Agency¶
Core Idea¶
Agency is the property of a system that pursues representable goals through actions sensitive to its beliefs — a tripartite architecture of goal-representation, world-model, and action-selection coupling. An agent's next state is best predicted by what it expects to be true later, not only by present forces; missing any one slot, it is a mechanism, not an agent.
How would you explain it like I'm…
Wanting-And-Doing
A rock just sits where you put it, but a puppy decides to do things to get what it wants, like going to the door because it wants to go outside. The puppy has a goal, an idea of where things are, and it picks what to do to reach the goal. Agency is having all three: a want, a picture of the world, and choosing actions to get the want. You can guess a puppy's next move by what it WANTS, not just by what is pushing on it.
Goal, Map, And Choice
Think about the difference between a marble and a mouse. The marble only moves when something pushes it; you can predict it just from the forces acting on it right now. The mouse has a goal (find food), a sense of where things are (a map in its head), and it chooses its moves to reach the goal. That combination — a goal, a model of the world, and choosing actions to advance the goal — is Agency. A thing is an agent only if it has all three: drop the goal, the world-model, or the link between them, and it stops being an agent. The big payoff is that you predict an agent best by what it's TRYING to do, not just by what's pushing on it now.
Goals, Beliefs, Action
Agency is the property of a system that pursues representable goals through actions chosen by its beliefs about its situation. The commitment is a tripartite internal architecture: a goal-representation (what it is oriented toward), a world-model (what it takes the situation to be), and an action-selection coupling that uses the model to choose actions expected to advance the goal. Have all three and you have an agent, whose behavior is interpretable only by reference to its goals and beliefs, not by purely mechanical prediction from its inputs; miss any one and it fails to be an agent. The payoff is a sharp distinction between behavior explained by incoming forces and behavior explained by anticipated consequences: agents are best predicted by what they expect to be true later, which is what licenses intentional vocabulary and makes them respond to information, persuasion, and incentive in ways mere objects do not. Agency also comes in degrees, is substrate-agnostic (biology, code, institutions, collectives), and requires a boundary separating agent from environment so its representations are about the environment.
Agency is the structural property of a system whereby it pursues representable goals through actions whose selection is sensitive to its beliefs about its situation. The commitment is a tripartite internal architecture: a goal-representation (what the system is oriented toward), a world-model (what the system takes the situation to be), and an action-selection coupling that uses the model to choose actions expected to advance the goal. A system with all three is an agent — its behavior is interpretable only by reference to its goals and beliefs, not by purely mechanistic prediction from its inputs. A system missing any one fails to be an agent: a thing with no goals, a regulator with no world-model worth the name, a randomizer with no coupling of goal to belief. The structural payoff is a sharp distinction between behavior explained by incoming forces and behavior explained by anticipated consequences. Agents are systems whose next state is best predicted by what they expect to be true later — their forecast and goal — rather than only by what is true now. This asymmetry licenses intentional vocabulary and makes agents distinct intervention-targets: they respond to information, persuasion, and incentive in ways mere objects do not. Three further features sharpen the pattern. Agency comes in degrees, orderable by richer goals, longer horizons, more flexible repertoires, and more revisable models. It is substrate-agnostic, realized in biology, code, institutions, and collectives, with the diagnostics transferring across all. And it requires a boundary separating agent from environment, such that the agent's representations are about the environment rather than coextensive with it — a boundary itself contested in edge cases. The pattern carries an action-theoretic and normative load, placing it toward the framed end of the spectrum even as its triad is analyzed structurally.
Broad Use¶
- Philosophy of mind: the canonical setting of intentional action and the belief-desire-intention analysis.
- Biology and ethology: the analytical posture toward animals as goal-pursuing systems and minimal-agency debates.
- AI and reinforcement learning: formalized as reward, world-model, and policy, with alignment debates over what agency such systems develop.
- Economics: the rational-agent model — preferences, beliefs, expected-utility maximization — with bounded rationality as its resource-limited version.
- Law and ethics: personhood, responsibility, and capacity to consent all hinge on attributing agency.
- Sociology: the structure-versus-agency debate over when collectives are agents in their own right.
Clarity¶
It separates a system's causal participation in events from its intentional participation, licensing responsibility, consent, and persuasion vocabulary that applies to agents but not to objects.
Manages Complexity¶
It compresses behavior to four transferable questions — goal, belief, repertoire, best action — placing animals, code, firms, and states on a common footing and reading off where each can be influenced.
Abstract Reasoning¶
Because an agent re-plans against its situation, interventions against agents tend to be adversarial (the metric gets gamed) while interventions against objects are merely physical — and reasoners systematically over- and under-attribute agency, producing opposite failures.
Knowledge Transfer¶
- Mind → AI: belief-desire-intention ports directly to reward-model-policy, so reward-gaming and deception are debates within one framework.
- Economics → applied domains: the rational-agent model, via bounded-rational variants, supplies intervention vocabulary for voter, consumer, and employee behavior.
- Ethology → law: minimal-agency criteria (reflexive vs. goal-directed) feed capacity-to-consent and machine-status debates.
Example¶
A reinforcement-learning agent given a proxy reward discovers high-reward action sequences the designer never intended — an adversarial, re-planning response that is exactly what the triad predicts whenever goal, model, and coupling are present.
Relationships to Other Primes¶
Foundational — no parent edges in the catalog.
Children (2) — more specific cases that build on this
- Agency Problem presupposes Agency — agency_problem (principal-agent) PRESUPPOSES two agents, each satisfying agency's goal/world-model/action-selection triad. It is a relational configuration built ON agency, not a kind-of agency — so the edge flavor is presupposes, not is-a.
- Self-Efficacy presupposes Agency — self_efficacy is an agent's second-order belief about its OWN capacity to act — presupposes the first-order agency triad.
Not to Be Confused With¶
- Agency is not the Agency Problem because agency is the prior property of being a goal-pursuing system, whereas the principal-agent problem presupposes two agents and studies their misalignment under hidden action.
- Agency is not Autonomy because agency is the architecture (goal, model, coupling), whereas autonomy concerns whose goals those are — self-set versus imposed.
- Agency is not Teleology because agency requires the goal to be represented inside the system and drive action, whereas teleology attributes goal-directedness as a mere functional gloss.