Integrating Large Language Models into the Chemistry and Materials Science Laboratory Curricula

Abstract

RecommendationsF rom robotic teachers to interactive tomes, personalized, artificial intelligence (AI)-based education has been a recurring theme in science fiction since the mid-20th century.With the recent rise of large language models (LLMs) such as ChatGPT and Gemini, these once-fantastical visions are rapidly shifting from fiction to fact.In the realm of chemistry and materials science education�and especially in courses with a laboratory component�the potential for revolutionizing education is immense.Laboratory experiences are meant to foster hands-on skills, reinforce core concepts, and develop students' identity as scientists. [1][2]2][3][4] Yet research consistently finds that traditional lab courses underperform in achieving deeper conceptual understanding and gains in critical thinking. 2,3,[5][6]6][7] In an era of LLMs, one can envision improving these student outcomes using AI-based feedback or personalized guidance to reshape laboratory courses.Alongside cautious optimism, however, higher education faces unsettling questions surrounding ethical usage and the penchant of LLMs to deliver hallucinations and misinformation.In lab-based courses, these risks become particularly fraught.Presently, the intersection of education and LLMs is a "wild west", where individual educators are experimenting with LLMs in classrooms and forging new approaches as they go.Best practices remain ad hoc, and new strategies�whether brilliant or flawed�are rarely shared widely.The result is a patchwork of innovation that calls for a more coherent framework and a better approach to sharing our successes and failures.In an early effort to answer this call, we convened a two-day workshop, "Integrating Large Language Models into the Chemistry Laboratory Curriculum", supported by the National Science Foundation (NSF).Our goal was to bring together a diverse group of graduate students, post-doctoral scholars, and faculty (41 participants, in total) with expertise spanning materials chemistry, chemical education, and computer science�each offering perspectives on laboratory teaching, undergraduate research experiences, and developing custom LLM-based tools.We spent much of the first half of the workshop sharing our prior experiences in areas ranging from evidence-based lab pedagogy, principles behind LLMs, and early attempts at incorporating LLMs in our courses.The second half was organized as a "hackathon", where we worked in groups to devise novel ways of integrating LLMs into chemistry education to ultimately create virtual laboratory teaching assistants or personalized tutoring systems.Because each team included both instructors armed with existing instructional resources and computer scientists with fluency in LLMs, we made surprisingly rapid progress in the short time

Topics & Concepts