Defining agency

With the ‘target list’ of natural and artificial agents in mind, and focusing on their common characteristics, I can now propose a working definition of agency. At its core, agency is the ability to do something (as opposed to the inability to do anything); agency is the ability to act (as opposed to utter passivity). And that action has certain qualities (it is not just any odd action): it serves some purpose, it has a direction towards some utility. So here is that definition, wrapped up in one short phrase:

Agency is the ability to effect non-random change in a given environment or situation.

As this is fairly dense, a little unpacking will help. But keep in mind the main tenets this definition is intended have: (a) in order to be inclusive of all types of agents, it refrains from any anthropomorphic concepts; (b) it is non-recursive, therefore it avoids any reference to the agent itself, be that explicit or implicit. Against that ambitious backdrop, let us walk through step by step.

“…the ability…” — Agency is a capability, either innate (in a natural agent) or given (to an artificial agent). It cannot be claimed by a ‘would-be’ agent as a self-empowering choice; it cannot be self-declared; it cannot be self-taught either. Such ‘self-elevation’ to agency would require having agency already — an obvious paradox. This realization is particularly relevant for artificial agents, because it affirms humanity’s responsibility for the agency any of those human-made artefacts will have. Artificial agency is an endowment that we give: either deliberately or unintentionally, be it as a conscious choice, out of convenient negligence, or through the accumulation of previous choices and decisions, which collectively set the conditions for agency to emerge.

“…to effect non-random change…” — It would be too tempting to claim simplistically that agency was the ability to “achieve an objective” or “influence a situation in a desirable way”. But the very ideas of ‘objectives’ and ‘desires’ are anthropomorphic and therefore unsuitable for an inclusive definition. Instead, we need to detach our thinking from our human reality and develop a fully abstract perspective of agency. That puzzle has three pieces.

  • “…change…” — Let us first consider the outcome of agency: What does it achieve? How does agency manifest itself? In the end, after agency “has been applied”, something is different than before, something has changed. Whether that change is good or bad is a question of perspective: while it will be good for the agent, it may be bad for (parts of) the environment. Just consider predator-prey relationships and their life-or-death consequences, and focus on the prey’s agency: if the prey escapes the attack, the predator goes hungry; but if the prey does not succeed, its death feeds the predator. — This is the first piece of the puzzle: Agency “somehow produces” a different outcome. The immediate follow-up question about the how leads straight to the second piece.
  • “…to effect…” — Let us then focus on the action verb in the definition: What does the agent actually do? I considered many different words, including “to cause” (which sounded too assertive, too much in control), “to initiate” (too weak), “to trigger” (too mechanistic), “to direct” (too authoritative), “to instigate” (getting closer), before settling on “to effect”. This verb describes any kind of possible action that might lead to change: either immediately and directly; or over time, indirectly, and through a cascade of events. Very strong agents may command the means and muster the authority to implement that change immediately and to its full extent (and I understand that as being included in “to effect”). However, less powerful agents (i.e., the majority of the natural and the artificial agents) can at best launch the beginning of a sequence of actions and activities that extend far beyond their own authority and control (think about swarms, symbioses, or the hierarchical organizations in our bureaucracies and military forces). — This is the second piece of the puzzle: Agency effects change “of a certain quality”. But how can we describe that quality? That question gets us to the final piece of the puzzle.
  • “…non-random…” — Apart from the output and the action, our definition of agency needs a qualifier. Agency is not about any kind of change; it’s about change that is useful or advantageous to the agent. Yet again, ‘utility’ and ‘advantage’ are anthromorphic concepts that can misguide us easily, and they both fall into the recursion trap that we need to avoid. I played with the idea of ‘directed’ change, but even that is too closely affiliated with the agent itself. Eventually, rather than defining the qualities the action has, I took the inverse approach and sought to characterise what quality the action does not have: the action is not subject to chance or happenstance, the action is not random. An observer will not be able to discern the presence of agency from a single action, but a number of actions should hint at “something that is present to exert a statistically relevant influence on the outcome”. This is the third and final piece of the puzzle: Agency effects non-random change.

“…in a given environment or situation.” — Agency can only exist within a given context, which frames the problem space as well as the solution space. What might seem like an afterthought is actually vital to the very concept of agency.

  • No agent exists in isolation. Any agent is always situated within “something bigger than itself”. That “something”, that context is dynamic, i.e., it may change over time. And those external changes —presenting either threats or opportunities— stir the agent into action.
  • For natural agents, that context is their physical, tangible environment. For humans, it is the tangible environment plus the —rather abstract— intangible situation (think about all the intricate organizational structures and processes we surround ourselves with). And that logic of the intangible situation applies to the artificial agents as well, given that most of them exist as highly abstracted concepts.
  • For the natural agents, there is even an additional way of demonstrating agency: many of them have some level of mobility, which allows them to leave one setting and move to another. That’s like “changing the environment”.

I can easily imagine that the very nature of this inclusive, non-recursive definition will leave some readers wondering whether something important might be missing. The refusal to address the agent’s purpose, mission, interest, desire or utility, as much as the insistence on non-random change, both seem to evade a clear answer. Yet this definition, in all its stubbornness, provides exactly the necessary basis for a meaningful characterisation that unifies all the natural and artificial agents discussed earlier.

With its focus on “what it is not“, this definition of agency frames the space in which the agents can exist.

This is the eighth in a series of posts on the agency and how it matters to innovation.

What's your view?

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.