Agency and the Changing Architecture of Behavioral Control
At what point can we say that an organism has control over its own behavior?
The conventional approach would be to begin with mind, intelligence, or consciousness. These concepts are useful; all are also deceptively difficult to define, measure, and compare across species.
Behavior is easier to observe, but behavior does not necessarily imply control.
Instincts, chemical signals, reflexes, and relatively fixed programs can generate incredibly complicated activity. Bacteria navigate chemical gradients. Ant colonies construct collective systems no individual ant understands. Animals converge upon prey. Humans build companies, governments, religions, economies, and societies.
Complexity alone does not tell us how any of this behavior is controlled or organized.
In The Evolution of Agency, developmental and comparative psychologist Michael Tomasello suggests another place for psychology to begin. Instead of asking only what an organism knows, thinks, or experiences, he asks a more operational question: How does an organism control its own behavior? Behind that sits an evolutionary question: How did this control become possible?
Tomasello proposes agency as a better organizing principle for psychology than either behavior or mentality alone. Mentality is difficult to access directly. Behavior tells us what an organism does under a set of conditions, but not how perception, motivation, decision-making, and action become organized toward an outcome. Agency concerns the architecture coordinating those processes.
His model proceeds through four broad forms of psychological agency: goal-directed, intentional, rational, and socially normative. These correspond, in his evolutionary reconstruction, primarily to ancient vertebrates, mammals, great apes, and humans. Each form enlarges the range of situations and possible actions available to an organism while introducing a new level of control over that expanding behavioral space.
This is not a complete summary of the book. What follows is my attempt to work through the model, identify what I found useful, and pull apart the questions it left me with.
Agency is the organization of behavior around goals through feedback control. An agent acts, perceives what happened, compares the result against some desired state, and uses that information to adjust what it does next.
There are many forms of regulation beneath this psychological definition.
Imagine a bacterium. Let’s call him Fred.
Fred is attracted to some chemicals and repulsed by others. Receptors and intracellular signaling pathways regulate changes in his movement. He is alive, organized, adaptive, and responsive to changes in his environment. Bacterial chemotaxis can even involve comparisons between current chemical conditions and those encountered moments earlier.
There is control here in a broad biological sense. Fred remains, however, a constrained biochemical system. Evolved mechanisms translate chemical conditions into changes in the probability and direction of movement. This is more complicated than a perfectly linear stimulus-response machine, but the behavior does not by itself demonstrate that Fred separately represents an outcome, compares alternative ways of reaching it, or evaluates whether his action succeeded.
Tomasello begins below agency with organisms he calls animate actors.
C. elegans, a tiny worm whose adult hermaphrodite nervous system contains 302 neurons, can integrate sensory information, navigate chemical gradients, approach nutrients, avoid harmful substances, and modify some behavior through experience. That is useful behavior, impressive behavior, even.
The question is how that behavior is organized.
In Tomasello’s account, psychological agency requires an organism to regulate action in relation to a represented goal and flexibly adjust its means as conditions change. An organism may sense, learn, move, and integrate information without possessing this particular architecture of control.
An organism can be alive, responsive, adaptive, and behaviorally complicated without necessarily possessing psychological agency.
The dividing line concerns the mechanism by which information becomes organized into action.
The first true agents in Tomasello’s reconstruction are ancient vertebrates, represented primarily by reptiles such as lizards. The evolutionary problem driving this transition is unpredictability.
Fixed behavioral programs work remarkably well when the important conditions of life remain stable. When food, danger, and opportunity appear in predictable forms, natural selection can prepare an organism with relatively specific responses. A variable environment presents a different problem. Moving targets change direction. Obstacles appear. Predators enter the scene. The same stimulus may demand approach in one situation and withdrawal in another.
Under these conditions, behavior must remain oriented toward an outcome while adapting to the changing circumstances surrounding it.
Imagine a lizard pursuing a cricket.
The cricket appears within a larger situation. Its distance, movement, accessibility, and relationship to nearby obstacles all matter. The presence of a predator may transform it from an opportunity into a danger.
The lizard’s perceptual system integrates these streams of information in relation to the goal of acquiring food. Through that goal, the environment becomes structured into opportunities, obstacles, and threats.
“Attention is thus a kind of goal-directed perception.”
- Michael Tomasello, The Evolution of Agency, p. 36
Attention selects and organizes information according to its relevance for action. Our cricket is reachable, distant, obstructed, escaping, or too dangerous to pursue. It gains meaning through its relationship to what the lizard is attempting to do: eat without being eaten.
The lizard acts, receives perceptual feedback, and adjusts. It may continue or stop, approach or inhibit, attack or withdraw. A visible cricket can produce different actions because behavior is regulated in relation to the goal and the larger situation in which that goal is being pursued.
A future state now helps organize behavior in the present. The organism does not simply encounter stimuli; it navigates situations.
Reacting to stimuli becomes navigating through situations.
The world no longer sits plainly as something that happens to collide with a sensory system. Attention constructs a usable environment from the information relevant to the organism’s goals, leaving other information outside the active control process.
Mammals add another level of behavioral organization. Tomasello calls this the executive tier.
A goal-directed agent adjusts immediate action in relation to an outcome. An intentional agent can generate or simulate several possible courses of action, compare their expected consequences, select one as an intention, and maintain that intention across time.
Consider a bear attempting to find food.
The bear may travel to a river where fish have previously been available. It can choose among several positions, wait for an appropriate moment, abandon an unsuccessful location, and move elsewhere while maintaining the larger goal.
The output of its decision-making system may therefore be a plan rather than an immediate movement. The intention to acquire food organizes a sequence of actions across time.
The bear may be hungry, tired, afraid of another bear, and aware of a different location where fish were previously available. These conditions produce competing action tendencies: pressures generated by emotional and motivational systems toward behaviors such as approaching, withdrawing, attacking, waiting, or investigating. The executive tier integrates these tendencies with memory and present perception, compares possible plans, and selects a course of action.
Tomasello connects this architecture to the evolution of mammalian life. Mammals tend to have longer periods of development, greater parental dependence, and more opportunity for individual learning than reptiles. The advantage of time and adolescent development grants ability to learn not only which things exist in their environment, but what their own actions do within it.
Relationships between action and result can then be carried into future situations and evaluated before behavior begins. This expands the organism’s behavioral space: the range of actions and plans available under a given set of conditions.
The value of this larger space depends upon the organism’s ability to organize it. Additional possibilities alone would produce confusion or inconsistency. Intentional agency gives those possibilities structure by allowing the organism to compare alternatives, form a plan, and maintain it long enough to produce an outcome.
Functionally, this control system can ask: What can I do? What might happen if I do it? Is that result desirable? Is another course of action available?
My interpretation is that evolution gains an advantage by producing organisms with a wider behavioral space and the internal control needed to navigate it. Instead of relying entirely upon natural selection to specify the appropriate response in advance, the organism can construct a response from memory, perception, motivation, and simulated outcomes.
The organism remains constrained by biology, learning, and circumstance. Within those constraints, however, it can hold several possibilities at once. It can want one thing while fearing another, hesitate between actions, inhibit a plan already underway, maintain an intention despite distraction, or redirect itself when the result diverges from what it expected.
Its internal operation has, to some degree, become available as information for its own control.
The next architecture is represented primarily by great apes and what Tomasello calls the reflective tier.
The intentional agent evaluates possible plans. The rational agent can inspect the basis upon which those plans are being evaluated.
One consequence is the ability to recognize uncertainty.
An intentional agent can select the best action currently available and revise that plan when it begins to fail. A rational agent may recognize, before committing, that the available information is insufficient. The question shifts from What should I do? to Do I know enough to decide?
Imagine an ape attempting to reach honey in a tree. An intentional agent may select climbing as part of a larger plan, feel a branch shift beneath it, and abandon the strategy when the action becomes dangerous.
A rational agent may test the branch before committing its weight.
Testing the branch does not directly obtain the honey. It produces information. The agent intervenes upon its uncertainty before the original plan succeeds or fails.
Rational agency begins when uncertainty itself becomes a condition the organism can recognize and act upon.
I take this as a form of second-order control. The agent evaluates its evaluation of the situation and changes its behavior to improve the information upon which a later decision will depend. It can functionally ask: What information is missing? Can I test this before acting? Why did the previous strategy fail? How can I gather greater certainty?
The world now contains relationships that can be investigated, tested, and manipulated. Tool use becomes especially revealing within this architecture. The agent acts upon a stick; the stick acts upon the environment; the resulting change contributes to the agent’s goal. A tool becomes an intermediary inside a causal chain, allowing behavior to be organized around a state of the world that does not yet exist.
Other agents can also become objects of causal and informational analysis. What can that individual see? What information do they possess? What are they attempting to do? What might they do next?
Great apes already possess important forms of social understanding. Tomasello’s claim is that causal and intentional relationships increasingly become available for representation, investigation, and control. The organism’s experiential world expands to include uncertainty, evidence, hidden causes, possible futures, and the intentions of other agents.
Finally, we come to humanity and a level of agency familiar from our own lives. Some goals cannot be achieved by one individual alone.
Imagine two early humans hunting together. One drives an animal in a particular direction. The other waits to intercept it.
For this activity to work, both participants must represent the shared outcome, their own role, the complementary role of the other person, and the expectation that both will remain committed until the activity is complete.
The activity contains I want food and you want food, but it is organized by something more: We are hunting.
A rational agent can represent something like I and you: my goal, your goal, what I know, and what you know. The socially normative agent adds we, a shared collective intentionality, and, from within that collective, me: myself as the person occupying a particular role for others.
If I stop chasing the animal for no reason, I have selected an ineffective action, disrupted your behavior, and violated a commitment upon which the shared activity depended. The failure becomes social because my role created expectations that organized what another person was willing and able to do.
Tomasello uses comparative work with children and chimpanzees to clarify this distinction. Chimpanzees can cooperate, recruit partners, and behave strategically. Human children, however, more readily treat joint activity as creating commitments, expectations, entitlements, and responsibility. A participant who contributed to the activity may be understood as deserving a share; someone who arrived afterward may not possess the same claim.
Tomasello’s argument is therefore not that chimpanzees lack cooperation. Their cooperation does not appear to be organized through the same mutually understood commitments and normative expectations. What looks like joint activity may instead reflect coordination or shared opportunism.
Humans treat the shared social world as a source of reasons and obligations. We predict one another’s behavior, and, we also judge it.
At first, accountability may exist between particular partners: I expect something from you because we agreed to act together. These expectations can then become generalized. The standard applies because this is how a person occupying this role is supposed to act.
Tomasello calls this normative rationality. Behavior can be evaluated according to whether it is expected, permitted, required, fair, or owed. A successful action may still be improper; a personally costly action may still be required.
The human experiential niche therefore contains promises, responsibilities, permissions, reputations, laws, institutions, identities, and roles. These are not physical objects in the ordinary sense, but they become causally effective because people organize their behavior around them.
A promise can constrain tomorrow before tomorrow exists. A job title can determine what actions are expected from a person. Money can move physical resources despite depending upon collectively maintained systems of representation. A company can direct activity, allocate resources, and constrain its participants despite having no biological body of its own.
Tomasello’s argument concerns the emergence of human normative agency. Thhrough this, iIt becomes possible to examine these shared systems cybernetically. Information moves through roles, decisions are distributed among participants, feedback modifies collective policy, and shared goals constrain individual behavior. In this limited sense, an organization can acquire a structure with causal consequences for the agents inside and outside it, even though every action still occurs through individual people.
Human agency constructs collective systems and then lives within the systems it constructed.
The concept tying much of the book together is Tomasello’s experiential niche.
An ecological niche concerns the environmental relationships through which an organism lives and survives. An experiential niche concerns the kinds of situations an organism can experience and enter.
A worm, a bee, a bat, an octopus, and a human may occupy the same physical environment while inhabiting dramatically different perceptual worlds. Reality contains vastly more information than any organism can detect or use. Some wavelengths, chemicals, movements, sounds, and social signals are available to one nervous system and absent from another.
The experiential niche also depends upon what detected information can become within an organism’s control system. A feature of the environment might become an opportunity or obstacle relative to a goal. An internal state might become one motivation among several competing possibilities. Uncertainty might become a reason to gather evidence. A promise or social role might become part of the situation itself.
These are different forms of experience made possible by different architectures of agency.
The concept reminds me of the Perception Box, a framework developed by Elizabeth R. Koch and introduced to me through Christof Koch of the Allen Institute. The Perception Box describes how incoming information is shaped by neural structure, prior experience, belief, expectation, narrative, and biological state.
I use the analogy somewhat more broadly than the original framework. The Perception Box helps describe how reality becomes available to an organism. The experiential niche describes the situations that can become meaningful and actionable once that information enters a particular architecture of control.
A goal-directed agent experiences opportunities and obstacles. An intentional agent gains a behavioral space containing alternative plans, competing motivations, and possible futures. A rational agent can encounter uncertainty, evidence, causal structure, and the limitations of its own knowledge. A socially normative agent inhabits a world of roles, commitments, permissions, obligations, and institutions.
Agency changes both what an organism can do and what kind of world it can inhabit.
The most useful move in Tomasello’s book is the shift away from cataloguing behavior and toward studying architectures of behavioral control.
Different species may perform radically different actions while relying upon structurally similar forms of organization. Conversely, two species may produce superficially similar behavior through very different systems.
Two predators may converge upon the same prey without jointly representing a shared goal. A bacterium and a lizard may both move toward food without controlling that movement in the same way. A mammal and an ape may both abandon an unsafe plan, while only one recognized and investigated its uncertainty before acting.
Tomasello calls these broad architectures psychological bauplans: organizational body plans for behavior.
Species need not perform the same acts to be playing structurally similar games. Different bodies and environments may produce different surface behavior while the underlying control problem remains something like pursue or withdraw, continue or inhibit, select this plan or another, act or gather more information, or cooperate or defect.
What changes across evolution is not only which decisions are made, but what kinds of decisions the organism is organized to make.
This is where Tomasello’s model connects most strongly to cybernetics. Every form of agency depends upon feedback: action, perception, comparison, adjustment. Evolution changes the kinds of variables available to that loop.
Goal-directed agents compare actions with goals. Intentional agents compare plans through their expected outcomes. Rational agents compare decisions with evidence and intervene when uncertainty remains too high. Socially normative agents compare individual conduct with shared commitments and standards.
Each architecture expands the organism’s behavioral space while adding a new method for regulating that expansion. Goals organize immediate actions. Intentions organize alternative plans. Reflection organizes uncertainty and information gathering. Norms organize individuals within shared activities.
Agency, in this account, evolves through the widening and successive organization of possible behavior.
Evolution moves through layered iteration, bounded by an organism’s history. New forms of agency are built above, around, and through older systems of control.
The executive tier reorganizes goal-directed action. The reflective tier monitors and revises executive decision-making. Socially normative systems regulate the decisions and actions of rational individuals.
Tomasello calls this hierarchical modularity. He also proposes trickle-down selection: when the success of a higher-level architecture depends upon how its lower-level components operate, selection acting upon the whole may gradually reorganize those components.
Humans still possess reflexes, conditioned associations, immediate desires, competing motivations, longer-term intentions, reflective judgments, and social obligations. These systems can cooperate, but they can also pull behavior in different directions.
I can understand my social responsibility while remaining hungry, tired, angry, or resentful. I can form a rational plan and fail to carry it out. I can recognize what I should do while another system organizes my behavior toward immediate relief.
Higher levels of control regulate the systems beneath them imperfectly. They create new possibilities without eliminating older pressures.
Human agency is a layered negotiation among systems operating at different scales of time and control.
This helps explain why possessing rational or socially normative agency does not guarantee rational or moral conduct. A higher tier makes a form of regulation possible; it does not ensure victory over every competing process.
If conscious experience integrates these systems of control, even something as ordinary as feeling hangry may reflect a shift in their relative influence. Immediate biological regulation becomes more behaviorally powerful, while the socially organized intention to remain patient loses control of the situation.
Hunger does not become a separate agent in any literal sense. The example instead suggests that what we experience as a unified self contains several forms of regulation competing to determine what the organism does next.
The larger evolutionary idea I take from the book is that uncertain environments favor organisms capable of constructing behavior during their own lives.
Fixed programs are effective when the environment repeats itself. Natural selection, however, cannot specify an answer for every possible circumstance. As unpredictability increases, there is value in widening the organism’s behavioral space, provided that the organism also develops the control required to navigate it.
This connects to W. Ross Ashby’s cybernetic principle of requisite variety: a control system must possess enough internal variety to answer the variety of conditions it encounters.
Behavioral variety without control would produce noise. Agency makes variety useful by organizing possibilities around goals, intentions, evidence, and commitments.
My description of this evolutionary trajectory is the death of specificity in favor of intentionality.
This does not mean the disappearance of instinct, reflex, or biological constraint. It means that biology becomes less responsible for specifying every behavioral answer in advance. Evolution increasingly constructs organisms able to pursue outcomes, compare possibilities, investigate uncertainty, and commit themselves to shared activity.
The same process eventually allows humans to create new experiential environments.
We build languages, companies, economies, governments, games, disciplines, laws, and identities. These systems depend upon representation and agreement, but they become causally real through the behavior organized around them.
Microsoft is a made-up word. A corporation is not an object found in nature. Yet corporations move resources, direct activity, employ people, create technologies, alter environments, and constrain the behavior of their participants.
Socially shared representations do not need to be physically objective to become causal.
The book is most useful when treated as a model of possible architectures rather than as a perfectly ordered ladder of species. I do not think the capacities associated with these forms of agency will always divide cleanly.
Different nervous systems may produce similar functional abilities through different evolutionary histories. Intentional, rational, and social capacities may occur in uneven combinations. Species such as crows, dolphins, and octopuses would likely complicate a simple progression from reptile to mammal to ape to human.
Tomasello intentionally focuses on the lineage leading toward humans. That choice makes the argument clearer, but it limits what the typology can establish about agency across life more generally.
I would therefore treat the four architectures as recurring organizational solutions, and perhaps as dimensions of control. rather than mutually exclusive boxes into which every species can be placed perfectly.
The distinction between stimulus-driven behavior and goal-directed control is conceptually meaningful. Empirically, the boundary may be difficult to locate.
We can observe whether an organism varies its behavior, persists after failure, takes a detour, transfers a learned strategy, tests a situation, or adapts its means while maintaining an apparent outcome. These behaviors provide evidence, but the internal organization producing them must still be inferred.
How much flexibility is enough to establish a represented goal? When does adaptive regulation become psychological agency? Does goal representation appear as a discrete capacity, or gradually across systems of increasing complexity?
The model provides a useful vocabulary for asking these questions. It does not make them easy to answer.
Agency does not solve consciousness.
An executive system integrates perception, memory, motivation, possible action, and expected outcomes. A reflective system makes aspects of its own decision-making available for monitoring. Increasingly agentive systems therefore seem to require broader integration of information, and perhaps a more unified space within which that information can shape behavior.
My own interests pull toward theories that attempt to explain such integration. Global neuronal workspace theory might frame consciousness in terms of distributed information becoming globally available. Integrated information theory might connect experience to the causal integration of the system itself.
Neither connection resolves the central problem.
The integration required for advanced agency may produce consciousness, require consciousness, or merely correlate with it. Tomasello’s account remains valuable because it allows us to study increasingly sophisticated forms of control without pretending that consciousness has already been explained.
Tomasello argues that the agent is a necessary level of psychological explanation, not a tiny person inside the brain.
A cell is composed of molecules, but a cellular explanation remains useful. An organism is composed of cells, but the organism remains a causally meaningful whole. An agent can likewise emerge from neural, biological, developmental, and social processes without becoming a fictional executive pulling levers behind the eyes.
Still, I am left with the phenomenological question: What is the relationship between this organized whole and the felt center of experience?
The agent may be an emergent causal organization. The self may also be a model constructed by that organization. These ideas can coexist.
The agent may be an illusion in one sense while remaining a causally useful organization produced by an organism controlling itself through time.
At the end of the book, Tomasello proposes that agency, not behavior or mentality, may be the proper first principle of psychology.
The proposal works because agency connects the mental and behavioral without reducing one to the other. Behavior tells us what an organism does. Mentality concerns the processes through which it perceives, remembers, evaluates, and decides. Agency concerns how those processes become organized into control.
The goal-directed agent regulates action toward an outcome. The intentional agent expands and navigates a space of possible plans. The rational agent monitors its own decision-making, recognizes uncertainty, and gathers information. The socially normative agent regulates behavior through shared goals, roles, commitments, and standards.
Each architecture increases what the organism can experience and control. The world gains opportunities, obstacles, alternative futures, hidden causes, uncertainties, intentions, responsibilities, and shared institutions.
Agency is not freedom from causality. It is the organization of causality through an organism capable of controlling what it does next within a causal world, with greater agency resulting in greater causality.