Cognitive Science Colloquium - Tali Sharot / Building efficient and useful knowledge systems
Cognitive Science Colloquium - Tali Sharot / Building efficient and useful knowledge systems
April 16 the Cognitive Science Colloquium finishes the semester with Tali Sharot from MIT and University Colleqe London, telling us about her work on how utility and emotion affect information seeking, belief formation, and decision making.
Human knowledge can be conceptualized as a network of interconnected concepts. An advantageous knowledge network will facilitate both coherent categorization and creative cross-domain inferences that facilitate innovation. Such efficiency is achieved when local clustering is balanced with long-range shortcuts—an architecture known as a small-world structure. We find that humans intuitively value information based on its ability to optimize such efficiency of knowledge organization (interestingly LLMs show similar tendencies). They combine this evaluation with two other key assessments of information - its usefulness in guiding action and its emotional impact - to guide information seeking. These three factors can be distilled into two fundamental components of value: extrinsic and intrinsic. Our findings reveal that these two components are represented by distinct brain systems, with the traditional reward system (e.g., VMPFC, striatum) selectively encoding intrinsic value. Finally, we introduce a novel tool that leverages language analysis algorithms to evaluate and score information based on its usefulness, emotional valence, and ability to enhance understanding. These scores are displayed alongside web search results, guiding users toward content that best fulfills their needs. Together, these findings reveal an adaptive prioritization of humans toward information that optimizes the different functions of the knowledge system.