What makes something valuable? Is value a stable property of the world, or does it depend on the goals we are trying to achieve?
In this Causerie session, I will discuss a goal-centric view of learning and decision-making in humans and AI, where value is not fixed but depends on what an intelligent agent is trying to do. When goals change, the same object, action, or outcome can take on a different meaning.
This perspective helps explain how agents select relevant information, reshape internal representations, and adapt flexibly to new demands. It also questions the idea, common in AI, that rewards are simply given by the environment, and instead treats value as something constructed relative to goals.
I will conclude with a brief presentation of one of my recent studies, focusing on how value representations support flexible learning and decision-making when goals shift.
About Ali Shiravand
Ali Shiravand is a PhD candidate and Normalien at the École Normale Supérieure (ENS) Paris, affiliated with the Laboratoire de Neurosciences Cognitives et Computationnelles (LNC2). Supervised by Prof. Stefano Palminteri and Prof. Benedetto De Martino (UCL), his doctoral research uses computational modeling to study how humans and artificial agents learn, represent value, and make decisions in complex environments. Originally trained in computer science, he became interested in human learning and decision-making through a broader curiosity about intelligence, and how these insights might help build more aligned and explainable AI systems.
Details
Date and Time: Thursday, 16th of April 2026 - 6:30 PM
Location: Sony CSL, 6 rue Amyot, 75005 Paris
Registration: luma - Registration is not mandatory, but it helps us better prepare the space and logistics.