Human and Machine Alignment

Catalog description

Humans and machine learning information processing systems create internal representations that enable them to perceive additional information, categorize it, make decisions, and plan actions. How can we measure the extent to which these representations align? Are machine learning representations that resemble human ones more effective? What insights about human representations can we gain from machine learning models? How can we make human and machine representations more similar? In what ways can we address ethical and safety concerns? With recent advancements in machine learning and cognitive science, these questions have become focal points of research. This course will explore this dynamic area, involving a close examination of foundational and recent papers in the emerging field of representational alignment, alongside cutting-edge research.