by Herbert Harris
In 2023, Stanford neuroscientist Robert Sapolsky’s bestseller ‘Determined’ declared free will to be a complete illusion. Sapolsky gathered a wealth of data from neuroscience to quantum mechanics in an effort to deliver a final knockout blow to our intuitive ideas of freedom. He then explores a wide range of ethical and social issues where our questionable notions of freedom have led to misguided and often inhumane policies and practices. Since its publication, the book continues to attract criticism for its deterministic stance. Experts from many fields have engaged in this lively discussion.
Watching the debate unfold shows that while we have strong ideas about what free will isn’t, we lack a clear understanding of what freedom actually could be. We agree that whatever it is, it would be incompatible with both mechanical determinism and total randomness. We also think that free will is connected to that vaguely self-conscious feeling that we are the originators of our actions. If free will exists, it probably exists in a middle ground that isn’t too deterministic nor too random. But what exactly is it?
Neuroscience may not be as conclusive about the end of free will as Sapolsky suggests, but it has not been particularly effective in producing alternative explanations. Science generally depends on deterministic approaches — such as reproducible experiments — to test hypotheses that can be proven false. It might be, as critics like Jessica Riskin argue, that science is the wrong place to seek an understanding of free will. However, a potential way forward could come from an unexpected source. In the early nineteenth century, philosopher G.W.F. Hegel developed a theory of freedom, defining it as a result of human social interactions.
Hegel famously argued that freedom isn’t innate but develops through social interaction. He starts with a depiction of humans caught in a Hobbesian struggle for power and survival, a dynamic he describes in his dialectic of the enslaver and the enslaved. Over time, and through often painful mutual engagement, individuals begin to see each other not just as obstacles or tools but as subjects with their own perspectives and intentions. From this social process, a new kind of self-awareness and agency emerges. Hegel believed that self-consciousness and freedom are interconnected and develop together. Without self-consciousness, freedom cannot exist, because we’d lack the reflective awareness needed to own our actions. Without freedom, self-consciousness remains incomplete, as we’d be trapped by external constraints or others’ perspectives.
Aspects of Hegel’s theory of the social nature of freedom might surprisingly align with modern neuroscience. In a previous post, I explained how Hegel’s idea of self-consciousness as a result of social interaction could be “naturalized” through neuroscience’s concept of active inference. Here, I outline how active inference could contribute to the emergence of freedom.
Based on ideas developed by neuroscientist Karl Friston and colleagues, active inference explains how brains don’t just react to sensory input; they actively predict it. When reality differs from those predictions, we either update our internal models or act to change the world to match our expectations. These predictions are constantly revised and integrated into dynamic internal models that represent not only our experiences but also the people, contexts, and goals we encounter. These models include representations of desired outcomes or future states, and it is the difference between these desired states and the current reality that drives action. In this framework, agency is not a reaction to the world, but a generative process aimed at minimizing surprise and achieving expected outcomes.
Active inference systems that initiate actions are similar to thermostats controlling room temperature. When the thermometer shows that the temperature is too low, the thermostat turns on the heat until the desired temperature is reached. This simple type of agency typically involves closed-loop control systems with fixed goals, like regulating blood pressure. It is termed first-order agency because it uses models that directly predict and control the system it manages. Second-order agency involves self-referential modeling of modeling processes (making models of models). In social interactions, this can mean several people modeling each other’s behavior and modeling themselves from the perspectives of others. The goals of interpersonal second-order agency are to influence how we are perceived and treated by others. We use feedback from others to update the predictive models we create of ourselves. Second-order agency is mainly mediated through language and behaviors that shape how others model us, leading us to develop representations of ourselves as seen through their eyes. Throughout our lives, other people act like psychological mirrors, providing us with information about ourselves that we can’t obtain through first-order modeling.
The following social interaction illustrates second-order agency. A person is at a job interview and wants to appear as a competent, ideal candidate. Their verbal and behavioral actions stem from second-order models they construct of themselves, influenced by ideals of what a successful applicant should look like to the interviewer. The aim is to shape the interviewer’s perception of the applicant. The interviewer’s responses reveal how her view of the applicant changes based on his actions. Her words, tone, and affect may show skepticism, enthusiasm, or boredom. The applicant perceives this feedback and incorporates it into his second-order self-modeling. Discrepancies between the applicant’s self-view, as reflected in the interviewer’s behavior, and the applicant’s predicted self (getting the job) generate error signals that drive his next actions.
The process repeats itself. The applicant models the interviewer, the interviewer models the applicant, and this cycle continues. Where does freedom fit into this? The circuitry of active inference is intricate, but there is no ghost in this machine. The social aspect of the interaction adds significant complexity. Having two minds interacting is a major step toward unpredictability, but an even more crucial factor is the social embeddedness of the interview. Social and cultural contexts introduce intentionality, meanings, and beliefs into the process. From the first handshake to the final moments, job interviews are filled with symbols and rituals that neither participant may fully understand. This web of hidden variables makes social interactions very hard to analyze or predict. Many opponents of Sapolsky-style determinism might accept this “practical indeterminacy” that influences human interactions. However, there is one more factor to consider.
The distinction between first- and second-order agency is important because it relates to similarities with formal languages like propositional logic. A first-order language usually includes a set of objects, some qualities and relationships that can be applied to those objects, and logical rules for building statements about the objects. Imagine a world filled only with balls, baseballs, footballs, soccer balls, and a set of procedures you can perform on them: throwing, catching, kicking. In a first-order language, you can discuss how these procedures directly apply to objects. You can say, “I can throw a baseball,” or “You can kick a soccer ball.” Each sentence describes an action applied to a specific thing.
In a second-order language, you gain the ability to talk about the procedures themselves. You can combine them, compare them, and generalize across them. Now you can say, “It’s harder to catch than to throw,” or “Everything you can do with a football, you can do with a baseball.” You can even define complex conditions: “In soccer, the only player who can always catch and kick the ball is the goalie.” Most importantly, you can talk about yourself. As in formal logic or computer science, introducing self-reference creates qualitatively new capacities. In human cognition, it allows us to reflect on our thoughts, to imagine others reflecting on us. This is the foundation of self-consciousness, but it is also the source of freedom.
Second-order agency opens up possibilities for creative, generative behaviors that are not limited to pre-set responses. The self-reference enabled by social feedback loops gives us the perspective to go beyond fixed goals and consider larger purposes. This model of socially embedded agency offers a naturalistic alternative to both metaphysical free will and strict determinism. It also has ethical implications. If agency is not confined to isolated brains but exists in relationships, then moral responsibility is also relational.
The process of mutual recursive self-modeling is a distant descendant of what Hegel called recognition. His famous dialectic of the enslaver and the enslaved describes how second-order capacities like freedom and self-consciousness can develop through social interactions from a primary struggle for power. Hegel viewed freedom and self-consciousness as historically arising rather than evolving, but they are fundamentally social in origin.
***
Enjoying the content on 3QD? Help keep us going by donating now.