Colin West - Impacts of AI on Physics EducationLarge Language Models and Conceptions of Reasoning in Introductory Physics
ORAL · Invited
Abstract
Colin West - Impacts of AI on Physics Education
The recent emergence of generative artificial intelligence (AI) systems, including large language models (LLMs), has produced an explosion of work from fields as diverse as computer science, philosophy, and linguistics exploring the extent to which such systems might be said to be "reasoning" or to show forms of comprehension and understanding. This conversation mirrors important discussions that have always taken place within the physics education community with respect to students themselves: in many ways, our best assessment tools are designed also to probe whether our students are "really" understanding what we teach them or merely able to reproduce the outward signifiers of comprehension. I will review the current state of evidence regarding AI systems and the extent of their abilities to create the appearance of "reasoning" in the context of basic physics. Then, I will summarize some of the major theoretical arguments for and against the idea that generative AI can reason and understand, and turn some of these theoretical lenses on the state of physics education itself to as: If we treated our students as we treat AI systems, would we believe that they were truly understanding physics?
The recent emergence of generative artificial intelligence (AI) systems, including large language models (LLMs), has produced an explosion of work from fields as diverse as computer science, philosophy, and linguistics exploring the extent to which such systems might be said to be "reasoning" or to show forms of comprehension and understanding. This conversation mirrors important discussions that have always taken place within the physics education community with respect to students themselves: in many ways, our best assessment tools are designed also to probe whether our students are "really" understanding what we teach them or merely able to reproduce the outward signifiers of comprehension. I will review the current state of evidence regarding AI systems and the extent of their abilities to create the appearance of "reasoning" in the context of basic physics. Then, I will summarize some of the major theoretical arguments for and against the idea that generative AI can reason and understand, and turn some of these theoretical lenses on the state of physics education itself to as: If we treated our students as we treat AI systems, would we believe that they were truly understanding physics?
–
Publication: West, Colin G. (2023). AI and the FCI: Can ChatGPT project an understanding of introductory physics?. arXiv preprint arXiv:2303.01067<br>West, Colin G. "Advances in apparent conceptual physics reasoning in ChatGPT-4." arXiv preprint arXiv:2303.17012 (2023) (under review)<br>West, Colin G. "Conceptions of student reasoning and comprehension in physics through the lens of large language models" (In Prep)
Presenters
-
Colin G West
University of Colorado, Boulder
Authors
-
Colin G West
University of Colorado, Boulder