In the late 1950s, biologist David Rogers of Vanderbilt University recorded a movie using an optical microscope in which a human immune cell called a neutrophil crawls amid red blood cells in pursuit of a single Staphyolococcus aureum bacterium. After chasing the zigzagging bacterium for some time, the relentless neutrophil catches its prey, engulfs and devours it.
Agency in action? A neutrophil (large amorphous cell) “chases” a bacterium (small dark dumbbell) amongst red blood cells (circular cells). Courtesy of the estate of David Rogers, Vanderbilt University.
This, at least, is how observers typically interpret what they are seeing. The immune cell is presumably sensing and following a chemical trail of some kind exuded from the bacterium, but it is nigh impossible to watch the movie and not frame it mentally with the narrative of a predator and its prey, each trying to out-maneuver the other. The movie seems to validate what Austrian biologist Karl Ludwig von Bertalanffy (a founder of the discipline of general systems biology, which drew on ideas from thermodynamics and cybernetics) said in 1969: “you cannot even think of an organism… without taking into account what variously and rather loosely is called adaptiveness, purposiveness, goal-seeking and the like.” (Bertalanffy 1969).
Should we even try to do so? The narrative the mind imposes on Rogers’ movie feels dangerously anthropomorphic, seeming to attribute to mere cells the kinds of capacities and behavioural drivers we might normally associate with the world of large and cognitively complex animals. But it is not obvious why, if we are prepared to recognize purpose, goals and agency in our own behaviour, we should make it inadmissible for “simpler” organisms. Perhaps, rather than wondering if we are reading too much purpose and agency into the neutrophil chase, a better question is to ask how agency might manifest both similarly and differently at different scales in time and space.
The predominant tendency in modern biology is, however (and pace Bertalanffy), to deny any need to invoke agency at all – to suppose that cells and bacteria, and perhaps the vast majority of living things, can be regarded as sophisticated machines. As Walsh (2015) says, “Organisms are fundamentally purposive entities, and [yet] biologists have an animadversion to purpose.”
The reasons for this aversion are complex. The “reductionistic turn” of a great deal of modern biology, in which explanations for phenomena are ultimately traced to the properties and interactions of molecules, does not obviously generate any prescription for agential concepts. If the causes of all biological phenomena flow from the bottom up, and one cannot reasonably attribute agency or purpose to molecules, how then can these attributes at higher levels be anything more than illusory? But such a case, once it is so explicitly stated, is easily demolished. Atoms and molecules do not (most would agree) possess consciousness or emotions either, but we do not in consequence feel compelled to dismiss these as real organismal features, relegating them to mere “as if” appearances.
Partly, too, the problem is that, in the absence of any widely used tools or concepts for describing properties such as agency and purposiveness, they are often said to be “emergent” in a somewhat ad hoc and vague manner. Without a theoretical framework for handling agency (indeed, without an agreed definition of what it is), that facility seems bound to appear rather tenuous and otiose.
There is also a common suspicion that talk of goals and purpose render biology teleological, thereby raising notions of design and of some over-arching “plan”. Invoking agency as a real biological phenomenon seems to some to open the door to forms of mysticism, such as the old notion that biological systems possess a vitalistic essence that sets them apart from abiotic systems. Worse, notions of purpose and goals might invite quasi-religious explanations of phenomena in the manner of intelligent design: to impute a kind of cosmic agency.
Traditionally, no role has been assigned to or deemed necessary for agency in what is often regarded as biology’s unifying schema: Darwinian evolution. Yet despite the successes of the gene-centred view of evolution, as encapsulated in the Modern Synthesis of genetics and Darwinism anointed by Julian Huxley in 1942, there remains disagreement about its causal structure or the scope of its explanatory power. Lewontin (1974) argued that
To concentrate only on genetic change, without attempting to relate it to the kinds of physiological, morphogenetic, and behavioural evolution that are manifest in the fossil record and the diversity of extant organisms and communities, is to forget entirely what it is we are trying to explain in the first place.
In the Neodarwinian Modern Synthesis, organismal behaviour and morphology become epiphenomena of the competition among gene variants for replicative success. In this view, there is no compelling reason to recognize agency as anything more than a collection of adaptive behaviours. Yet attempts to describe and explain agency as a real biological phenomenon – akin, say, to immunity, metabolism, or cognition – need not obviously pose a threat to the Modern Synthesis (although some argue that such an account will demand that this picture be modified or extended). They simply to refocus the biological lens so that the primary question becomes how living entities (including neutrophils) do what we see them do. As Lewontin implies, this demands a shift from a gene-centred to an organism-centred view.
In this survey I shall discuss some of the approaches that have been taken to naturalize agency: to accept it as a real phenomenon that can be understood without appeal to extra-physical influences. I shall also examine how notions of biological causation and evolutionary change might look different if agency is admitted as a real property of living things.
What is Agency?
Agency is defined by Webster’s dictionary as “the capacity to act or exert power”, and in robotics and AI research a system that can act in any way in response to environmental stimuli is sometimes considered agential. But in biology, typically something more is demanded. The definition offered by Sultan et al. (2022) is typical: they say biological agency is “the capacity of a system to participate in its own persistence, maintenance and function by regulating its own structures and activities in response to the conditions it encounters.” The several definitions listed by Moreno (2018) are similar, and many mention the goal-directedness of agents and their interactions with their environment. It may be that we should seek no harder than this: that a cluster of overlapping definitions enables a more fruitful and inclusive investigation than a premature attempt to impose rigid boundaries.
In the broadest view, agency might be seen as one of the defining characteristics of living entities. Whereas typical definitions of life tend to invoke capabilities such as metabolism, replication and evolution, the notion of agency describes the ends to which such capabilities are put. Agency frames the living entity as a doing thing.
At the same time, understanding agency must place the focus not on the what of doing, but the how. “Agency is thus not about all of the many and varied things that organisms do—from building anthills to caching nuts—but rather about how they do them”, says Tomasello (2022). “Individuals acting as agents direct and control their own actions.”1
While the existence of biological agency seems intuitive, does the notion truly add anything to biology that is not explained by a reductive, mechanistic account of how its parts interact? What about organisms is not understood that a theory of agency is needed to explain? How would biology look different if it recognized agents as real entities?
To find answers, we first need to be clear about which entities possess types or degrees of agency. Some insist that it obtains only if the agents display deliberate intention, perhaps even consciousness. But here Aristotle’s injunction is surely still worth heeding: “It is absurd to suppose that purpose is not present because we do not observe the agent deliberating.” [Physics II: 8])
The anthropocentric viewpoint that humans, if not uniquely agential, have a special variety of it, has been the traditional one historically (Riskin 2016). Aristotle distinguished humans from other living beings by the fact that we alone possess a rational soul: the capacity to reason. For Descartes, meanwhile, agency as a behaviour distinguished from machine-like stimulus-response (even if that included feelings and emotions) was exhibited only by humans, by virtue of our immortal soul: this supplied the theologically necessary capacity to choose.
Such distinctions are less apparent (if not absent entirely) today, when human behaviour and attributes are considered particular cases of more general features of living things. Many behavioural and cognitive scientists even now regard consciousness as a matter of degree, shared by at least some other metazoan species. In any event, biological agency is rarely considered now to be contingent on conscious intention.
What is less clear is whether life is “agency all the way down”. Might it, for example, demand a nervous system capable of formulating and seeking to attain goals? Even bacteria can be considered decision-making entities (Ben-Jacob et al. 2014) whose behaviour depends not just on external circumstances but on their own internal state, informed by external data gathered through sensory systems. Some consider that this mode of operation be best regarded within the framework of cognition rather than mere mechanism. Agency too might be regarded as a matter of degree, extending in some measure even to the simplest forms of life.
Perhaps, however, it goes no further. Moreno (2018) argues that non-living systems such as cellular automata (computational simulations in which the states of cells on a grid depend on those of their neighbours) and active matter (non-living particles that have some means of propulsion) do not display true agency because the individual component “particles” cannot be considered to possess goals. Viruses, meanwhile, lack the autonomy that characterizes true agency. Whether true agency (as opposed to a simulacrum of it imposed by the designer) can be designed into systems such as artificial intelligence remains to be seen – but to assess the prospects clearly we will need a better understanding of wherein this property resides.
One answer to the question of what a theoretical framework for conceptualizing agency might bring to biology is that it might foster more predictive capability. The distinction between physics and biology is sometimes illustrated via the thought experiment of repeating Galileo’s (almost certainly apocryphal) Tower of Pisa experiment by dropping a cannonball and a pigeon. The trajectory of the cannonball is wholly predicted by the Newtonian laws of motion; the same cannot be said for the pigeon, even though it does not violate any physical laws. To the extent that we can predict what the pigeon will do at all, we implicitly invoke its agency. To explain why it does not simply plummet, it is not enough to invoke aerodynamics; we must also in effect allow that the pigeon does not want to plummet. It manifests its agency by virtue of having goals.
Walsh (2015) distinguishes object theories, which describe the behaviour of objects according to laws external to the system (typically Newton’s laws), and agent theories, in which actions are events “that occur as a consequence of agents’ pursuit of their own purposes” and are internal to the system. Even for a relatively simple and well-studied biological phenomenon such as bacterial chemotaxis – the movement of bacteria in response to a gradient in chemical concentration – the exact function and mechanism of such agential, goal-directed behaviour is not fully understood, nor is it wholly predictable on the basis of the stimuli alone (Neilson et al. 2011; Samanta et al. 2017).
The challenge of identifying and explaining agential behaviour is not a mere academic exercise. If, for example, we try to treat a cancer cell as an object that behaves as it does because of causal laws governed by genetic mutations, then judging from present experience we may not get very far towards finding a cure. If our theory incorporates some notion of the cancer cell as an agent interacting with its environment (including other cancer cells), with goals that involve not just survival and replication but also development, and with all the responsiveness and adaptation of its internal state that this entails, we might do better (Sonnenschein & Soto 2020). As Barbara McClintock expressed it in her 1983 Nobel lecture, “A goal for the future would be to determine the extent of knowledge the cell has of itself and how it utilizes this knowledge in a “thoughtful” manner when challenged.”
Sultan et al. (2021) argue that an understanding of agency would allow the role of the environment to be better incorporated into biology. Currently, environment tends to feature in evolutionary theory as a “given” to which an organism must respond. Of course, it is understood that biological agents may alter their surroundings – but there is no systematic way of theorizing about that process. Agents might deplete resources, but also enrich them, for example through excretion of nutrients. They might restructure the environment in more profound ways, altering local climate or geomorphology. Some of this is incorporated in a rather ad hoc fashion into the evolutionary concept of niche construction or the notion of an “extended phenotype” (Dawkins 1982). But there is no systematic way of predicting or describing such interactions or the principles underpinning them.
For example, an agent intent on self-preservation and maintenance might conceivably respond to environmental change (a rise in temperature or salinity, say) in several ways:
By developing a capacity to buffer its internal states against external fluctuations or shifts.
By activity that restores the previous environmental conditions in a homeostatic manner (Dyke & Weaver 2013).
By migrating to a different environment with more amenable conditions (as in chemotaxis).
Which of these strategies is preferred in a given circumstance? There is no general framework for answering that question.
Walsh (2015) argues that, in the end, recognizing and explaining agency is not just an instrumental desideratum; ultimately it is a part of the scientific quest to understand. “If agency is a real, natural phenomenon, and our scientific theories cannot countenance it, then our understanding of the world is destined to be impoverished”.