Against responsibility

I am sur­rounded by well-mean­ing peo­ple try­ing to take re­spon­si­bil­ity for the fu­ture of the uni­verse. I think that this at­ti­tude – promi­nent among Effec­tive Altru­ists – is caus­ing great harm. I no­ticed this as part of a broader change in out­look, which I’ve been try­ing to de­scribe on this blog in man­age­able pieces (and some­times failing at the “man­age­able” part).

I’m go­ing to try to con­tex­tu­al­ize this by out­lin­ing the struc­ture of my over­all ar­gu­ment.

Why I am worried

Effec­tive Altru­ists of­ten say they’re mo­ti­vated by util­i­tar­i­anism. At its best, this leads to things like Katja Grace’s ex­cel­lent anal­y­sis of when to be a veg­e­tar­ian. We need more of this kind of prin­ci­pled rea­son­ing about trade­offs.

At its worst, this leads to some peo­ple angst­ing over whether it’s eth­i­cal to spend money on a cup of coffee when they might have saved a life, and oth­ers us­ing the greater good as li­cense to say things that are not quite true, so­cially pres­sure oth­ers into bear­ing in­ap­pro­pri­ate bur­dens, and make ever-in­creas­ing claims on re­sources with­out a cor­re­spond­ingly strong ver­ified track record of im­prov­ing peo­ple’s lives. I claim that these ac­tions are not in fact morally cor­rect, and that peo­ple keep wind­ing up en­dors­ing those con­clu­sions be­cause they are us­ing the wrong cog­ni­tive ap­prox­i­ma­tions to rea­son about moral­ity.

Sum­mary of the argument

  1. When peo­ple take re­spon­si­bil­ity for some­thing, they try to con­trol it. So, uni­ver­sal re­spon­si­bil­ity im­plies an at­tempt at uni­ver­sal con­trol.

  2. Max­i­miz­ing con­trol has de­struc­tive effects:

    • An ad­ver­sar­ial stance to­wards other agents.

    • De­ci­sion paral­y­sis.

  3. Th­ese failures are not ac­ci­den­tal, but baked into the struc­ture of con­trol-seek­ing. We need a prac­ti­cal moral philos­o­phy to de­scribe strate­gies that gen­er­al­ize bet­ter, and benefit from the ex­is­tence of other benev­olent agents, rather than treat­ing them pri­mar­ily as threats.

Re­spon­si­bil­ity im­plies control

In prac­tice, the way I see the peo­ple around me ap­ply­ing util­i­tar­i­anism, it seems to make two im­por­tant moral claims:

  1. You—you, per­son­ally—are re­spon­si­ble for ev­ery­thing that hap­pens.

  2. No one is al­lowed their own pri­vate per­spec­tive—ev­ery­one must take the pub­lic, com­mon per­spec­tive.

The first prin­ci­ple is al­most but not quite sim­ple con­se­quen­tial­ism. But it’s im­por­tant to note that it ac­tu­ally doesn’t gen­er­al­ize; it’s mas­sive dou­ble-count­ing if each in­di­vi­d­ual per­son is re­spon­si­ble for ev­ery­thing that hap­pens. I worked through an ex­am­ple of the dou­ble-count­ing prob­lem in my post on match­ing dona­tions.

The sec­ond prin­ci­ple fol­lows from the first one. If you think you’re per­son­ally re­spon­si­ble for ev­ery­thing that hap­pens, and obliged to do some­thing about that rather than weigh your taste ac­cord­ingly – and you also be­lieve that there are ways to have an out­sized im­pact (e.g. that you can re­li­ably save a life for a few thou­sand dol­lars) – then in some sense noth­ing is yours. The money you spent on that cup of coffee could have fed a poor fam­ily for a day in the de­vel­op­ing world. It’s only jus­tified if the few min­utes you save some­how pro­duce more value.

One way of re­solv­ing this is sim­ply to de­cide that you’re en­ti­tled to only as much as the global poor, and try to do with­out the rest to im­prove their lot. This is the rea­son­ing be­hind the no­to­ri­ous de­mand­ing­ness of util­i­tar­i­anism.

But of course, other peo­ple are also mak­ing sub­op­ti­mal uses of re­sources. So if you can change that, then it be­comes your re­spon­si­bil­ity to do so.

In gen­eral, if Alice and Bob both have some money, and Alice is mak­ing poor use of money by giv­ing to the So­ciety to Cure Rare Diseases in Cute Pup­pies, and Bob is giv­ing money to com­par­a­tively effec­tive char­i­ties like the Against Malaria Foun­da­tion, then if you can cause one of them to have ac­cess to more money, you’d rather help Bob than Alice.

There’s no rea­son for this to be differ­ent if you are one of Bob and Alice. And since you’ve already re­jected your own pri­vate right to hold onto things when there are stronger global claims to do oth­er­wise, there’s no prin­ci­pled rea­son not to try to re­al­lo­cate re­sources from the other per­son to you.

What you’re will­ing to do to your­self, you’ll be will­ing to do to oth­ers. Re­spect­ing their au­ton­omy be­comes a mere mat­ter of ei­ther self­ishly in­dulging your per­sonal taste for “de­on­tolog­i­cal prin­ci­ples,” or a con­ces­sion made be­cause they won’t ac­cept your lead­er­ship if you’re too de­mand­ing—not a prin­ci­pled way to co­op­er­ate with them. You end up try­ing to force your­self and oth­ers to obey your judg­ment about what ac­tions are best.

If you think of your­self as a benev­olent agent, and think of the rest of the world and all the peo­ple in it in as ob­jects with reg­u­lar, pre­dictable be­hav­iors you can use to im­prove out­comes, then you’ll feel morally obliged—and there­fore morally sanc­tioned—to shift as much of the lo­cus of con­trol as pos­si­ble to your­self, for the greater good.

If some­one else seems like a bet­ter can­di­date, then the right thing to do seems like throw­ing your lot in with them, and trans­fer­ring as much as you can to them rather than to your­self. So this at­ti­tude to­wards do­ing good leads ei­ther to per­sonal con­trol-seek­ing, or sup­port of some­one else’s bid for the same.

I think that this rea­son­ing is tac­itly ac­cepted by many Effec­tive Altru­ists, and ex­plains two seem­ingly op­po­site things:

  1. Some EAs get their act to­gether and make power plays, im­plic­itly claiming the right to de­ceive and ma­nipu­late to im­ple­ment their plan.

  2. Some EAs are par­a­lyzed by the im­pos­si­bil­ity of weigh­ing the con­se­quences for the uni­verse of ev­ery act, and col­lapse into per­pet­ual scrupu­los­ity and anx­iety, miti­gated only by some­one else claiming le­gi­t­i­macy, tel­ling them what to do, and tel­ling them how much is enough.

In­ter­est­ingly, peo­ple in the sec­ond cat­e­gory are some­what use­ful for peo­ple fol­low­ing the strat­egy of the first cat­e­gory, as they demon­strate de­mand for the ser­vice of tel­ling other peo­ple what to do. (I think the right thing to do is largely to de­cline to meet this de­mand.)

Ob­jec­tivists some­times crit­i­cize “al­tru­is­tic” ven­tures by in­sist­ing on Ayn Rand’s defi­ni­tion of al­tru­ism as the drive to self-ab­ne­ga­tion, rather than benev­olence. I used to think that this was ob­nox­iously miss­ing the point, but now I think this might be a fair de­scrip­tion of a large part of what I ac­tu­ally see. (I’m very much not sure I’m right. I am sure I’m not de­scribing all of Effec­tive Altru­ism – many peo­ple are do­ing good work for good rea­sons.)

Con­trol-seek­ing is harmful

You have to in­ter­act with other peo­ple some­how, since they’re where most of the value is in our world, and they have a lot of causal in­fluence on the things you care about. If you don’t treat them as in­de­pen­dent agents, and you don’t already rule over them, you will de­fault to go­ing to war against them (and more gen­er­ally try­ing to at­tain con­trol and then make all the de­ci­sions) rather than trad­ing with them (or let­ting them take care of a lot of the de­ci­sion­mak­ing). This is bad be­cause it de­stroys po­ten­tial gains from trade and di­vi­sion of la­bor, be­cause you win con­flicts by de­stroy­ing things of value, and be­cause even when you win you un­nec­es­sar­ily be­come a bot­tle­neck.

Peo­ple who think that con­trol-seek­ing is the best strat­egy for benev­olence tend to adopt plans like this:

Step 1 – ac­quire con­trol over ev­ery­thing.

Step 2 – op­ti­mize it for the good of all sen­tient be­ings.

The prob­lem with this is that step 1 does not gen­er­al­ize well. There are lots of differ­ent goals for which step 1 might seem like an ap­peal­ing first step, so you should ex­pect lots of other peo­ple to be try­ing, and their in­ter­ests will all be di­rectly op­posed to yours. Your meth­ods will be nearly the same as the meth­ods for some­one with a differ­ent step 2. You’ll never get to step 2 of this plan; it’s been tried many times be­fore, and failed ev­ery time.

Lots of differ­ent types of peo­ple want more re­sources. Many of them are very tal­ented. You should be skep­ti­cal about your abil­ity to win with­out some mas­sive ad­van­tage. So, what you’re left with are your prox­i­mate goals. Your im­pact on the world will be de­ter­mined by your means, not your ends.

What are your means?

Even though you value oth­ers’ well-be­ing in­trin­si­cally, when pur­su­ing your prox­i­mate goals, their agency mostly threat­ens to muck up your plans. Con­se­quently, it will seem like a bad idea to give them info or leave them re­sources that they might mi­suse.

You will want to make their be­hav­ior more pre­dictable to you, so you can in­fluence it bet­ter. That means tel­ling sim­plified sto­ries de­signed to cause good ac­tions, rather than to di­rectly trans­mit rele­vant in­for­ma­tion. With­hold­ing, rather than shar­ing, in­for­ma­tion. Mes­sage dis­ci­pline. I wrote about this prob­lem in my post on the hu­mil­ity ar­gu­ment for hon­esty.

And if the words you say are tools for caus­ing oth­ers to take spe­cific ac­tions, then you’re cor­rod­ing their use­ful­ness for liter­ally true de­scrip­tions of things far away or too large or small to see. Peter Singer’s claim that you can save a life for hun­dreds of dol­lars by giv­ing to de­vel­op­ing-world char­i­ties no longer means that you can save a life for hun­dreds of dol­lars by giv­ing to de­vel­op­ing-world char­i­ties. It sim­ply means that Peter Singer wants to mo­ti­vate you to give to de­vel­op­ing-world char­i­ties. I wrote about this prob­lem in my post on bind­ings and as­surances.

More gen­er­ally, you will try to min­i­mize oth­ers’ agency. If you be­lieve that other peo­ple are moral agents with com­mon val­ues, then e.g. with­hold­ing in­for­ma­tion means that the friendly agents around you are more poorly in­formed, which is ob­vi­ously bad, even be­fore tak­ing into ac­count trust con­sid­er­a­tions! This plan only makes sense if you ba­si­cally be­lieve that other peo­ple are moral pa­tients, but in­de­pen­dent, friendly agents do not ex­ist; that you are the only per­son in the world who can be re­spon­si­ble for any­thing.

Another spe­cific be­hav­ioral con­se­quence is that you’ll try to ac­quire re­sources even when you have no spe­cific plan for them. For in­stance, GiveWell’s im­pact page tracks costs they’ve im­posed on oth­ers – money moved, and at­ten­tion in the form of vis­its to their web­site – but not in­de­pen­dent mea­sures of out­comes im­proved, or the op­por­tu­nity cost of peo­ple who made a GiveWell-in­fluenced dona­tion. The im­pli­ca­tion is that peo­ple weren’t do­ing much good with their money or time any­way, so it’s a “free lunch” to gain con­trol over these.<fn>Their an­nual met­rics re­port goes into more de­tail and does track this, and finds that about a quar­ter of GiveWell-in­fluenced dona­tions were re­al­lo­cated from other de­vel­op­ing-world char­i­ties (and an­other quar­ter from de­vel­oped-world char­i­ties).</​fn> By con­trast, the Gates foun­da­tion’s Valen­tine’s day re­port to War­ren Buffet tracks noth­ing but de­vel­op­ing-world out­comes (but then ab­surdly takes credit for 100% of the im­prove­ment).

As usual, I’m not pick­ing on GiveWell be­cause they’re un­usu­ally bad – I’m pick­ing on GiveWell be­cause they’re un­usu­ally open. You should as­sume that similar but more se­cre­tive or­ga­ni­za­tions are worse by de­fault, not bet­ter.

This kind of di­ver­gent strat­egy doesn’t just di­rectly in­flict harms on other agents. It takes re­sources away from other agents that aren’t defend­ing them­selves, which forces them into a more ad­ver­sar­ial stance. It also earns jus­tified mis­trust, which means that if you fol­low this strat­egy, you burn co­op­er­a­tive bridges, forc­ing your­self farther down the ad­ver­sar­ial path.

I’ve writ­ten more about the choice be­tween con­ver­gent and di­ver­gent strate­gies in my post about the ne­glect­ed­ness con­sid­er­a­tion.

Sim­ple patches don’t undo the harms from ad­ver­sar­ial strategies

Since you’re benev­olent, you have the ad­van­tage of a goal in com­mon with many other peo­ple. Without aban­don­ing your ba­sic ac­quisi­tive strat­egy, you could try to have a se­cret hand­shake among peo­ple try­ing to take over the world for good rea­sons rather than bad. Ideally, this would let the benev­olent peo­ple take over the world, co­op­er­at­ing among them­selves. But, in prac­tice, any sim­ple shib­bo­leth can be faked; any­one can say they’re ac­quiring power for the greater good.

It’s a com­mon­place in var­i­ous dis­cus­sions among Effec­tive Altru­ists, when some­one iden­ti­fies an in­di­vi­d­ual or or­ga­ni­za­tion do­ing im­por­tant work, to sug­gest that we “per­suade them to be­come an EA” or “get an EA in the or­ga­ni­za­tion”, rather than di­rectly about ways to open up a di­alogue and co­op­er­ate. This is straight­for­wardly an at­tempt to get them to agree to the same shib­bo­leths in or­der to co­or­di­nate on a power-grab­bing strat­egy. And yet, the stan­dard of ev­i­dence we’re us­ing is mostly “iden­ti­fies as an EA”.

When Gleb Tsipursky tried to ex­tract re­sources from the Effec­tive Altru­ism move­ment with straight­for­ward low-qual­ity mime­sis, mouthing the words but not re­ally adding value, and grossly mis­rep­re­sent­ing what he was do­ing and his level of suc­cess, it took EAs a long time to no­tice the pat­tern of mis­be­hav­ior. I don’t think this is be­cause Gleb is es­pe­cially clever, or be­cause EAs are es­pe­cially bad at notic­ing things. I think this is be­cause EAs iden­tify each other by easy-to-mimic shib­bo­leths rather than mean­ingful stan­dards of be­hav­ior.

Nor is Effec­tive Altru­ism unique in suffer­ing from this prob­lem. When the Ro­man em­pire be­came too big to gov­ern, grad­u­ally em­per­ors hit upon the solu­tion of di­vid­ing the em­pire in two and pick­ing some­one to gov­ern the other half. This oc­ca­sion­ally worked very well, when the two em­per­ors had a strong pre­ex­ist­ing bond, but gen­er­ally they dis­trusted each other enough that the two em­pires be­haved like ri­val states as of­ten as they be­haved like al­lies. Even though both em­per­ors were Ro­mans, and of­ten close rel­a­tives!

Us­ing “be­lieve me” as our stan­dard of ev­i­dence will not work out well for us. The Pres­i­dent of the United States seems to have fol­lowed the strat­egy of say­ing the thing that’s most con­ve­nient, whether or not it hap­pens to be true, and won an elec­tion based on this. Others can and will use this strat­egy against us.

We can do better

The above is all a symp­tom of not in­clud­ing other moral agents in your model of the world. We need a moral the­ory that takes this into ac­count in its de­scrip­tions (rather than hav­ing to do a de­tailed calcu­la­tion each time), and yet is scope-sen­si­tive and con­se­quen­tial­ist the way EAs want to be.

There are two im­por­tant desider­ata for such a the­ory:

  1. It needs to take into ac­count the fact that there are other agents who also have moral rea­son­ing. We shouldn’t be sad to learn that oth­ers rea­son the way we do.

  2. Grace­ful degra­da­tion. We can’t be so trust­ing that we can be defrauded by any­one will­ing to say they’re one of us. Our moral the­ory has to work even if not ev­ery­one fol­lows it. It should also de­grade grace­fully within an in­di­vi­d­ual – you shouldn’t have to be perfect to see benefits.

One thing we can do now is stop us­ing wrong moral rea­son­ing to ex­cuse de­struc­tive be­hav­ior. Un­til we have a good the­ory, the an­swer is we don’t know if your clever ar­gu­ment is valid.

On the ex­plicit and sys­tem­atic level, the di­ver­gent force is so dom­i­nant in our world that sincere benev­olent peo­ple sim­ply as­sume, when they see some­one overtly op­ti­miz­ing for an out­come, that this per­son is op­ti­miz­ing for evil. This leads to per­cep­tive peo­ple who don’t like do­ing harm, like Venkatesh Rao, to ex­plic­itly ad­vise oth­ers to min­i­mize their mea­surable im­pact on the world.

I don’t think this im­pact-min­i­miza­tion is right, but on cur­rent mar­gins it’s prob­a­bly a good cor­rec­tive.

One en­courag­ing thing is that many peo­ple us­ing com­mon-sense moral rea­son­ing already be­have ac­cord­ing to norms that re­spect and try to co­op­er­ate with the moral agency of oth­ers. I wrote about this in Hum­ble Char­lie.

I’ve also be­gun to try to live up to co­op­er­a­tive heuris­tics even if I don’t have all the de­tails worked out, and help my friends do the same. For in­stance, I’m happy to talk to peo­ple mak­ing giv­ing de­ci­sions, but usu­ally I don’t go any farther than con­nect­ing them with peo­ple they might be in­ter­ested in, or coach­ing them through heuris­tics, be­cause do­ing more would be harm­ful, it would de­stroy in­for­ma­tion, and I’m not om­ni­scient, oth­er­wise I’d be richer.

A move­ment like Effec­tive Altru­ism, ex­plic­itly built around overt op­ti­miza­tion, can only suc­ceed in the long run at ac­tu­ally do­ing good with (a) a clear un­der­stand­ing of this prob­lem, (b) a so­cial en­vi­ron­ment en­g­ineered to ro­bustly re­ject cost-max­i­miza­tion, and (c) an in­tel­lec­tual tra­di­tion of op­ti­miz­ing only for ac­tu­ally good things that peo­ple can an­chor on and learn from.

This was only a sum­mary. I don’t ex­pect many peo­ple to be per­suaded by this alone. I’m go­ing to fill in the de­tails in the fu­ture posts. If you want to help me write things that are rele­vant, you can re­spond to this (prefer­ably pub­li­cly), let­ting me know:

  • What seems clearly true?

  • Which parts seem most sur­pris­ing and in need of jus­tifi­ca­tion or ex­pla­na­tion?

(Cross-posted at my per­sonal blog.)