Abstract

Social theory has identified agency with social change and dynamism, bringing tension and possibility to a world where social structures are reproduced. The concept of agency can rescue us from the notion that we are simply the product of our conditioning (zombies of embodied habits), and stands in opposition to ontologies that foreground practices at the expense of subjects. While a humanist conception links agency to purposive action, an expansive (post-humanist) definition elides the question of intentionality, and links agency with action, irrespective of purpose. According to this view, rather than being an exclusive human property, agency is all around us, and society has always consisted of relations between human and non-human actors. We should keep in mind that agency is not absolute or independent, but contextual and relational. If we conceive of agency in this way, we can see the stakes of some of the current debates about AI: to what extent will these systems act as agents of change in our world, and how will AI affect (enhance, extend, supplant, or constrain) human agency?

***

Agency to Change the World

How did Western intellectuals go from believing agency to be the exclusive property of the human subject, to considering whether algorithmic agents, or AI systems, also have agency? One understanding is that as AI increasingly approximates human intelligence, it attains attributes formerly reserved for humanity. But an argument can be made that AI today, even in its narrowest forms, already exercises agency, and that humans were never particularly special to begin with.

It is commonly said that people exercise agency to achieve their desires, goals, and interests. In sociology, agency has long been seen as the source of change in society. Agency is why society does not remain in a steady state, despite all the ways that social structures are reproduced. Without agency, we would all be pawns shaped and manipulated by larger forces that often precede our existence: children molded into reproductions of their parents; compliant, orderly workers reproduced by the educational system to passively accept ideologies that justify why the existing order is natural, desirable, or worthy of being preserved. Agency refers to our ability to change this social structure, to disagree with our parents, use education to advance knowledge, achieve social mobility, critique ideology, and challenge government.

There is a longstanding debate in social theory about the relationship between agency and social structure,[1] which has largely gone stale and unresolved. But agency continues to provide the tension that prevents a totalizing view of structure. Not everyone agrees that agency is required to understand humanity or the relationship between individuals and society, but social theories that do without agency, or that provide an impoverished view of agency, paint a deterministic picture. Individuals are conceived not as subjects, but through their habits and practices, or as the effects of the social structures that produce them. Without agency, we are zombies, automata, or cultural dupes.[2]

Amidst some of the debates about agency and structure in the 1980s and 1990s, a new conception (often associated with Bruno Latour and Actor-Network Theory)[3] began to take hold. The provocative argument was that agency was not confined to humans, but that society was composed of both human and non-human “actants”.[4] Agency was defined roughly with action, and the ability to affect the world. If a human worker was replaced by an object (even an inanimate one)[5] that could play the same role, then that object similarly exercised agency. Because in many of our interactions with the world, whether in laboratory experiments or farming,[6] humans cannot fully predict or control the outcome, nature also has agency – co-creating the world with us. Questions of intentionality and purposiveness are elided through this focus on action.

Whether in its humanist or post-humanist form, Western theory’s interest in agency has also been subjected to significant critique. The idea of an autonomous human subject is arguably a historical invention – a distinctly Western, masculine, individualistic vision of man. Feminist theorists advanced these arguments decades ago, pointing to the often unacknowledged work (disproportionately performed by women) of nurturing and caring for ‘autonomous’ subjects. Complicating but not necessarily rejecting the ideal of autonomy, these authors advanced a concept of agency that situates it firmly within social relationships.[7] Relational and non-human conceptions of agency are common in Indigenous and non-Western ontologies,[8]rooted in the understanding that the world is agentially alive, and that humanity is inexorably linked to and dependent on these forces. In an article section titled Columbus Discovers Non-Human Agency,[9] three authors influenced by Indigenous feminist literature point to the Eurocentric and settler colonial bias of a recent turn in social theory. In this ‘new materialism’, authors influenced by Latour and feminist STS have made expansive claims about agency that may be innovative for social theory, but which are quite traditional for unacknowledged indigenous ontologies.

At this point it is worth reflecting on these divergent conceptions of agency. Along one dimension outlined above, they run the range from treating agency as a distinctly human property, linked to subjectivity, consciousness and intentionality, to a broader view of agency as whatever has effects on the world (and ourselves). At its broadest, we are not agents at all: distinctions of subjects and objects are dissolved, and the entire universe becomes a quantum soup of intra-active becoming.[10] But somewhere between this posthuman extreme and the reassurance of conventional humanism, we can return to a view of agency that encompasses both humans and AI, as agents that change the world, and are entwined in relations with one another. 

Today, developers are building robots that learn about and interact with their environment – an environment that includes other robots as well as humans. Machine learning enables AI systems to pursue goals in ways that humans could not anticipate, even if their goals were initially formulated by humans. We now regularly interact with various kinds of AI, or are subject to decisions made by these systems. Finally, the distinctiveness or exceptionality of the human subject has been repeatedly problematized by advances in AI and in our understanding of other organisms. In this context, conceiving of agency as the ability to change the world remains valuable for considering issues common to humanity and AI.

Conceptualized as a means of social change, we can see that agency is not a human birthright, and is not equally distributed across humanity. Structured inequality provides opportunities to some, which are denied to others. Where a person is born, and how they are nurtured or socialized, has great consequences for the choices and capacities available to them – including the impact a person can have on reshaping pre-existing structures. Agency depends on our relationship to these structures, as well as to each other. Hence, agency varies across positions in society and is subject to change. We can engineer technologies and social systems to enhance human agency, to provide capabilities for transformation of individual or collective conditions; or we can design to preserve and reinforce existing power structures. Similarly, it is valuable to conceptualize the agency of AI through its ability to affect the world, change itself, and change human lives, irrespective of consciousness or intentionality. If we conceive of agency in this way, we can see the stakes of some of the current debates about AI: to what extent will these systems be agents of change in our world, and how will AI affect human agency? What decisions will AI make on behalf of humans, and how will these sociotechnical systems reconfigure the possibilities available to us?

[1] Margaret S. Archer, “Morphogenesis versus Structuration: On Combining Structure and Action,” The British Journal of Sociology 61, no. s1 (2010): 225–52, https://doi.org/10.1111/j.1468-4446.2009.01245.x.

[2] Douglas V. Porpora, Reconstructing Sociology: The Critical Realist Approach (Cambridge: Cambridge University Press, 2016).

[3] Bruno Latour, “On Actor-Network Theory: A Few Clarifications,” Soziale Welt 47, no. 4 (January 1, 1996): 369–81, https://doi.org/10.2307/40878163.

[4] Edwin Sayes, “Actor–Network Theory and Methodology: Just What Does It Mean to Say That Nonhumans Have Agency?,” Social Studies of Science 44, no. 1 (2014): 134–49.

[5] Bruno Latour, “Where Are the Missing Masses? The Sociology of a Few Mundane Artifacts,” in Shaping Technology/Building Society: Studies in Sociotechnical Change, ed. Wiebe E. Bijker and John Law (Cambridge, MA: MIT Press, 1992), 225–58.

[6] Michel Callon, “Some Elements of a Sociology of Translation: Domestication of the Scallops and the Fishermen of St Brieuc Bay,” The Sociological Review 32 (1984): 196–233, https://doi.org/10.1111/j.1467-954X.1984.tb00113.x.

[7] Catriona Mackenzie and Natalie Stoljar, eds., Relational Autonomy: Feminist Perspectives on Automony, Agency, and the Social Self (Oxford: Oxford University Press, 2000), http://ebookcentral.proquest.com/lib/ubc/detail.action?docID=430598.

[8] Jason Edward Lewis et al., “Making Kin with the Machines,” Journal of Design and Science, July 16, 2018, https://doi.org/10.21428/bfafd97b.

[9] Jerry Lee Rosiek, Jimmy Snyder, and Scott L. Pratt, “The New Materialisms and Indigenous Theories of Non-Human Agency: Making the Case for Respectful Anti-Colonial Engagement,” Qualitative Inquiry, February 27, 2019, 1077800419830135, https://doi.org/10.1177/1077800419830135.

[10] In Barad’s “agential realism”, drawing on Latour and quantum theory, “agency is not an attribute but the ongoing reconfigurings of the world. The universe is agential intra-activity in its becoming”. Karen Barad, Meeting the Universe Halfway: Quantum Physics and the Entanglement of Matter and Meaning (Durham: Duke University Press, 2007), 141.