli­cated later.

I say all this just to make clear what I mean when I talk about how the brain’s “wiring.” By “wiring,” I mean all the fea­tures of the phys­i­cal struc­tures that con­nect neu­rons to each other and which are rele­vant for un­der­stand­ing how the brain works. I mean to all the things I’ve men­tioned above, and any­thing I may have omit­ted. It’s im­por­tant to have a word to talk about this wiring, be­cause what (ad­mit­tedly lit­tle) we un­der­stand about how the brain works we un­der­stand in terms of this wiring.

For ex­am­ple, the ner­vous sys­tem ac­tu­ally first be­gins pro­cess­ing vi­sual in­for­ma­tion in the retina (the part of the eye at the back where our light re­cep­tors are). This is done by what’s known as the cen­ter-sur­round sys­tem: a patch of light re­cep­tors, when ac­ti­vated, ex­cites one neu­ron, but nearby patches of light re­cep­tors, when ac­ti­vated, in­hibit that same neu­ron (some­times, the ex­ci­ta­tory roles and in­hibitory roles are re­versed).

The effect of this is that what the neu­rons are sen­si­tive to is not light it­self, but con­trast. They’re con­trast de­tec­tors. And what al­lows them to de­tect con­trast isn’t any­thing mag­i­cal, it’s just the wiring, the way the neu­rons are con­nected to­gether.

This tech­nique of get­ting neu­rons to serve spe­cific func­tions based on how they are wired to­gether shows up in more com­pli­cated ways in the brain it­self. There’s one line of ev­i­dence for spe­cial­iza­tion of brain re­gions that I saved for this sec­tion, be­cause it also tells us about the de­tails of how the brain is wired. That line of ev­i­dence is record­ings from the brain us­ing elec­trodes.

For ex­am­ple, dur­ing the 50’s David Hubel and Torsten Weisel did ex­per­i­ments where they par­a­lyzed the eye mus­cles of an­i­mals, stuck elec­trodes in pri­mary vi­sual ar­eas of the an­i­mals’ brains, and then showed the an­i­mals var­i­ous images to see which images would cause elec­tri­cal sig­nals in the an­i­mals’ pri­mary vi­sual ar­eas. It turned out that the main thing that causes elec­tri­cal sig­nals in the pri­mary vi­sual ar­eas is lines.

In par­tic­u­lar, a given cell in the pri­mary vi­sual area will have a par­tic­u­lar ori­en­ta­tion of line which it re­sponds to. It ap­pears that the way these line-ori­en­ta­tion de­tect­ing cells work is that they re­ceive in­put from sev­eral con­trast de­tect­ing cells which, them­selves, cor­re­spond to re­gions of the retina that are them­selves all in a line. A line in the right po­si­tion and ori­en­ta­tion will ac­ti­vate all of the con­trast-de­tect­ing cells, which in turn ac­ti­vates the line-ori­en­ta­tion de­tect­ing cell. A line in the right po­si­tion but wrong ori­en­ta­tion will ac­ti­vate only one or a few con­trast-de­tect­ing cells, not enough to ac­ti­vate the line-ori­en­ta­tion de­tect­ing cell.

[If this is un­clear, a di­a­gram like the one on Wikipe­dia may be helpful, though Wikipe­dia’s di­a­gram may not be the best.]

Another ex­am­ple of a trick the brain does with neu­ral wiring is lo­cat­ing sounds us­ing what are called “in­ter­au­ral time differ­ences.” The idea is this: there is a group of neu­rons that re­ceives in­put from both ears, and speci­fi­cally re­sponds to si­mul­ta­neous in­put from both ears. How­ever, the ax­ons run­ning from the ears to the neu­rons in this group of cells vary in length, and there­fore they vary in how long it takes them to get a sig­nal from each ear.

This means that which cells in this group re­spond to a sound de­pends on whether or not the sound reaches reaches the ears at the same time or at differ­ent times, and (if at differ­ent times) on how big the time differ­ence is. If there’s no differ­ence, that means the sound came from di­rectly ahead or be­hind (or above or be­low). A big differ­ence, with the sound reach­ing the left ear first, in­di­cates the sound came from the left. A big differ­ence, with the sound reach­ing the right ear first, in­di­cates the sound came from the left. Small differ­ences in­di­cate some­thing in be­tween.

[A di­a­gram might be helpful here too, but I’m not sure where to find a good one on­line.]

I’ve made a point to men­tion these bits of wiring be­cause they’re cases where neu­ro­scien­tists have a clear un­der­stand­ing of how it is that a par­tic­u­lar neu­ron is able to fire only in re­sponse to a par­tic­u­lar kind of stim­u­lus. Un­for­tu­nately, cases like this are rel­a­tively rare. In other cases, how­ever, we at least know that par­tic­u­lar neu­rons re­spond speci­fi­cally to more com­plex stim­uli, even though we don’t know why. In rats, for ex­am­ple, there are cells in the hip­pocam­pus that ac­ti­vate only when the rat is in a par­tic­u­lar lo­ca­tion; ap­par­ently their pur­pose is to keep track of the rat’s lo­ca­tion.

The vi­sual sys­tem gives us some es­pe­cially in­ter­est­ing cases of this sort. We know that the pri­mary vi­sual cor­tex sends in­for­ma­tion to other parts of the brain in two broadly-defined path­ways, the dor­sal path­way and the ven­tral path­way. The dor­sal path­way ap­pears to be re­spon­si­ble for pro­cess­ing in­for­ma­tion re­lated to po­si­tion and move­ment. Some cells in the dor­sal path­way, for ex­am­ple, fire only when an an­i­mal sees an ob­ject mov­ing in a par­tic­u­lar di­rec­tion.

The most in­ter­est­ing cells of this sort that neu­ro­scien­tists have found so far, though, are prob­a­bly some of the cells in the me­dial tem­po­ral lobe, which is part of the ven­tral path­way. In one study (Quiroga et al. 2005), re­searchers took epilep­tic pa­tients who had had elec­trodes im­planted in their brains in or­der to lo­cate the source of their epilepsy and showed them pic­tures of var­i­ous peo­ple, ob­jects, and land­marks. What the re­searchers found is that the neu­ron or small group of neu­rons a given elec­trode was read­ing from typ­i­cally only re­sponded to pic­tures of one per­son or thing.

Fur­ther­more, a par­tic­u­lar elec­trode of­ten got read­ings from very differ­ent pic­tures of a sin­gle per­son or thing, but not similar pic­tures of differ­ent peo­ple or things. In one no­to­ri­ous ex­am­ple, they found a neu­ron that they could only get to re­spond to ei­ther pic­tures of ac­tress Halle Berry or the text “Halle Berry.” This in­cluded draw­ings of the ac­tress, as well as pic­tures of her dressed as Cat­woman (a role which she had re­cently performed at the time the study was performed), but not other draw­ings or other pic­tures of Cat­woman.

What’s go­ing on here? Based on what we know about the wiring of con­trast-de­tec­tors and ori­en­ta­tion-de­tec­tors the fol­low­ing con­jec­ture seems highly likely: if we were to map out the brain com­pletely and then fol­low the path along which vi­sual in­for­ma­tion is trans­mit­ted, we would find that neu­rons grad­u­ally come to be wired to­gether in more and more com­plex ways, to al­low them to grad­u­ally be­come spe­cific to more and more com­plex fea­tures of vi­sual images. This, I think, is an ex­tremely im­por­tant in­fer­ence.

We know that ex­pe­rience can im­pact the way the brain is wired. In fact, some as­pects of the brain’s wiring seem to have evolved speci­fi­cally to be able to change in re­sponse to ex­pe­rience (the main wiring of that sort we know about is called Heb­bian synapses, but the de­tails aren’t im­por­tant here). And it is ac­tu­ally some­what difficult to draw a clear line be­tween fea­tures of the brain that are in­nate and fea­tures of the brain that are the product of learn­ing, be­cause some fairly ba­sic fea­tures of the brain de­pend on out­side cues in or­der to de­velop.

Here, though, I’ll use the word “in­nate” to re­fer to fea­tures of the brain that will de­velop given the over­whelming ma­jor­ity of the con­di­tions an­i­mals of a given species ac­tu­ally de­velop un­der. Un­der that defi­ni­tion, a “Halle Berry neu­ron” is highly un­likely to be in­nate, be­cause there isn’t enough room in the brain to have a neu­ron spe­cific to ev­ery per­son a per­son might pos­si­bly learn about. Such neu­ral wiring is al­most cer­tainly the re­sult of learn­ing.

But im­por­tantly, the un­der­ly­ing struc­ture that makes such learn­ing pos­si­ble is prob­a­bly at least some­what com­pli­cated, and also spe­cial­ized for that par­tic­u­lar kind of learn­ing. This is be­cause such per­son-spe­cific and ob­ject-spe­cific neu­rons are not found in all re­gions of the brain, there must be some­thing spe­cial about the me­dial tem­po­ral lobe that al­lows such learn­ing to hap­pen there.

Similar rea­son­ing ap­plies to re­gions of the brain that we know even less about. For ex­am­ple, it seems likely that Broca’s area and Wer­nicke’s area both con­tain spe­cial­ized wiring for han­dling lan­guage, though we have lit­tle idea how that wiring might perform its func­tion. Given that hu­mans seem to have a con­sid­er­able in­nate knack for learn­ing lan­guage (Pinker 2007), it again seems likely that the wiring is some­what com­pli­cated.

3. On some prob­le­matic com­ments by Eliezer

I agree with Sin­gu­lar­ity In­sti­tute po­si­tions on a great deal. After all, I re­cently made my first dona­tion to the Sin­gu­lar­ity In­sti­tute. But here, I want to point out some prob­le­matic neu­ro­science-re­lated com­ments in Eliezer’s de­bate with Robin Han­son:

If you ac­tu­ally look at the genome, we’ve got about 30,000 genes in here. Most of our 750 megabytes of DNA is repet­i­tive and al­most cer­tainly junk, as best we un­der­stand it. And the brain is sim­ply not a very com­pli­cated ar­ti­fact by com­par­i­son to, say, Win­dows Vista. Now the com­plex­ity that it does have it uses a lot more effec­tively than Win­dows Vista does. It prob­a­bly con­tains a num­ber of de­sign prin­ci­ples which Microsoft knows not. (And I’m not say­ing it’s that small be­cause it’s 750 megabytes, I’m say­ing it’s gotta be that small be­cause at least 90% of the 750 megabytes is junk and there’s only 30,000 genes for the whole body, never mind the brain.)

That some­thing that sim­ple can be this pow­er­ful, and this hard to un­der­stand, is a shock. But if you look at the brain de­sign, it’s got 52 ma­jor ar­eas on each side of the cere­bral cor­tex, dis­t­in­guish­able by the lo­cal pat­tern, the tiles and so on, it just doesn’t re­ally look all that com­pli­cated. It’s very pow­er­ful. It’s very mys­te­ri­ous. What can say about it is that it prob­a­bly in­volves 1,000 differ­ent deep, ma­jor, math­e­mat­i­cal in­sights into the na­ture of in­tel­li­gence that we need to com­pre­hend be­fore we can build it.

Though this is not ex­plicit, there ap­pears to be an in­fer­ence here that, in or­der for some­thing so sim­ple to be so pow­er­ful, it must in­cor­po­rate many deep in­sights into in­tel­li­gence, though we don’t know what most of them are. There are sev­eral prob­lems with this ar­gu­ment.

First of all, it is not true that the fact that that brain is di­vided into only 52 ma­jor ar­eas is ev­i­dence that it is not very com­plex, be­cause know­ing about the com­plex­ity of its macro­scopic or­ga­ni­za­tion tells us noth­ing about the com­plex­ity of its micro­scopic wiring. The brain con­sists of tens of billions of neu­rons, and a sin­gle neu­ron can make hun­dreds of synapses with other neu­rons. The de­tails of how synapses are set up vary greatly. The fact is that un­der a micro­scope, the brain at least looks very com­plex.
The ar­gu­ment from the small size of the genome is more plau­si­ble, es­pe­cially if Eliezer is think­ing in terms of Kol­mogorov com­plex­ity, which is based on the size of the small­est com­puter pro­gram needed to build some­thing. How­ever, it does not fol­low that if the genome is not very com­plex, the brain must not be very com­plex, be­cause the brain may be built not just based on the genome, but also based on in­for­ma­tion from the out­side en­vi­ron­ment. We have good rea­son to think this is how the brain is ac­tu­ally set up, not just in cases we would nor­mally as­so­ci­ate with learn­ing and mem­ory, but with some of the most ba­sic and near-uni­ver­sal fea­tures of the brain. For ex­am­ple, in nor­mal mam­mals, the neu­rons in the vi­sual cor­tex are or­ga­nized into “oc­u­lar dom­i­nance columns,” but these fail to form if the an­i­mal is raised in dark­ness.
More im­por­tantly, there is no rea­son to think get­ting a lot of power out of a rel­a­tively sim­ple de­sign re­quires in­sights into the na­ture of in­tel­li­gence it­self. To use Eliezer’s own ex­am­ple of Win­dows Vista: imag­ine if, for some rea­son, Microsoft de­cided that it was very im­por­tant for the next gen­er­a­tion of its op­er­at­ing sys­tem to be highly com­press­ible. Microsoft tells this to its pro­gram­mers, and they set about look­ing for ways to make an op­er­at­ing sys­tem do most of what the cur­rent ver­sion of Win­dows does while be­ing more com­press­ible. They end up do­ing a lot of things that are only ap­pli­ca­ble to their situ­a­tion, and couldn’t be used to make a much more pow­er­ful op­er­at­ing sys­tem. For ex­am­ple, they might look for ways to re­cy­cle pieces of code, and make par­tic­u­lar pieces of code do as many differ­ent things in the pro­gram as pos­si­ble.
In this case, would we say that they had dis­cov­ered deep in­sights into how to build pow­er­ful op­er­at­ing sys­tems? Well no. And there’s rea­son to think that life on Earth uses similar tricks to get a lot of ap­par­ent com­plex­ity out of rel­a­tively sim­ple ge­netic codes. Genes code for pro­tein. In a phe­nomenon known as “al­ter­na­tive splic­ing,” there may be sev­eral ways to com­bine the parts of a gene, al­low­ing one gene to code for sev­eral pro­teins. And even a sin­gle, spe­cific pro­tein may perform sev­eral roles within an or­ganism. A re­cep­tor pro­tein, for ex­am­ple, may be plugged into differ­ent sig­nal­ing cas­cades in differ­ent parts of an or­ganism.
Eliezer’s com­ments about the com­plex­ity of brain are only a small part of his ar­gu­ments in the de­bate, but I worry that com­ments like these by peo­ple con­cerned with the fu­ture of Ar­tifi­cial In­tel­li­gence are harm­ful in­so­far as they may lead some peo­ple (par­tic­u­larly neu­ro­scien­tists) to con­clude AI-re­lated fu­tur­ism is a bunch of con­fu­sions based in ig­no­rance. I don’t think it is, but a neu­ro­scien­tist tak­ing the Han­son-Yud­kowsky de­bate as an in­tro­duc­tion to the is­sues could eas­ily con­clude that.
Of course, that’s not the most im­por­tant rea­son for peo­ple with an in­ter­est in AI to un­der­stand the ba­sics of neu­ro­science. The most im­por­tant rea­son is that un­der­stand­ing some neu­ro­science will help clar­ify your think­ing about the rest of cog­ni­tive sci­ence.


Pinker, S. 2007. The Lan­guage In­stinct. Harper Peren­nial Modern Clas­sics.

Quiroga, R. Q. et al. 2005. In­var­i­ant vi­sual rep­re­sen­ta­tion by sin­gle neu­rons in the hu­man brain. Na­ture, 435, 1102-1107.