Grounding Affect in Agentivity: Manipulating Sense of Agency and Decision-Making to Identify Behavioral and Intracranial Neurocomputational Signatures of Depression

Major depressive disorder is widely recognized clinically, yet its explanation remains elusive. Its heterogeneity, variable treatment response, and overlap across affective, cognitive, motivational, and self-referential symptoms have made it difficult to localize within any single domain. This dissertation examines one possible reason for that difficulty: affect, agency, and decision-making are typically investigated in relative isolation. Across theoretical, behavioral, and intracranial computational levels of analysis, this dissertation asks whether these domains can be related more systematically through agentivity: the ways in which one is oriented as an agent toward the world. To this end, I develop a multi-level account of affect grounded in agentivity, with depression as the main clinical case. Over the last decades, the interdisciplinary study of affect has seen the proliferation rather than integration of theoretical perspectives thanks to differences in metaphysical and mechanistic assumptions about what affective phenomena are and how they work. Chapter 2 addresses the problem of affect itself by developing the Human Affectome framework in collaboration with over 170 affective researchers. Here, an assumption about the purpose of affect is leveraged as a teleological principle to synthesize a common set of metaphysical and mechanistic assumptions about affective phenomena. In particular, affective phenomena can be organized into concerns—which are manners of being oriented actionably toward the world—or features—which are dimensions that characterize that agentivity. This framework therefore offers a principled agenda and launchpad for both organizing existing perspectives and generating new ones, suggesting that disorders such as depression can be reinterpreted through the lens of agentive organization. Chapter 3 investigates whether implicit sense of agency—operationalized as the intentional binding effect in an interval estimation task—can be manipulated via morphosyntactic priming. Participants completed trials in which they either pressed a button themselves or observed the computer press it for them, after which a fixed delay occurred followed by an auditory tone. They then estimated the interval between action and outcome using a continuous slider. These events were preceded by sentence primes crossing Voice (Active, Passive) with Point of View (1st, 2nd, 3rd, and Agentless)—with agentless active primes taking the form of unaccusative constructions and agentless passive primes omitting the by-phrase. Depressive symptom severity was also modeled as a clinical moderator. Linear mixed-effects models replicated the intentional binding effect in both baseline and linguistic trials, with self-initiated actions associated with more negative estimation error than computer-initiated actions. Contrary to preregistered hypotheses, Voice and Point of View did not significantly modulate intentional binding in the full model. In addition, follow-up analyses suggested that depressive symptom severity did not attenuate intentional binding globally; rather, binding remained broadly consistent across conditions. However, intentional binding was selectively flattened after first-person active voice priming at high depressive symptom severity, suggesting a context-dependent relation between depressive symptoms and morphosyntactic framing of implicit agency. Chapter 4 characterizes the intracranial EEG spectral signatures of uncertainty about what to do as a computational quantity under active inference. Ten patients with refractory epilepsy completed a three-armed bandit task during stereotactic EEG monitoring. An active inference model that was selected via BIC over reinforcement learning and Bayesian alternatives captured participant behavior by minimizing expected free energy to ascertain posterior beliefs for the three available a