Agents and Institutions
This essay is the first in a two-part series exploring institution design in the polycene. It delves into the nature of agents, their roles, and their interplay with institutional landscapes. The insights gained from understanding these agents have far-reaching implications for various fields, including human behavior, decision-making, and institution design. Read Part 2 at Small and Large Worlds.
This essay is based on transcripts from a series of original audio recordings of our conversations with Jon with minor edits for clarity and length.
Defining Agents
In grappling with the complexities of institutional design in the context of the polycrisis, I find myself drawn to the problem of agency and agents. To begin, let’s acknowledge that the question of “agents” and “agency” is complex and much debated. This isn’t about producing a general theory of agency or necessarily resolving the debate about what counts or does not count as an agent (forests, viruses, simians) but rather to tackle this problem from the perspective of some usable archetypes. What I am interested in is trying to capture certain aspects of agency. I’m thinking about this through two general kinds of agents: model-bound agents and unbound agents. These archetypes don’t necessarily appear in pure form out in the wild of the world, but are useful to think through.
Model-Bound Agents: Enabled and Limited by Structure
By model-bound agents, I am referring to entities that act within a system, guided by certain rules, representational structures, or informal norms embodied as instinct and habitus. They can be individuals, groups, or even artificial constructs like algorithms. Model-bound agents are framed by an enabling institutional infrastructure, that is, they operate within defined institutional boundaries and are guided by specific models that have a certain degree, but not unlimited degree, of latitude. My working example is the physician in the clinic.
As the name suggests, model-bound agents are both constrained and enabled by the models they operate within. For instance, we stipulate that doctors should practice medicine based on science and evidence. Yes, there's judgment, discretion, and contextual adaptation, as well as enormous limits to both current science and evidence based on histories of racism, homophobia, sexism, reductive understandings of disease, and so on. The idea of evidence-based medicine has significant shortcomings, but no one wants to return to the days (only a few decades ago) when the most powerful doctor in the room decided on the “correct” treatment by fiat. Of course, there are cases that lie at the borders of current knowledge where medical practice becomes a form of experimentation. All of this being said, we insist on the existence of a border between the practice of medicine—based on particular models and operating within a particular cultural/legal/technological infrastructure—and malpractice for ethical, scientific, and pragmatic reasons.
Unbound Agents: Navigating Complexity
Are there other kinds of agents that can move more freely within and across institutional infrastructures? I believe, yes.
During the first years of the COVID-19 pandemic, I worked at the clinic-level with community health workers and peer recovery specialists, who are an example of what I am calling unbound agents. I largely worked with people who are formerly incarcerated or people in recovery from addiction. They were figuring out in real-time how to access institutional resources as institutions closed down or collapsed around them, and I was able to watch them (and work with them) as they strategized across this landscape. In part, they had the knowledge and literacy to operate in this fashion because they had lived through this institutional landscape as patients and as people who had experienced the violence of the system in multiple ways first-hand. This collective experience of survival has generated a form of systems mapping from below.
These unbound agents, operating at the edge of the medical and treatment systems and, in some ways, against these same institutions, understood how it functioned in real-time and on the ground better than the people who were charged with running them. Executives have paper models, maps of departments, and decision trees. Their sense of the actual complex web of informality that holds these institutions together was limited—in part, because people operating in the institution at lower levels actively dissimulate these networks of informality. People whose survival depended on being able to navigate through these institutions (and the attendant concerns of poverty, racism, and environmental injustice) are tacticians of informality. They move through these spaces in unique ways that incorporate dimensions that elude the models of model-bound agents. Their tactics are also sometimes informed by a community memory surrounding these institutions, often stretching back decades.
Unbound but Not Uniform: A Closer Examination of Subtypes
The term “unbound agents” may suggest a homogenous group, but the reality is more complex. To truly understand the impact and potential of these agents, we need to break them down into more specific categories. Here, we’ll focus on Chaos Agents, Hackers, and Reflexive Agents
Chaos Agents: Manipulating Responses
Chaos agents exploit the frameworks that guide model-bound agents. They understand the rules and constraints within which these agents operate and manipulate them to serve their own ends. For example, Donald Trump realized that by polarizing public opinion and driving the profit-based news cycle, he could control the narrative. The act of making half of the U.S. despise him created a powerful relationship that he could manipulate to his advantage. Chaos agents are masters of emotional and psychological manipulation, they instrumentalize reactivity—the automatic or fixed dimensions of model-based agents—by triggering the activation of models outside the institutional contexts and infrastructures within which they are designed to function. They control you by controlling your reaction.Hackers: Exploiting Infrastructure
Hackers are agents who understand the underlying technical infrastructure of an institution and exploit it to serve their own ends, often undermining the general model the institution aims to uphold. A prime example is the Federalist Society's strategy to pack the Supreme Court with conservative judges. By doing so, they took an institution meant to regulate the conditions of democracy (basically, to ensure that the historically white and wealthy elite which is allowed to play the game of politics plays by the same rules) and turned it into a tool for right-wing, minoritarian governance. Hackers identify loopholes or weak points in the institutional framework and exploit them, often causing the institution to act in ways contrary to its dominant semantics, that is, its intended purpose.Reflexive Agents: Real-Time Adaptation
Reflexive agents are perhaps the most adaptive of all. They reflect on institutional dynamics in real-time and translate these reflections into actionable insights. Community health workers and peer recovery specialists serve as an example here. Another example might be special forces in some military contexts. Community Health Workers react in real time to institutional breakdown or failure, building ad hoc networks and extemporizing temporary paths across the institutional landscape. Unlike model-bound agents, who are limited by their pre-existing frameworks, reflexive agents work case-to-case, troubleshooting singular situations, and start their process after the point that the governing model of the institution has already failed. They sometimes find themselves working to achieve an institution’s stated goals, for example providing health care for poor and underserved communities, against the actual practices of institutions themselves. They make use local or informal knowledge, which are representations of complex feedback loops that the models of model-bound agents fail to capture.Nurturing Institutional Immune Systems
The contrast between these two types of agents leads us to propose a division of labor—perhaps even a fundamental principle of institutional design for the polycrisis—between model-based agents and unbound agents. Model-based agents enact existing frameworks within an institutional infrastructure, while unbound agents traverse institutions through non-linear paths, seeing and responding to what remains invisible to the institution’s own models (No institution is capable of developing a complete, real-time model of itself as an actor in the world—trying to do so results in an infinite regress.) If we could find structures and protocols to coordinate this division of labor, we might be able to harness the strengths of both.
Unfortunately, our current institutions are often deeply hostile to unbound agents, even when they have created the positions that they occupy, viewing them as threats. Yet unbound agents could function as a kind of institutional immune system—much like some companies utilize computer hackers to probe the weaknesses of their security systems. There are some frameworks for integrating model-bound and reflexive agents, like multidisciplinary care teams in medicine. But we need more.
The Fallacy of “Formal” Institutions
As many thinkers have observed, institutions are not merely formal structures with defined rules and regulations. They are complex, adaptive systems that encompass both formal and informal dimensions. The formal models, maps of departments, and decision trees are only part of the story. They are also shaped by hidden processes, externalities, and a complex web of relationships that go beyond these formal models. Understanding this complexity is essential for designing institutions that can adapt to the uncertainties of our world. And we cannot understand this complexity without incorporating these different modalities of agency (hackers, chaos agents, reflexive agents) into our analysis from the beginning.
In the same vein, the relationship between institutions and actors is dynamic and multifaceted. Agents within institutions are not merely following predefined paths. They are navigating a complex landscape, strategizing, and adapting in real-time. They are not only adapting their own strategies to novel situations, they are working to transform the institution itself. Institutions have become battlegrounds for competing models pursued by competing agents—one factor that has led to large-scale institutional paralysis.
The unbound agent’s ability to navigate, critique, troubleshoot, and find ways to improve institutions is an extraordinary resource that institutions could potentially tap into. At the same time, actors are not merely passive recipients of institutional rules; they actively shape and influence the institutions they traverse. This interplay between institutions and actors offers a rich avenue for institutional self-reflexivity. By tapping into the knowledge and insights of actors, especially unbound agents, institutions can learn about their failures through how they are being “hacked” by unbound agents to live up to promises they made but are failing to deliver in practice.
Unbound Agents and Institutional Reflexivity
Finally, a useful idea for me (that comes out of agent-based modeling) is that agents are effective actors in the world to the extent that they have an accurate model of themselves in the world. So a collective property of the kinds of agency discussed here is that they are enabled by the fact that the agent has a representation of themselves existing and acting in their reality. This is especially interesting because it means that, if we expand people’s models, that expands the range of their agency. Conversely, constricting or controlling people's models becomes a form of political control.
The complexity of institutions and actors challenges us to experiment with new principles of design that incorporate several modalities of agency. We should design with model-bound agent, hackers, chaos agent, and reflexive agents in mind. Each of these modalities of agency are generated as possibilities by all but the most simple of institutions or systems.
We must recognize the nonlinear paths, the hidden processes, and the informal dimensions that shape our institutional landscapes. By embracing this complexity, we can design institutions that are more adaptive and resilient. At the same time, we still want institutions to provide stable contexts for action—we want to be treated effectively when we go into a doctors office—and some form of accountability when our models fail. We can harness the strengths of both model-based agents and unbound agents, coordinate their division of labor, and create institutions that may be better equipped to navigate the uncertainties of this urgent moment in our shared history.