Goal-based representations in decision-making

In this thesis, I investigate the computational and neural mechanisms by which the brain supports flexible, goal-directed decision-making. Adaptive behaviour requires the brain to construct and maintain internal models that are dynamically shaped by current goals. This work integrates computational modelling, behavioural tasks, and neuroimaging to explore how goals modulate neural representations, how value is encoded, and how confidence influences learning and choice across different cognitive domains. First, using fMRI and a value-based decision-making task, we demonstrate that a dedicated brain system, centred on the ventromedial prefrontal cortex (vmPFC), encodes the "usefulness" of a stimulus in alignment with a behavioural goal. This suggests that the brain constructs value representations that are highly contextual and goal-dependent, as opposed to simply monetary value or reward. Second, through two virtual maze navigation experiments and computational modelling, we show that participants' spatial representations are best described by a single predictive map, rather than by the creation of separate maps for each goal. Finally, we explore the domain-generality of metacognition and find evidence for a "confidence leak," whereby confidence in one task systematically influences choice behaviour in a subsequent, unrelated task, in a different cognitive domain (i.e, value and perception). This finding suggests a shared, domain-general confidence signal that, while beneficial for cognitive control, can also lead to cross-task biases. Together, these findings provide a framework for understanding how the brain supports flexible, goal-directed behaviour by dynamically shaping internal models, value signals, and metacognitive evaluations in response to changing goals and environmental demands.