Content area
Goals are a prevalent idea across the cognitive sciences, and as the basis for agentic and motivated behavior, have been of interest to psychologists since the field’s inception. The human concept of a goal is remarkably flexible: from the representation of a single goal, be it self-proposed or externally provided, we can plan how to pursue the goal, propose similar ones, evaluate progress toward a goal, and infer how well it aligns with observed behavior. How do we do it? What sort of representation might offer such behavioral flexibility? How do we even define a goal? Figure 1 summarizes my answer (and this thesis). In Chapter 1, I review the use of the term ‘goal’ across the cognitive sciences and artificial intelligence to motivate the gap this thesis attempts to bridge. Its ubiquity notwithstanding, the term goal is often left undefined in psychological work, and when definitions are provided, they can suffer from inconsistency and a lack of technical rigor. Unlike psychology, approaches in artificial intelligence have offered technically precise definitions of goals, driven by the necessity to implement and evaluate ideas in code. However, here the definitions provided are often ones of convenience, reducing goals to target states of the world to achieve. While technically successful, these definitions fail to capture the rich, creative, and often idiosyncratic goals people routinely create for themselves and for others. This dissertation offers a path forward by studying human-created goals, proposing a program-based representation that can capture the complexity of human-created goals, demonstrating that these representations can be used to generate new goals or infer existing ones, and taking a foray into studying similar questions in large language models. Chapter 2 describes my initial investigation of human goals, their representations as programs, and the roles cognitive capacities, such as common sense and compositionality, play in creating them. Chapter 3 offers additional details on these reward-producing programs and proposes a computational model capable of generating coherent and creative human-like goals. In Chapter 4, I provide evidence that these representations can generalize, modeling human-created goals as reward-producing programs in a different environment and using these programs to perform goal inference from behavior. Chapter 5 turns to study task representations in large language models as a precursor to further investigation of how they might represent goals with human-like richness. Finally, Chapter 6 summarizes insights from this work and open questions left unanswered.