Building Agency: Toward a Mechanistic Understanding of Human Autonomous Goal Pursuit

Author: Aenugu, Sneha R.

Year: 2027

Degree: Dissertation (Ph.D.)

Advisor: O'Doherty, John P.

Committee Members: Sprenger, Charles David; Perona, Pietro; Doyle, John Comstock; O'Doherty, John P.

Option: Social and Decision Neuroscience

DOI: 10.7907/gh3v-dt73

Abstract

Agency is unique to biological life and reaches its pinnacle in humans, whose behavior is marked by the ability to set elaborate long-term goals and pursue them efficiently. This thesis develops a mechanistic account of human agency by investigating the computational and neural mechanisms that enable humans to select among competing long-term goals, sustain commitment over extended periods, monitor goals relative to alternatives, and flexibly switch when necessary.

To study extended goal pursuit in a controlled setting, I developed behavioral paradigms inspired by board games: simulated worlds in which participants freely set and switch between goals. Because these goals unfold over time, successful performance requires balancing stability in commitment with flexibility in response to changing circumstances. By intermittently devaluing goals during the games, I created situations in which these competing demands could be directly measured.

To explain how humans navigate this stability–flexibility tradeoff, I introduced the computational construct of goal momentum. Goal momentum integrates both goal progress and the velocity of progress to approximate time-to-goal-completion, a quantity relevant for agents optimizing discounted rewards. I show that momentum explains human overpersistence and preferences for progressing goals better than alternative accounts of goal selection, while also providing a parsimonious approximation of goal completion prospects.

I then extend the momentum framework to hierarchical goal pursuit. In a behavioral paradigm where extended goals are composed of extended subgoals, momentum accounts for persistence patterns at both goal and subgoal levels. Translating this paradigm to functional Magnetic Resonance Imaging (fMRI), I identify neural correlates of momentum computations and dissociable neural subsystems supporting goal and subgoal pursuit.

Finally, I lay the groundwork for a circuit-level account of momentum through a dynamical systems perspective. I show that goal commitment manifests as attractor modes in decision policies modulated by drives incorporating both progress and progress velocity, and hypothesize that switching costs during goal transitions arise from the intrinsic dynamics of goal commitment.

Taken together, this work advances a mechanistic account of human agency and provides a foundation for investigating the principles underlying artificial agency.

Files