Welcome to the Computation & Cognition Lab at KAIST AI. Our mission is to advance AI by combining computational methods with insights from human cognition and development. We take a data-centric view of intelligence: what a machine can learn is fundamentally shaped by the data and experiences it encounters. Social interaction is one of the most powerful ways to create such experiences. Other intelligent systems, and the artifacts they create, can serve as rich sources of knowledge, helping machines learn more effectively and efficiently than by discovering everything on their own.

This perspective motivates two complementary directions in our research: machine social learning and social world modeling. In machine social learning, we study how models can develop stronger cognitive capabilities by learning from other models, from themselves, and through interaction with their environments. Our work spans synthetic data, distillation, self-improvement, memory, and inference-time interaction. In social world modeling, we study how machines can understand and reason about people and other agents—their beliefs, intentions, relationships, norms, behavior, and embodied interactions.

Together, these directions ask how machines can both learn from other intelligences and understand the social worlds they inhabit, with applications ranging from reasoning and interactive agents to human behavior simulation and physical AI.

Our current research interests include, but are not limited to:

• Developing reasoning and cognitive capabilities in foundation models and agents.
• Scalable data synthesis, selection, and curriculum design.
• Distillation, self-improvement, and learning from model-generated feedback.
• Social world modeling for simulation, virtual agents, and physical AI.
• Systematic evaluation of foundation models and agents.