EAGER: Feyenman: Developing and Evaluating Computational Transparency Approaches for Understanding AI Systems

AI is rapidly becoming a defining technology of modern society. Artificial intelligence is rapidly expanding into medicine, science, engineering, education, software development, and countless other domains, transforming how individuals interact with information and how organizations conduct work. The huge growth in AI usage has not been matched by public understanding of how AI systems actually work. Individuals primarily experience the outputs of AI systems without insight into the inference and training processes, probabilistic reasoning, agentic tools, and other mechanisms that require vast amounts of parallel and distributed computation. This project investigates whether making hidden AI computation visible can help people better understand how AI systems operate. By making hidden computation visible, Feyenman seeks to help people better understand both the opportunities and implications of AI as it becomes an increasingly important part of our daily lives. More broadly, the project investigates whether computational transparency can foster a more informed public dialogue about AI and its role in society. In this project, the team will investigate, design, implement, analyze, and demonstrate the Feyenman cluster—a 30-node computing system that combines motion, light, and sound to visualize the hidden parallel and distributed computation underlying modern AI systems. Through a series of interactive experiments, we will evaluate whether making AI computation visible improves understanding of how AI systems generate outputs and why they exhibit particular capabilities and limitations. Consistent with findings from science communication, informal learning, and AI-literacy research, Feyenman emphasizes active participation, visibility into otherwise hidden computational processes, and mental-model development. Beyond the creation of the artifact itself, the primary research outcomes of this project are an empirical understanding of how computational transparency influences mental-model formation, a framework for evaluating computational transparency approaches, design principles for representing AI computation through motion, light, and sound, and the first framework for assessing the mental models participants develop through experiential immersion in AI computation. The results will provide a foundation for future research, educational exhibits, and public-engagement activities that seek to make AI systems and distributed computation more understandable, accessible, and relevant to discussions about their role in society and the human condition. 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.