CAREER: Scaffolding Intelligence: How Structured Learning Guides the Formation of Neural Representations for Flexible Cognition

Better teaching and more capable artificial intelligence (AI) both depend on understanding how learning can become faster, more efficient, and more flexible. Structured learning, in which information is presented in a meaningful order so that learners first understand simpler concepts and then build toward more complex ones, is known to improve learning. However, it is still not well understood how this kind of scaffolding helps the brain turn experience into flexible knowledge that can be used to intelligently solve new problems. This question is especially important because humans are far better than current AI systems at applying what they have learned to new situations, whereas many of the current AI systems require substantial retraining even when only small parts of a task or environment change. This project investigates how different parts of the brain work together during structured learning to build knowledge about how a task is organized. By revealing how structured learning helps the brain build flexible and accurate knowledge from fewer experiences, this research will advance understanding of efficient human learning and intelligence, inform better teaching, and provide insights relevant to AI research on how systems might learn from less data and perform better in new situations. The project examines how the structure of learning experiences shapes the brain mechanisms that support abstraction, generalization, and flexible decision-making. It tests the idea that well-structured curricula help the brain organize knowledge compositionally, so that smaller pieces of knowledge act like building blocks that can be combined in new ways to solve problems that were not directly taught. Across three related studies, the research asks when structured and unstructured learning begin to diverge, how structured experience changes how the brain represents tasks in the hippocampus and prefrontal cortex, and how prior structured learning shapes the brain representations that support more efficient learning in new, more complex tasks. To address these questions, the project combines behavioral experiments with magnetoencephalography (MEG) and functional magnetic resonance imaging (fMRI) to track when brain representations of task structure emerge, how they are organized, and how they transfer across tasks under different learning conditions. There are plans to use computational models and analyses of brain activity patterns to track how learners come to understand hidden task structure and how the brain’s representation of that knowledge develops over the course of learning. Together, these studies aim to establish how structured learning shapes brain representations that allow reusable knowledge to be flexibly combined across tasks, supporting efficient learning and problem solving in new situations. In doing so, the project aims to advance understanding of the brain mechanisms that support intelligence and flexible decision-making, while providing insights relevant to education and AI research on how learning can become more efficient and generalize more effectively. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2539348 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Seongmin Park | Institution: Virginia Polytechnic Institute and State University, BLACKSBURG, VA | Award Amount: $766,200 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2539348 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2539348.html