On the Nature of Agency

Epistemic sta­tus: Fairly high con­fi­dence. Prob­a­bly not com­plete, but I do think pieces pre­sented are all part of the true pic­ture.

Agency and be­ing an agent are com­mon terms within the Effec­tive Altru­ist and Ra­tion­al­ist com­mu­ni­ties. I have only used heard them used pos­i­tively, typ­i­cally to praise some­one’s virtue as an agent or to de­cry the lack of agents and agency.

As a con­cept, agency is re­lated to plan­ning. Since I’ve been writ­ing about plan­ning of late, I thought I’d at­tempt a break­down of agency within my gen­eral plan­ning paradigm. I apol­o­gize that this write-up is a lit­tle rushed.

Ex­am­ples of Agency and Non-Agency

Keep­ing the ex­po­si­tion con­crete, let’s start with some in­stances of things I ex­pect to be de­scribed as more or less agen­tic.

Things likely to be de­scribed as more agen­tic:

  • Set­ting non-triv­ial, non-stan­dard (am­bi­tious) goals and achiev­ing them.

  • Drop­ping out of school and found­ing a startup.

  • Be­com­ing pres­i­dent.

  • Mak­ing a mil­lion dol­lars.

  • Self-teach­ing rather than en­rol­ling in courses.

  • Build­ing and cus­tomiz­ing your own things rather than pur­chas­ing pre-made.

  • Re­search­ing niche ar­eas which you think are promis­ing in­stead of the pop­u­lar ones.

  • Dis­re­gard­ing typ­i­cal re­la­tion­ship struc­tures and cre­at­ing your own which work for you.

  • Notic­ing there are is­sues in your work­place and in­sti­gat­ing change.

  • Notic­ing there are is­sues in your so­ciety and in­sti­gat­ing change.

  • Gam­ing the sys­tem (es­pe­cially when the sys­tem is un­just).

  • Read­ing di­verse ma­te­ri­als to form your own mod­els and opinions.

  • Be­ing able to re­ceive a task with limited in­struc­tion and able to ex­e­cute it com­pe­tently to a high stan­dard.

  • Ig­nor­ing con­ven­tional ad­vice and in­vent­ing your own bet­ter way.

  • Hav­ing hereti­cal thoughts and be­liev­ing things other peo­ple think are wrong or crazy.

  • Strong will­ing­ness to trust one’s own opinion de­spite the views of oth­ers and even ex­perts.

  • Ac­com­plish­ing any difficult task which most peo­ple are un­able to.

Things likely to be de­scribed as less agen­tic:

  • Spend­ing your life work­ing on the fam­ily farm like your par­ents be­fore you.

  • Un­ques­tion­ingly adopt­ing the views of your friends, fam­ily, faith, or other au­thor­i­ties.

  • Pro­ceed­ing through school, un­der­grad­u­ate, grad­u­ate de­gree as the sim­plest path­way.

  • Stick­ing to con­ser­va­tively pres­ti­gious or sta­ble pro­fes­sions like medicine, law, nurs­ing, con­struc­tion, teach­ing, or even pro­gram­ming.

  • Seek­ing so­cial ap­proval for one’s plans and ac­tions or at least en­sur­ing that one’s ac­tions do not leave the range of typ­i­cally so­cially-ap­proved ac­tions.

  • Pri­ori­tiz­ing safety, se­cu­rity, and sta­bil­ity over gam­bits for greater gain.

  • Re­quiring spe­cific di­rec­tion, in­struc­tion, train­ing, or guidance to com­plete novel tasks.

  • Fol­low­ing cur­rent fads or what’s in fash­ion. Gen­er­ally high imi­ta­tion of oth­ers.

I have not worked hard to craft these lists so I doubt they are prop­erly com­pre­hen­sive or rep­re­sen­ta­tive, but they should suffice to get us on the same page.

At times it has been pop­u­lar, and ad­mit­tedly con­tro­ver­sial, to speak of how some peo­ple are PCs (player char­ac­ters) and oth­ers are mere NPCs (non-player char­ac­ters). PCs (agents) do in­ter­est­ing things and save the day. NPCs (non-agents) fol­lowed scripted, bor­ing be­hav­iors like stock and man the village store for the du­ra­tion of the game. PCs are the heroes, NPCs are not. (It is usu­ally the case that any­one is ac­com­plished or im­pres­sive is granted the ti­tle of agent.)

The In­gre­di­ents of Agency

What causes peo­ple in one list to be agen­tic and those in other to be not so? A ready an­swer is that peo­ple be­ing agen­tic are will­ing to be weird. The ex­am­ples di­vide nicely along con­for­mity vs non­con­for­mity, do­ing what ev­ery­one else does vs forg­ing your own path.

This is em­phat­i­cally true—agency re­quires will­ing­ness to be differ­ent—but I ar­gue that it is in­ci­den­tal. If you think agency is about be­ing weird, you have missed the point. Though it is not overly ap­par­ent from the ex­am­ples, the core of agency is about ac­com­plish­ing goals strate­gi­cally. Fore­most, an agent has a goal and is try­ing to se­lect their ac­tions so as to ac­com­plish that goal.

But in a way, so does ev­ery­one. We need a lit­tle more de­tail than this stan­dard defi­ni­tion that you’ve prob­a­bly heard already. Even if we say that a com­puter NPC is mind­lessly ex­e­cut­ing their pro­gram­ming, a hu­man shop­keeper le­gi­t­i­mately does have their own goals and val­ues to­wards which their ac­tions con­tribute. It should be un­con­tro­ver­sial to say that all hu­mans are choos­ing their ac­tions in a way that digi­tal video game NPCs are not. So what makes the differ­ence be­tween a bor­ing hu­man shop­keeper and Barack Obama?

It is not that one chooses their ac­tions and the other does not at all, but rather the pro­cess by which they do so.

First, we must note that plan­ning is re­ally, su­per-duper, frick­ing hard. Plan­ning well re­quires the abil­ity to pre­dict re­al­ity well and do some se­ri­ously in­volved com­pu­ta­tion. Given this, one of the eas­iest ways to plan is to model your plan off some­one else’s. It’s even bet­ter if you can model your plan of those ex­e­cuted by dozens, hun­dreds, or thou­sands of oth­ers. When you choose ac­tions already taken by oth­ers, you have ac­cess to some re­ally good data about will hap­pen when you take those ac­tions. If I want to go to grad school, there’s a large sup­ply of peo­ple I could talk to for ad­vice. By imi­tat­ing the plans of oth­ers, I en­sure that I prob­a­bly won’t get any worse re­sults than they did, plus it’s eas­ier to know which plans are low-var­i­ance when lots of peo­ple have tried them.

The differ­ence is that agents are usu­ally ex­e­cut­ing new com­pu­ta­tion and tak­ing risks with plans that have much higher un­cer­tainty and higher risk as­so­ci­ated. The non-agent gets to rely on the fact that many peo­ple’s model thought par­tic­u­lar ac­tions were a good idea, whereas the agent much more needs to rely on their own mod­els.

Con­sider the archetyp­i­cal founders drop­ping out of col­lege to work on their idea (back be­fore this was a cool, ad­mirable archetype). Most peo­ple were fol­low­ing a path­way with a pre­dictably good out­come. Woz­niak, Jobs, and Gates prob­a­bly would have grad­u­ated and got­ten fine jobs just like peo­ple in their refer­ence class. But they in­stead calcu­lated that a bet­ter op­tion for them was to drop out with the at­ten­dant risk. This was a course of ac­tion that stemmed from them think­ing for them­selves what would most lead to­wards their goals and val­ues. Bring­ing their own mod­els and com­pu­ta­tion to the situ­a­tion.

This bumps into an­other fea­ture of agency: agents who are run­ning their own ac­tion-se­lec­tion com­pu­ta­tion for them­selves rather than imi­tat­ing oth­ers (in­clud­ing their past selves) are able to be a lot more re­spon­sive to their in­di­vi­d­ual situ­a­tion. Plans made my the col­lec­tive have limited abil­ity to in­clude pa­ram­e­ters which cus­tomize the plan to the in­di­vi­d­ual.

Re­turn­ing to the ques­tion of will­ing­ness to be weird: it is more a pre­req­ui­site for agency than the core defi­ni­tion. An agent who is try­ing to ac­com­plish a goal as strate­gi­cally as pos­si­ble and who is run­ning new com­pu­ta­tion and perform­ing a search for the op­ti­mal plan for them—they sim­ply don’t want to be re­stricted to any ex­ist­ing solu­tions. If an ex­ist­ing solu­tion is the best, no prob­lem, it’s just that you don’t want to throw out an op­ti­mal solu­tion just be­cause it’s un­usual.

What other peo­ple do is use­ful data, but to an agent it won’t in­her­ently be a limi­ta­tion. (Ad­mit­tedly, you do have to ac­count for how other peo­ple will re­act to your de­viance in your plans. More on this soon.)

Mini-sum­mary: an agent tries to ac­com­plish a goal by run­ning rel­a­tively more of their own new com­pu­ta­tion/​plan­ning rel­a­tive to pure imi­ta­tion of cached plans of oth­ers or their past selves; they will not dis­card plans sim­ply be­cause they are un­usual.

Now why be agen­tic? When you imi­tate the plans of oth­ers, you pro­tect against down­side risk and likely won’t get worse than most. On the other hand, you prob­a­bly won’t get bet­ter re­sults ei­ther. You cap your ex­pected out­comes within a com­fortable range.

I sus­pect that among the traits which cause peo­ple to ex­hibit the be­hav­iors we con­sider agen­tic are:

  • A sense that more is pos­si­ble. They be­lieve that there are reach­able out­comes much bet­ter than the ex­ist­ing de­fault.

  • An as­pira­tion, striv­ing, or am­bi­tion to the more which can they en­vi­sion.

  • Some­thing to pro­tect.

  • Con­versely, com­pla­cency is the en­emy of agency.

There has to be some­thing which makes a per­son want to in­vest the effort to come up with their own plans rather than march­ing along the beaten paths with ev­ery­one else.

Or maybe not, maybe some peo­ple have pow­er­ful and ac­tive minds so that it’s rel­a­tively cheap to them to be think­ing fresh for them­selves. Maybe in their case, the im­pe­tus is bore­dom.

An agent must be­lieve that more is pos­si­ble, and more cru­cially they must be­lieve that it pos­si­ble for them to cause that more. This cor­re­sponds to the lo­cus of con­trol and self-effi­cacy vari­ables in the core self-eval­u­a­tions frame­work.

Fur­ther, any agent whose sig­nifi­cant work you’re able to see has likely pos­sessed a good mea­sure of con­scien­tious­ness. I’m not sure if lazy ge­niuses might count as an ex­cep­tion. Still, I ex­pect a strong cor­re­la­tion here. Most peo­ple who are con­scien­tious are not agents, but those agents you who ob­serve are prob­a­bly con­scien­tious.

The last few traits could be con­sid­ered “pos­i­tive traits” ac­tive traits that agents must pos­sess. There are also “nega­tive traits”, traits that most peo­ple have and agents must have less of.

Agents strive for more, but the price they pay is a will­ing­ness to risk get­ting even less. If you drop out of col­lege, you make mil­lions of dol­lars or you might end up broke and with­out a de­gree. When you make your own plans, pos­si­bly go off the beaten path, there is like­li­hood of failure. What’s worse, if you fail then you can be blamed for your failure. Pity may be with­held be­cause you could have played it safe and gone along with ev­ery­one else, and in­stead, you de­cided to be weird.

Across all the differ­ent situ­a­tions, agents might be risk­ing money, home, re­spect, limb, life, love, ca­reer, free­dom and all else they value. Not ev­ery­one has the con­sti­tu­tion for that.

Now, just a bit more needs to be said around agents and the so­cial situ­a­tion. Above it was im­plied that the plans are of oth­ers are es­sen­tially or­thog­o­nal to those of an agent. They’re not limited by them. That is true as far as the plan­ning pro­cess goes, but as far as en­act­ing one’s plans goes, it takes a lit­tle more.

An agent doesn’t just risk that their un­usual plans might fail in ways more stan­dard plans don’t, they also have to risk they will 1) lose out on ap­proval be­cause they are not do­ing the stan­dard things, 2) ac­tively be pun­ished for be­ing a de­viant with their plans.

If there is sta­tus at­tached to go­ing along cer­tain pop­u­lar path­ways, e.g. work­ing in the right pres­ti­gious or­ga­ni­za­tions, then any­one who de­cides to fol­low a differ­ent plan that only makes sense to them must nec­es­sar­ily forego sta­tus they might have oth­er­wise at­tained. (Per­haps they are gam­bling that they’ll make more even­tu­ally on their own path, but at least at first they are fore­go­ing.) This cre­ates a strong filter that agents are those peo­ple who were ei­ther in­differ­ent to sta­tus or will­ing to sac­ri­fice it for greater gain.

Ideally it would only be po­ten­tially fore­gone sta­tus which would af­fect agents, in­stead there is the fur­ther el­e­ment that de­viance is of­ten ac­tively pun­ished. It’s the stereo­type that the es­tab­lish­ment strikes out against the anti-es­tab­lish­ment. Every­one group will have its known truths and its taboos. Ar­ro­gance and hubris are sins. We are hyp­ocrites who si­mul­ta­neously praise those who have gone above and be­yond while sneer­ing at those who at­tempt to do the same. Agents must have thick skin.

In­deed, agents must have thick skin and be will­ing to gam­ble. In con­trast, imi­ta­tion (which ap­prox­i­mates non-agency) serves the mul­ti­fold func­tion of a) sav­ing com­pu­ta­tion, b) re­duc­ing risk, and c) guard­ing against so­cial op­pro­brium and even op­ti­miz­ing for so­cial re­ward.

Every­day Agents

I fear the above dis­cus­sion of agency has tended too grandiose, too much to­wards rev­olu­tion­ar­ies and founders of billion dol­lar com­pa­nies. Really though, we need agency on much more mun­dane scales too.

Con­sider that an agen­tic em­ployee is a supremely use­ful em­ployee since:

  • If you give them a task with limited in­struc­tion, they will use their own new com­pu­ta­tion/​plan­ning to figure out how to ex­e­cute it well. They don’t need things step by step.

  • They will sup­ply their own sense that more is pos­si­ble and push for ex­cel­lence.

  • They will take ini­ti­a­tive to make things bet­ter be­cause of their sense that more is pos­si­ble.

  • They will not be in­hibited by ex­ces­sive fear of failure or your dis­ap­proval be­cause they did some­thing other than fol­low ex­plicit in­struc­tions.

  • They’re will­ing to take on new and un­usual tasks and learn new skills be­cause:

    • They have high self-efficacy

    • They’re in the habit of think­ing their own fresh thoughts in­stead of imi­tat­ing and en­act­ing ready-made plans.

    • They’re will­ing to fail in the course of trial-and-er­ror to figure things out.

An agen­tic em­ployee is the kind of em­ployee who doesn’t suc­cumb to defen­sive de­ci­sion-mak­ing.

Why Agency is Uncommon

The dis­cus­sion so far can be sum­ma­rized neatly by say­ing what is which makes agency un­com­mon:

  • Agency is rare be­cause it in­volves plan­ning for your­self, go­ing off the beaten the path rather than imi­tat­ing and copy­ing the plans of oth­ers or your past self. Plan­ning for your­self is re­ally, re­ally hard.

    • It re­quires the skill of plan­ning for your­self.

    • It re­quires the ex­pen­di­ture of effort to do so.

  • Agency re­quires both a sense that more is pos­si­ble and a striv­ing to reach that more. Con­versely, agency is poi­soned by the pres­ence of com­pla­cency.

  • Agency re­quires be­lief in one’s self-effi­cacy and that one is the lo­cus of con­trol in their life.

  • Agency re­quires lower than av­er­age risk-aver­sion since at­tempt­ing po­ten­tially non-stan­dard plans means risk­ing non-stan­dard failure.

    • In par­tic­u­lar, it re­quires low so­cial risk-aver­sion.

    • This ap­plies at both the macro and micro scale.

  • Agency re­quires con­scien­tious­ness.

  • Agency re­quires a re­silience to so­cial sac­ri­fice ei­ther pas­sively via fore­gone sta­tus or ap­proval or ac­tively via the pun­ish­ment re­ceived for de­vi­at­ing from the norm.

Agent/​Non-Agent More and Less Agentic

This post is pri­mar­ily writ­ten in terms of agents and non-agents. While con­ve­nient, this lan­guage is dan­ger­ous. I fear that when be­ing an agent is cool, ev­ery­one will think them­selves is one and go to sleep each night con­grat­u­lat­ing them­selves for be­ing an agent un­like all those bad dumb non-agents.

Bet­ter to treat agency is a spec­trum upon which you can be scor­ing higher or lower on any given day.

  • How agen­tic was I to­day?

  • Was I be­ing too risk-averse to­day?

  • Was I wor­ry­ing too much about so­cial ap­proval?

  • Am I try­ing to think with fresh eyes and from first prin­ci­ples of new ways I could ac­com­plish my goals? Or am I just re­hash­ing the same pos­si­bil­ities again and again?

Ad­den­dum: Mys­te­ri­ous Old Wizards

A friend of mine has the hy­poth­e­sis that a pri­mary way to cause peo­ple to be more agen­tic is to have some­one be their mys­te­ri­ous old wiz­ard a la Gan­dalf, Dum­ble­dore, and Quir­rell. A mys­te­ri­ous old wiz­ard shows up, be­lieves in some­one, and prob­a­bly says some mys­te­ri­ous stuff, and this help in­duces agency.

I can see this work­ing. This might have hap­pened to me a bit, too. If some­one shows up and is suffi­ciently high-sta­tus in your mind, and they tell you that you are ca­pa­ble of great things, they can cause all the fol­low­ing:

  • You al­low your­self to be­lieve more is pos­si­ble be­cause the wiz­ard be­lieves it too.

  • You be­lieve that you are ca­pa­ble (self-effi­cacy, lo­cus of con­trol) be­cause the wiz­ard does.

  • You are will­ing to go on your quest de­spite so­cial op­pro­brium be­cause now you only care about the so­ciety of you and the wiz­ard, not any­one else.

I can see it work­ing.