本文深入探讨了“能动性”(Agency)这一概念及其相关变体，如“稳健的能动性”、“意向性”等。文章从社会学角度出发，阐述了能动性作为个体实现潜力的能力，并受认知结构、社会认知及环境等多重因素影响。文章还对比了“能动性”与“意向性”，指出“意向性”强调行为的可预测性，与目标实现紧密相连。作者通过对不同情境下“能动性”的分析，如网球运动员的例子，质疑了传统定义，并提出了用“影响”代替“能动性”的观点。最后，作者反思了如何更有效地使用“能动性”及其相关概念，以便更好理解行动背后的意图。

💪 能动性(Agency)是指个体行动的能力，受到认知、社会认知和环境等多重因素影响。它关乎个体是否有能力和资源来实现自己的潜力，受歧视、偏见等负面因素限制。

🎯 意向性(Intentionality) 指的是系统行为的可预测性，即通过假设系统具有意图、信念和欲望来预测其行为。与此相对的是“斯芬克斯式”行为，即基于触发-行动模式而非目标驱动的行为。

💡 稳健的能动性(Robust Agency) 指的是在不同领域和情境下都能根据意图行动，即使方法改变也能保持意图不变。它体现了行动的资源性和创新性，有时其方法过于复杂，以至于不那么智能的个体难以理解其意图。

🤔 作者认为，当观察者能够通过赋予系统或个体意图来更好地预测其行为时，就存在能动性。即使系统行为未能实现其目标，只要其行为可以基于意图预测，也应被认为是能动性的体现。

🌟 作者提出用“影响”(Influence) 来代替“能动性”，以描述改变环境的潜力和能力，无论是否与意图一致。同时，提出用“成功”(Successful)或“明智”(Phronesis)来描述实现意图的行动。

Published on January 11, 2025 3:22 AM GMT

And by "we" I mean "I". I'm the one struggling.

Agency with a 'y', basically means “The condition of being in action; operation.” Or the means or mode of acting, the context I hear it used most often is sociologically:

"Agency is the capacity of individuals to have the power and resources to fulfill their potential." One's agency is one's independent capability or ability to act on one's will. This ability is affected by the cognitive belief structure which one has formed through one's experiences, and the perceptions held by the society and the individual, of the structures and circumstances of the environment one is in and the position one is born into."

People who are discriminated against for job opportunities have limited agency owing to the prejudices or negative perceptions people have which may be at odds with their ability: this robs them of the power to fulfill their potential.

People who internalize prejudicial ideas of others about them or negative self-image may be unnecessarily impeding their own agency. Arnold Schwarzenegger in his motivational speeches speaks about how he didn't allow these prejudicial ideas about his long name and non-native English Accent, or even his Olympic physique (which was a world away from the 'stars' of the day like Jack Nicholson and Dustin Hoffman) to impede on his career: this I think serves as a quick example of the impact of internalizing or not internalizing social beliefs on agency. He claims he did not allow these prejudices to rob him of his potential.

HERE’S WHERE I GET CONFUSED: "AGENCY", "ROBUST AGENCY", "INTENTIONALITY", "AGENTIC", "AGENTICNESS"….

Agency sometimes “Robust Agency” is mentioned a lot on here. But not only that "Agentic", with a "c", indicates something very different:

"the more you can predict its actions from its goals since its actions will be whatever will maximize the chances of achieving its goals.

This stems, at least some what from Daniel Dennet's Intentionality Systems Theory which tautologically posits that a system has intention if assuming it has intentions, beliefs, and desires allows us to better predict it's behavior. Compare to something which is sphexish - that appears to act not with intention, but on certain trigger-action cues which may or may not act towards overarching goals. Or to put it another way: imputing onto a system/person an intention doesn't help you predict it's behaviour any better than using the physical, or design stance.

If I understand this right, when an A.I. consistently acts intentionally in a way that gets results or consequences that accord with your intentions (as a developer let's say) then it is both agentic and Corrigible. My batch file for renaming photos is not Agentic but Sphexish, because it is a series of if-then statements and not more easily predicted by imputing intentions to it. A rogue A.I. may be agentic, but it is incorrigible.

A Robust Agent is broadly agentic  - as in acts in accord with the intentions we impute to on it - across a lot of different domains and, if I understand correctly, maintains the same intentions even if it changes methods^[1] The robustness is indicative of the resourcefulness and innovation it displays in working towards it's intentions by any number of means.
 

In some cases a highly Robust Agent may appear to not be acting intentionally on first blush, because it's methodology is so advanced in comparison to a less robust or intelligent agent won't be able to discern how their method of achieving their intention connects to their actions^[2]. The assumption is that a robust agent, to a similarly effective agent would appear to still be acting intentionally.

However, I've seen other definitions for Agency (with a 'y') floating about, for example this one seems to say we only notice agency when it acts towards some goal better than us:

We perceive agency when something is better at doing something than us; we endorse some aspect of its reasoning or activity. (Endorse as a way of achieving its goals, if not necessarily our own.)

As a side note I disagree with this. Rafael Nadal will recognize the agency of a lesser Tennis Player, either as an observer to a game or playing against them, especially if they manage to score against him. If Nadal wins the game and therefore is doing “tennis” better that doesn't mean he will not recognize the agency of the losing player.

That is... Unless you want to get really reductionist and atomic and therefore say that the lesser player exhibits Agency when, and only when, they score (better serve, less mobile etc. etc.), and when Nadal scores the lesser player is acting Sphexish? But also, vice versa, Nadal when he loses to this player is acting Sphexish, but when he scores isn’t. What about when they rally? Are they both Sphexish… I suppose on some level they are because there are trained to operate on reflexes that are faster than their ability to consciously process the trajectory of the ball and intuitively compute the next moves of each other. 

Presumably both players are acting intentionally, they both intend to win -and they both are deploying the most effective methods they have at their disposal. The fact that Nadal is better doesn't negate the agenticness of the other player: we still can impute the intention to win on the other player. 

While here Agency (again with a 'y') is defined in a way which resembles robust agency but not what I might call general agency or non-robust agency:

By ‘agency’ I mean the ability to get what you want across different contexts – the general skill of coming up with ambitious goals and actually achieving them, whatever they are. In my experience, I’ve found this is learnable

But the author also uses 'agentic' (with a 'c') to describe a personality pattern than can be misconstrued as bossy, but is more positively seen as confident, and less likely to seek permission before acting. I find this confusing but don't necessarily disagree with it. Because on first blush this is something I'd more commonly refer to as "initiative":

energy or aptitude displayed in initiation of action

Initiative or enthusiastic agentic-ness is only agentic (in the sense of intention) if it is successful. Indiscriminately trying different things and failing all of or most of them would appear to be the opposite of agency if agency means to act in accordance with goals or intentions. Indeed, this tends towards Sphexiness. In an A.I. this may be seen as incorrigible.

Going back to Nadal, an amateur tennis player who whacks the ball indiscriminately would not be agentic. A novice player who only always reacts in the same way, so much so that Nadal would managed to notice their tactics, and play against them, the novice player would be acting Sphexishly, but Nadal would be displaying some kind of intentionality by changing his tactics in reaction to this.

I'm not sure, if you take a lot of initiative, and it is successful, are you agentic? If you choose not to act, to be patient, then aren't you also acting in accordance with your intentions if it is the best method?

WHAT DO WE DO ABOUT THIS?

What would make the word "Agency", and its variants, maximally "agentic" in the sense that it makes it easier for others to impute our intentions when using it?

I don't know. But I'm keen to hear what other have to say about this. Perhaps there is value in the nuance, perhaps the word agency is robust because of how it is applied? Or is it a amorphous porridge word that is too malleable for it's own good?
 

What I do feel qualified to point out is where the consensus lies: that agency is when an observer imputing an intention onto a system or person better predicts what they might do than a lower stance like the design or physical stance. This includes highly sophisticated agents which have specific domain expertise, or are so robust across several domains, that they may appear to be incomprehensible to us – but to a similar level of agent would appear to have intention.

Going forth, I’m going to avoid using the word ‘agency’ and instead describe the potential or capacity to affect change to one’s environment – be it in alignment with intentions or not – as “influence”.

While imputing intentions onto an agent, where the agent realizes their intentions I might call Phronesis. Maybe even more generally “successful”? I have to decide which one is less burdened by connotations. (i.e. how meta is this; I have to decide which one better suites my intentions!).

This leaves another category – there needs to be a word to describe when a system can be predicted by an intention, but fails to realize that intention. To put it another way, imputing some goal, desires and beliefs to the systems allows us to predict how it will act – even if those actions are ineffectual in realizing it’s goals. If I was being annoying, I’d suggest “gentic” (i.e. “a-gent” is the opposite of “gent”, like how 'acyclical' is the opposite of 'cyclical'). But “unsuccessful” or “ineffectual” might be better. I don’t know yet. Maybe I'm displaying a distinct lack of agency?

 

1. ^{^}

   (I believe in Classical Philosophy this is analogous to Prudence or Phronesis – which, at least according to Aristotle is the aptitude for deciding on the course of action which best suites the interests of the agent which are not determined by specific disciplines or crafts (techne): the agent may be a human, a military general on behalf of his army, or a polis).

2. ^{^}

   (Now: I would caution isn’t it possible for non-robust agents to act intentionally and yet appear not be, as they are in possession of specific domain knowledge? For example, imagine a child who is exceedingly good at macramé so much so that it is inconceivable when they first start what they are doing: they would be a less robust agent than the average adult, but a particularly strong non-robust agent at knotting macramé.) 

Discuss