# Causality and its harms

I’ll as­sume that you do not hold fast to a rigor­ous sys­tem of meta­physics, in which case I think you can fancy me and ac­cept that, if I so de­sire to strug­gle, I could re­duce the con­cept of causal­ity to one (or a chain of) prob­a­bil­is­tic re­la­tion­ships be­tween events.

Here’s a naive defi­ni­tion of causal­ity: `P(E) ~= 1 | C` (where “|” stands for “given”). If I can say this, I can most cer­tainly say that `C` causes `E`, at least in a sys­tem where a thing such as “time” ex­ists and where `C` hap­pens be­fore `E`.

It should be noted this doesn’t im­ply `P(C) ~= 1 | E`.

Ob­vi­ously, this defi­ni­tion doesn’t cover all or even most of the things we call causal and can be taken into an ex­treme con­text where it holds in spite of no causal re­la­tion­ship. I’m just start­ing with it be­cause it’s a point from which we can build up to a bet­ter one. Let’s roll with it and look at an ex­am­ple.

## 1. Hy­po­thet­i­cal num­ber one

Hu­man height can be viewed as a func­tion of hap­logroup alle­les + shoe size, that is to say, we can pre­dict h fairly well given those 2 pa­ram­e­ters.

Height can also be viewed as a func­tion of mTORC1 ex­pres­sion + HGH blood lev­els. Let’s say some as­say that tells us the lev­els of mTORC1 and HGH pre­dicts height equally well to hap­logroup alle­les + shoe size.

But I think most sci­en­tists would agree the later are to some ex­tent “causal” for height while the former aren’t. Why?

Well con­sider 2 hy­po­thet­i­cal ex­per­i­ment:

1. Take a 30yo hu­man of av­er­age height and use a sal­ine solu­tion pump to in­flate their feet to 1.5x their size. Then use {mag­i­cal pre­ci­sion nu­cleotide ad­di­tion and dele­tion vec­tor} to re­move all his hap­logroup alle­les and in­sert those most as­so­ci­ated with in­creased height.

2. Take a 30yo hu­man of av­er­age height and then use {mag­i­cal pre­ci­sion nu­cleotide ad­di­tion and dele­tion vec­tor} to over­ex­press the heck out of mTORC1 and HGH.

In which case do we ex­pect to see an in­crease in height? That might in­di­cate causal­ity.

Trick ques­tion, of course, in nei­ther.

We’d ex­pect to see in­creased height in the sec­ond case if the hu­man was, say, 13 in­stead of 30.

The causal­ity here is P(E) given C un­der some spe­cific con­di­tions. Where C has hap­pened be­fore P(E), even if C hap­pened 10 years ago and C hap­pen­ing again now would not af­fect E.

Also, see physics for situ­a­tions where that doesn’t quite cut it ei­ther be­cause the tem­po­ral re­la­tion­ship “seems” in­verted [cita­tion needed].

Causal­ity is what we call `P(E) ~= 1 | C` hap­pened in the past in a world where we have “in­tu­itive” tem­po­ral re­la­tion­ships AND we can be pretty cer­tain about the en­vi­ron­ment.

Hit glass with a ham­mer and it breaks, ex­cept for the fact that this only hap­pens in a fairly spe­cific en­vi­ron­ment with some types of glass and some types of ham­mer. Move the en­vi­ron­ment 300 me­ters un­der­wa­ter, make the glass bul­let­proof, make the ham­mer out of alu­minum or let an in­fant wield it and we lose “causal­ity”, even though the event de­scribed is still the same.

But even that doesn’t cut it in terms of how weird causal­ity is.

## 2. Causal­ity and effect size

Now let’s move onto the idea of “cause” that doesn’t fall into the whole `P(E) ~= 1 | C`. This is not so easy be­cause there are two differ­ent ways by which this could hap­pen.

The easy one in­volves `E` as a con­tin­u­ous value rather than a bi­nary event. In the pre­vi­ous height ex­am­ple, `E` could have been an in­crease as counted by a % of the sub­ject’s ini­tial height or in cen­time­ters.

But this is easy be­cause we can just ab­stract away `E` as some­thing like “and in­crease in height by be­tween 10 and 25%”, ba­si­cally strap some con­fi­dences ranges on a nu­mer­i­cal in­crease and it’s back to be­ing a bi­nary cause.

The hard one in­volves events that hap­pen due to a given cause only very sel­dom.

For ex­am­ple, if we are trav­el­ing through a moun­tain pass af­ter a heavy snow­storm and our friends start yo­del­ing very loudly we might say some­thing like:

Shut up, you might cause an avalanche

But this seems like the ex­act kind of situ­a­tion where we’d be bet­ter suited say­ing:

Shut up, there’s a very spu­ri­ous cor­re­la­tion be­tween loud singing and avalanches and I’d rather not take my chances.

Well, maybe we say “cause” only for the sake of brevity? Doubt­fully.

I think the rea­son we say “cause” here rather than “cor­re­late with” is that we seem to have some un­der­ly­ing in­tu­ition about the laws of the phys­i­cal world, that al­lows us to see the mechanism by which yo­del­ing might put in mo­tion a se­ries of events (us­ing the “naive” strong defi­ni­tion of causal­ity) which end up be­ing “causal” of an avalanche us­ing the naive defi­ni­tion used be­fore (e.g. some to do with very strong echos + very un­sta­ble snow cov­er­ing on a steep slope).

Con­versely, if we saw a black cat jump out of the snow and just re­al­ized to­day is Fri­day 13th we might start be­ing a bit afraid of an avalanche hap­pen­ing, maybe even more so than if our friends start yo­del­ing. But I think even the most su­per­sti­tious per­son would shy away from call­ing black cats in ran­dom places and cer­tain dates “causal” of avalanches.

But then again, if yo­del­ing can cause an avalanche by this defi­ni­tion, so can the but­terfly flap­ping its wings in China the ac­tion of which snow­balled into a slight di­rec­tion of cur­rent in your moun­tain pass which (cou­pled with the yo­del­ing) could cause the avalanche.

Heck, maybe the slops were avalanche se­cure for some ob­tuse rea­son, but then some­one moved some medium-sized rocks a few weeks ago and ac­ci­den­tally re­ally harmed the avalanche-re­lated struc­tural sta­bil­ity.

## 3. Causal­ity or­der, time and replication

Ok, this sec­tion could be ex­panded in an ar­ti­cle on its own, but I don’t find it as in­ter­est­ing as the last, so I will try to keep it brief. To patch up our pre­vi­ous model we need to in­tro­duce the idea of or­der, time, and repli­ca­tion.

Why is the wind causal of an avalanche but not the but­terfly flap­ping its wings?

Well, be­cause of the or­der those events hap­pened in. Given `P(E) = x | C1` and `P(E) = x | C2` but `P(E) = x | C1 & C2` the “cause” of `E` will be whicever of the two causes hap­pened “first”.

Some­times we might change this defi­ni­tion in or­der to bet­ter di­vert our ac­tion. If you push some­one on a sub­way track and he is sub­se­quently un­able to climb back in time and gets hit by a sub­way, you could hardly say to a judge:

Well, your honor, based on prece­dence, I think it’s fair to say that it was his failure to get off the tracks that caused him to be hit. Yes, my push­ing might have caused him to fail at said task… But if we go down that slip­pery slope you could also place blame on my boss, for mak­ing me an­gry this morn­ing and thus caus­ing the pushy be­hav­ior.

Similarly with the yo­del­ing caus­ing the avalanche, rather than the yo­del­ing caus­ing some in­ter­me­di­ary phe­nomenon chain which ends up with one of them caus­ing the avalanche.

We say yo­del­ing causes the avalanche be­cause “yo­del­ing” is an ac­tion­able thing, the re­ver­ber­a­tion of sound through a valley once it leaves the lips, not so much.

A cause is defined based on how easy it is to repli­cate (or, in the case of the track-push­ing, how easy it is to avoid it ever again be repli­cated).

Bar­ring ease of repli­ca­tion, some spa­tiotem­po­ral or­der­ing of the events seems to be preferred.

We usu­ally want the cause to be the “sim­plest” (easy to state and repli­cate) nec­es­sary and suffi­cient con­di­tion to get the effect (Okams Ra­zor + Falsifi­a­bil­ity).

That is to say, cross­ing the US bor­der is what “causes” one to be in the USA.

Tak­ing a plane to NYC also causes one to be in the USA, but it ex­plains fewer ex­am­ples and is a much more com­plex ac­tion. So I think we pre­fer to say that the bor­der cross­ing is the “cause” here.

In­tro­duc­ing spa­tial-tem­po­ral or­der via ap­peal­ing to the sci­en­tific method doesn’t make a lot of sense, but it’s quite an amaz­ing heuris­tic once you think about it.

Two causes seem linked and equally easy to repli­cate, what is a good heuris­tic by which we can get the least amount of ex­per­i­men­tal er­ror if our repli­ca­tion as­sump­tion is wrong?

Well, repli­cat­ing the one that’s clos­est in space and time to the event ob­served (harder if one is close in space and the other in time, but this is so sel­dom the case I’d call it an edge case, and heuris­tic aren’t made for that).

Or, what if we can’t de­cide on how easy they are to repli­cate? Or think they are linked but can’t be sure?

Well, again, the spa­tial-tem­po­ral heuris­tic cleaves through the un­cer­tainty and tells us we are most likely to ob­serve the de­sired effect again (or stop it) by act­ing upon the cause clos­est to it in space and time.

In­ter­est­ing… and get­ting more com­plex.

But at this point causal­ity still sounds kinda rea­son­able.

Granted, we haven’t got­ten into the idea of on­go­ing cause-effect re­la­tion­ships. I’ve kind of as­sumed very com­plex cause-effect re­la­tion­ships can be split into hun­dreds of lit­tle “naive” cau­sa­tions and that some­how hun­dreds of naive cau­sa­tions can add up to a sin­gle big­ger cause.

But those things aside, I think there’s one fi­nal (po­ten­tially most im­por­tant ?) point I want to con­sider:

## 4. (Par­tially real) Hy­po­thet­i­cal num­ber two

As­sume we have 5 camps that ar­gue about what is the causes of hu­man vi­o­lence, from peo­ple at­tack­ing their spouse to sado­masochism, to mass shoot­ings, to drunken fist­fights, to gang wars.

• The blankslateist: Violence is caused by a lack of ed­u­ca­tion and a so­ciety that per­pet­u­ates it. Raise kids in a non-vi­o­lent en­vi­ron­ment, ed­u­cate them about the use­less­ness of vi­o­lence and im­prove their em­pa­thy and we’d ba­si­cally get rid of all the vi­o­lence.

• The economist: Violence is caused by mon­e­tary and so­cial sta­tus mo­ti­vated causes. The rob­ber threat­ens peo­ple with a gun be­cause he wants money. The drunk fist­fighter is mo­ti­vated by some fake ideal of “mas­culine so­cial sta­tus”, in a run­away-er­ro­neous way even the school shooter might be viewed as such. A gang threat­ens to kill busi­ness own­ers for pro­tec­tion taxes and en­gages in war with ri­val gangs as to not lose it’s “cus­tomers”. Provide eco­nomic and so­cial in­cen­tives which always make vi­o­lence an ob­vi­ously sub­op­ti­mal choice and you’d get rid of it.

• The ge­netic de­ter­minis­tic: Violence is caused by ge­netic, it’s baked into hu­man na­ture, it goes down be­cause of evolu­tion­ary cir­cum­stances not fa­vor­ing it and in­creases for the same rea­sons. There’s no cure to vi­o­lence other than figur­ing out how to ge­net­i­cally en­g­ineer non-vi­o­lent hu­mans.

• The Freudian: Violence is caused by our deeper an­i­mal self, our sub­con­scious an­i­mal feel­ing cas­trated by the pact our con­scious self has to sign with mod­ern so­ciety. Violence is just a form of psy­chosis, and out­burst harken­ing back to times im­memo­rial. Treat psy­chosis us­ing ther­apy that al­lows peo­ple to ex­plore and un­der­stand their mind and you get rid of vi­o­lence.

• The Statis­ti­cian: Violence is mainly caused by lead. Get rid of lead pol­lu­tion and you’ve solved most cases of vi­o­lence.

In the first 4 cases, we see an ex­am­ple of what we rec­og­nize as causal­ity.

The blankslateist seems to cor­rectly figure out some strong causes, but he’s much too ideal­ist in hop­ing one can de­sign the cul­tural con­text and ed­u­ca­tion sys­tems that would rid us of vi­o­lence. After all, we’ve been at it for a long while and no mat­ter how much money one throws at ed­u­ca­tion it doesn’t seem to stick.

The economist has found some causes, but they are high-level causes he uses for ev­ery­thing and his solu­tion is too vague to be ap­pli­ca­ble.

The ge­netic de­ter­minist seems to have cause and effect back­ward. He doesn’t un­der­stand the fact that hu­mans self-seg­re­gate into com­mu­ni­ties/​tribes based on phe­no­type, and some com­mu­ni­ties are forced into situ­a­tions that pro­mote vi­o­lence. His solu­tion seems to us morally ab­hor­rent and likely not to work un­less you liter­ally en­g­ineer a pop­u­la­tion of iden­ti­cal hu­mans. Even then, they’d likely find ways to make tribes and some of those tribes would be forced or ran­domly stum­ble into a cor­rupt equil­ibrium that pro­motes vi­o­lence.

The Freudian’s ex­pla­na­tion is out­right silly to mod­ern ears, but again, he seems to be get­ting at some­thing like a cause, even though it’s so ab­stract he might have well pointed to “God” as the cause. Con­versely, since his cause is so vague, so is his solu­tion.

But the statis­ti­cian seems to not even un­der­stand causal­ity. He’s con­fus­ing a cor­re­la­tion for cau­sa­tion.

Lead is not a cause of vi­o­lence, maybe it’s a proxy at best, an en­vi­ron­men­tal haz­ard that en­courages cer­tain be­hav­ior pat­terns, but a cause, nah, it’s...

1. Lead level (even if we only track mea­sures in the air) cor­re­lated with ag­gra­vated as­sault more so then an­tibiotics are with bac­te­rial in­fec­tion sur­vival. [link]

2. Strongly cor­re­lated (both high p-value and large effect size) with vi­o­lent crime as far back as the 20th cen­tury and the low­er­ing of crime rates as the cen­turies progress match its de­crease. [link]

3. Life­time ex­po­sure is strongly cor­re­lated (both high p-value and large effect size) with vi­o­lent crim­i­nal be­hav­ior. [link]

4. Strongly cor­re­lated in a fairly ho­moge­nous pop­u­la­tion with small vari­a­tions in lead ex­po­sure (same city) with gun vi­o­lence, homi­cide, and rape. [link]

Huh, I won­der if the other 4 can claim any­thing similar. And this is just me search­ing ar­bi­trary pri­mary sources on google scholar.

You can find hun­dreds of stud­ies look­ing at lead lev­els in the en­vi­ron­ment and body and their cor­re­la­tion with crime. In­clud­ing at the fact that de­creas­ing lead lev­els seem to de­crease vi­o­lence in the same de­mo­graphic and area where vi­o­lence pro­lifer­ated when lead lev­els were high.

The lead blood level in a tod­dler tracks vi­o­lent crime so well it’s al­most un­be­liev­able. Most drug com­pa­nies or ex­per­i­men­tal psy­chol­o­gist can’t hack their way into some­thing that looks 1/​3rd as con­vinc­ing as this graph.

Did I men­tion the in­ter­ven­tions that re­move lead by re­plac­ing the pipe or ban­ning leaded gasoline and see a sharp drop in crime rate only a few years af­ter­ward?

To my knowl­edge, what lit­tle cor­re­la­tion ed­u­ca­tion has with vi­o­lence van­ishes when con­trol­ling for so­cioe­co­nomic sta­tus.

Poverty is sur­pris­ingly un­cou­pled from vi­o­lence when looked at in the ab­stract (e.g. see rates of vi­o­lence in poor Asian coun­tries vs poor Euro­pean coun­tries and poor vs rich cities), when it can be con­sid­ered a proxy for vi­o­lence, the lead-vi­o­lence cor­re­la­tion would eat it up as just a con­founder.

Psy­cho­an­a­lyst ther­apy doesn’t seem what­so­ever re­lated to vi­o­lence, though due to the kind of peo­ple that usu­ally get it, it’s hard to de­con­found past a point.

One could ar­gue genes are re­lated to vi­o­lence from a snap­shot at a sin­gle point in time, but look­ing at vi­o­lence drop­ping in the same pop­u­la­tion over just a sin­gle gen­er­a­tion this doesn’t seem so good.

So, if we could cut vi­o­lent crime by 50% in a pop­u­la­tion by re­duc­ing serum lead lev­els to ~0 (a rea­son­able hy­poth­e­sis, po­ten­tially even un­der­stated)… then why can’t most peo­ple de­clare, with a straight face and proud voice, that lead is the sin­gle most im­por­tant cause of vi­o­lence? Why would any­one dis­agree with such a blatantly ob­vi­ous `C => P(E) ~= 1` state­ment? (Where `E` is some­thing like “re­duc­tion in vi­o­lent crime by be­tween 30 and 80%)

What if I make my hy­poth­e­sis stronger by adding some nu­tri­tional ad­vice to the mix? Some­thing like: re­duce lead blood level to ~0, re­duce boron blood level to as lit­tle as pos­si­ble, in­crease io­dine and omega-3 in­take to 2x RDA in ev­ery sin­gle mem­ber of a pop­u­la­tion.

If, this in­ter­ven­tion re­duced vi­o­lence in all pop­u­la­tions by ~90%, would I be able to claim:

Hey guys, I figured out the cause of hu­man vi­o­lence, ap­par­ently, it has to do with too much resi­d­ual lead and boron in the body cou­pled with lack of io­dine and omega-3. Good news, with a 5-year in­ter­ven­tion that costs less than 1% of the yearly US bud­get we can likely end al­most all crime and war.

I’d wa­ger the an­swer is, no and I think it’s no mainly for mis­guided rea­sons. It has to do with the aes­thet­ics we as­so­ci­ate with a cause. It’s the same rea­son why the but­terfly effect sounds silly.

Violence seems like such a fun­da­men­tal hu­man prob­lem to us that it seems silly be­yond be­lief that the cause was just some resi­d­ual heavy metal all along, or at least for the last 200 years or so.

Yet… I see no rea­son not to back up this claim. It seems a much stronger cause than any­thing else peo­ple have come up with based on the sci­en­tific ev­i­dence. It re­spects our pre­vi­ous defi­ni­tion of causal­ity, it gets ev­ery­thing right. Or, at least, much more so than any other hy­poth­e­sis one can test.

So re­ally, `P(E) ~= 1 | C` is not enough even if we use the sci­en­tific method to find the sim­plest `C` pos­si­ble. In­stead, it has to be some­thing like `P(E) ~= 1 | C` where `C` re­spects {spe­cific hu­man in­tu­ition that rea­sons about the kind of things that are meta­phys­i­cally valid to be causes for other things}.

This is where we get into is­sues be­cause “{spe­cific hu­man in­tu­ition that rea­sons about the kind of things that are meta­phys­i­cally valid to be causes for other things}” varies a lot be­tween peo­ple for ba­si­cally no rea­son.

It varies in that a physi­cist, chemist and biol­o­gist might think differ­ent of what a valid cause is. It also varies in that a per­son that grew up dis­ad­van­taged their whole life might have a fun­da­men­tally differ­ent un­der­stand­ing of “what a hu­man can cause” than some­one that grew up as the son of a pow­er­ful poli­ti­cian.

It varies based on com­plex tax­onomies of the world, the kind that clas­sifies things into lev­els of “im­por­tance” and tells us that a cause which is too many lev­els of im­por­tance bel­low an effect can­not be a “real cause”.

If e.g. love, vi­o­lence, and death are “in­tu­itive im­por­tance level 100”, then ed­u­ca­tion, eco­nomics, and so­cial sta­tus might be “in­tu­itive im­por­tance level 98″. On the other hand, lead blood lev­els, what we eat for break­fast, or our labrador’s own­er­ship sta­tus are closer to “in­tu­itive im­por­tance level 10”.

To say that some­thing that’s “in­tu­itive im­por­tance level 98” can cause some­thing that’s “in­tu­itive im­por­tance level 100“ sounds plau­si­ble to us. To say that some­thing that’s “in­tu­itive im­por­tance level 10” can cause some­thing that’s “in­tu­itive im­por­tance level 100” is blas­phemy.

## 5. Why I find causal­ity harmful

I ad­mit that I can’t quite pain a com­plete pic­ture of causal­ity in ~3000 words, but the more edge cases I’d cover, the leak­ier a con­cept causal­ity would seem to be­come.

Causal­ity seems like a sprawl­ing mess that can only be defined us­ing very broad statis­ti­cal con­cepts, to­gether with a spe­cific per­son’s or groups in­tu­ition about how to in­ves­ti­gate the world. And all of that is cou­pled pro­tected by a vague pseudo-re­li­gious veil that dic­tates taboos about what kind of things are “pure enough” or “im­por­tant enough” to serve as causes to other things on the same spec­trum of “im­por­tance” or “pu­rity”.

I cer­tainly think that causal­ity is a good lay­man term that we should keep us­ing in day to day in­ter­ac­tions. If my friend wants to in­gest a large quan­tity of cyanide I want to be able to tell them “Hey, you shouldn’t do that, cyanide causes de­cou­pling of mi­to­chon­drial elec­tron trans­port chains which in turn cause you to die in hor­rible agony”.

But if a sci­en­tist is study­ing “cyanide’s effects upon cer­tain mi­to­chon­drial res­pi­ra­tory com­plexes” I feel like this kind of re­search is rigor­ous enough to do away with the con­cept of causal­ity.

On the other hand, re­plac­ing causal­ity with very strict math­e­mat­i­cal for­mu­las that are tightly linked to the type of data we are look­ing at doesn’t seem like a solu­tion ei­ther. It might be a solu­tion in cer­tain cases, but it would make a lot of liter­a­ture pointlessly difficult to read.

How­ever, there might be some mid­dle ground where we re­place the ideas of “cause” and “causal­ity” with a few sub­species of such. Sub­species that could also stretch the defi­ni­tion to in­clude things like lead caus­ing vi­o­lence or but­terflies flap­ping their wings caus­ing thun­der­storms.

Maybe I am wrong here, I cer­tainly know it would be hard for me to stop us­ing causal lan­guage. But I will at least at­tempt to re­duce my us­age of such and/​or be more rigor­ous when I do end up us­ing it.

• TL;DR: Causal­ity is an ab­strac­tion, a fea­ture of our mod­els of the world, not of the world it­self, and some­times it is use­ful, but other times not so much. No­tice when it’s not use­ful and use other mod­els.

• Agreed. As­sign­ing causal­ity re­quires hav­ing made a choice about how to carve up the world into cat­e­gories so one part of the world can af­fect an­other. Without hav­ing made this choice we lose our nor­mal no­tion of causal­ity be­cause there are no things to cause other things, hence causal­ity as nor­mally for­mu­lated only makes sense within an on­tol­ogy.

And yet, there is some un­der­ly­ing phys­i­cal pro­cess which drives our abil­ity to model the world with the idea that things cause other things and we might rea­son­ably point to it and say it is the real causal­ity, i.e. the as­pect of ex­is­tence that we per­ceive as change.

• And yet, there is some un­der­ly­ing phys­i­cal pro­cess which drives our abil­ity to model the world with the idea that things cause other things and we might rea­son­ably point to it and say it is the real causal­ity, i.e. the as­pect of ex­is­tence that we per­ceive as change.

Hmm. Imag­ine the world as fully de­ter­minis­tic. Then there is no “real causal­ity” to speak of, ev­ery­thing is set in stone, and there is no differ­ence be­tween cause and effect. The “un­der­ly­ing phys­i­cal pro­cess which drives our abil­ity to model the world with the idea that things cause other things” are es­sen­tial in be­ing an em­bed­ded agent, since agency equals a per­ceived world op­ti­miza­tion, which re­quires, in turn, pre­dictabil­ity (from the in­side the world), but I don’t think any­one has a good han­dle on what “pre­dictabil­ity from in­side the world” may look like. Off hand, it means that there is a sub­set of the world that runs a coarse-grained simu­la­tion of the world, but how do you rec­og­nize such a simu­la­tion with­out already know­ing what you are look­ing for? Any­way, this is a bit of a tan­gent.

• Imag­ine the world as fully de­ter­minis­tic. Then there is no “real causal­ity” to speak of, ev­ery­thing is set in stone, and there is no differ­ence be­tween cause and effect.

If cau­sa­tion is un­der­stood in terms of coun­ter­fac­tu­als — X would have hap­pened if Y had hap­pened — then there is still a differ­ence be­tween cause and effect. A model of a world im­plies mod­els of hy­po­thet­i­cal, coun­ter­fac­tual wor­lds.

• If cau­sa­tion is un­der­stood in terms of coun­ter­fac­tu­als — X would have hap­pened if Y had hap­pened — then there is still a differ­ence be­tween cause and effect. A model of a world im­plies mod­els of hy­po­thet­i­cal, coun­ter­fac­tual wor­lds.

Yes, in­deed, in terms of coun­ter­fac­tu­als there is. But coun­ter­fac­tu­als are in the map (well, to be fair a map is a tiny part of the ter­ri­tory in the agent’s brain). Which was my origi­nal point: causal­ity is in the map.

• The map and the ter­ri­tory are not sep­a­rate mag­is­te­ria. A good map, or model, fits the ter­ri­tory: it al­lows one to make ac­cu­rate and re­li­able pre­dic­tions. That is what it is, for a map to be a good one. The things in the map have their coun­ter­parts in the world. The good­ness of fit of a map to the world is a fact about the world. Cau­sa­tion is there also, just as much as pi­anos, and grav­i­ta­tion, and quarks.

• This claim..

A good map, or model, fits the ter­ri­tory: it al­lows one to make ac­cu­rate and re­li­able pre­dic­tions. That is what it is, for a map to be a good one.

is not ob­vi­ously equiv­a­lent to this claim:-

The things in the map have their coun­ter­parts in the world.

Cau­sa­tion is there also, just as much as pi­anos, and grav­i­ta­tion, and quarks.

If you ac­cept use­ful­ness in the map as the sole crite­rion for ex­is­tence in the ter­ri­tory, then cau­sa­tion is there, along with much else, in­clud­ing much that you do not be­lieve in ,and much that is mu­tu­ally con­tra­dic­tory.

• Hmm. Imag­ine the world as fully de­ter­minis­tic. Then there is no “real causal­ity” to speak of, ev­ery­thing is set in stone, and there is no differ­ence be­tween cause and effect

There’s a differ­ence be­tween strict causal de­ter­minism and block uni­verse the­ory. Un­der causal de­ter­minism, fu­ture events have not hap­pened yet,and need to be caused, even though there is is only one way they can turn out. Whereas un­der the block uni­verse the­ory , the fu­ture is already “there”—on­tolog­i­cally fixed as well as episte­molog­i­cally fixed.

• Which is the cor­rect the­ory—the first para­graph or the sec­ond?

There is plenty of ev­i­dence that hu­man no­tions of causal­ity are in­fluenced by hu­man con­cerns, but it doesn’t add up to the con­clu­sion that there is no causal­ity in the ter­ri­tory. The com­par­i­son with on­tol­ogy is apt: just be­cause ta­bles and chairs are hu­man level on­tol­ogy, doesn’t mean that there’s no quark level on­tol­ogy to the uni­verse.

• What would it even mean to say a the­ory of causal­ity is “cor­rect” here? We’re talk­ing about what makes sense to ap­ply the term causal­ity to, and there’s mat­ter of cor­rect­ness at that level, only of use­ful­ness to some pur­pose. It’s only af­ter we have some sys­tem­atized way of fram­ing a ques­tion that we can ask if some­thing is cor­rect within that sys­tem.

• Cor­rect­ness as op­posed to use­ful­ness would be cor­re­spon­dence to re­al­ity.

There’s a gen­eral prob­lem of how to es­tab­lish cor­re­spon­dence, a prob­lem which ap­plies to many things other than causal­ity. You can’t in­fer that some­thing cor­re­sponds just be­cause it is use­ful, but you also can’t in­fer that some­thing does not cor­re­spond just be­cause it is use­ful—“in the map” does not im­ply “not in the ter­ri­tory”.

• +1 on the sprawl­ing mess. What I have per­son­ally found use­ful is figur­ing out what is go­ing on in terms of men­tal heuris­tics when some causal ex­pla­na­tions seem ‘bet­ter’ than oth­ers. Which in­volves type er­rors and de­grees of free­dom.

• Can you please elab­o­rate on type er­rors and de­grees of free­dom in terms of men­tal heuris­tics? I am not sure if I followed

• Type er­ror: con­sider Aris­to­tle’s 4 causes. If I ask you a why ques­tion about one kind of cause and you give me an ex­pla­na­tion about an­other kind of cause there has been a type er­ror.

De­grees of free­dom: if there are more de­grees of free­dom in your ex­pla­na­tion than in the thing you are at­tempt­ing to ex­plain then you can always get the an­swer you want. Con­sider as­trol­ogy. A good ex­pla­na­tion has fewer de­grees of free­dom than the thing it is ex­plain­ing and thus cre­ates com­pres­sion and pre­dic­tion power, i.e. it elimi­nates more pos­si­ble wor­lds whereas bad ex­pla­na­tions leave you with the same num­ber of pos­si­ble wor­lds as you started with.

• If I ask you a why ques­tion about one kind of cause and you give me an ex­pla­na­tion about an­other kind of cause there has been a type error

“At Milliways, you can go as many times as you like with­out meet­ing your­self, be­cause of the em­bar­rass­ment that would cause”.

• I have the op­po­site im­pres­sions. Science should em­brace causal­ity more and do it bet­ter. And as a lay­man term it should be re­fined so that we stop talk­ing about the causes of any event as a cake where each slice has a name and only one name.

I find it hard to sum­ma­rize why, at least right now, but my view is sorta similar to Pearl’s (though I don’t to­tally like how he puts it). Hope­fully later I’ll re-read this more at­ten­tively and com­ment some­thing more pro­duc­tive (if no one has done a strictly bet­ter job already).

• I be­lieve the thing we differ on might just be a se­man­tic, at least as far as re­defi­ni­tion goes. My fi­nal con­clu­sion is around the fact that the term is bad be­cause it’s ill-defined, but with a stronger defi­ni­tions (or ideally mul­ti­ple defi­ni­tions for differ­ent cases) it would be use­ful, it would also, how­ever, be very for­eign to a lot of peo­ple.

• P(E) ~= 1 | C (where “|” stands for “given”). If I can say this, I can most cer­tainly say that C causes E

Well… un­less P(E) also ~= 1 | !C be­cause P(E) ~= 1 and C is irrelevant

• Cor­rected the word­ing to be a bit “weaker” on that claim, but also, it’s just a start­ing point and the fi­nal defi­ni­tion I dis­pute against doesn’t rest on it.