Keynote 4: Towards a science of human simulation by Dr. Tim Althoff

Simulating human individuals is a double-edged sword. While it offers significant potential benefits—such as empowering underrepresented voices, enabling personalized safety assessments, and improving accessibility—it also carries significant risk of automated exclusion and misrepresentation. Modeling and supporting diverse individuals is therefore a critical frontier for artificial intelligence research. Current large models often treat humanity as a monolith, optimizing for a fictitious "average user" while failing to capture the complexities of any single individual. To advance the vision of making human simulation a rigorous science, I will discuss the urgent need for systematic evaluation methods grounded in specific individuals, goals, and applications. As a first step toward principled evaluation, I will introduce a novel large-scale benchmark. I will describe our motivation, key design decisions, and the opportunities we hope this shared resource will create for the broader community.