Close

Presentation

From Textbooks to Tutors: Can Generative Language Models Engage and Empower Students?
DescriptionTextbooks often fail to engage students and go unread. Increasingly instructors are making textbooks optional and relying on presentation slides to convey course content. While slides may provide an easily digested summary of textbook highlights, relying on them as the sole way of communicating the content of a field may result in a superficial understanding.

Plato might have predicted the failure of textbooks. He saw the advent of books as a threat to memory and critical thought. Many focus on his critique of books from the perspective of replacing human memory with writing. Perhaps his greater concern may have been the tendency of books to replace interactive discussions with a tutor with linear monologues that fail to engage with the interests and aptitudes of students. Textbooks typically adopt a didactic approach to teaching: students are passive recipients of instruction. The advent of AI has the potential to change textbooks from an instrument of the didactic method to an instrument of the Socratic method in which students and tutors engage in a dialog and active learning (Oluwatoyin, 2015). Technology may have now matured to address Plato's concern.

Generative large language models (LLMs) can enable flexible conversational interactions with an AI tutor. LLMs have the potential to better engage students in complex material by conveying the details of the textbook in a way that merely reading the material or classroom instructions cannot. Like the primer in Neil Stephenson's science fiction masterpiece Diamond Age, an AI tutor might be able to answer students' questions, ask students questions to probe their knowledge, and point students to content to expand their knowledge. These models might even satisfy Socrates with how they enable his method (Gregorcic & Polverini, 2023).

This paper describes a tutor that complements an introductory human factors textbook based on generative large language model, often referred to as AI. This AI tutor builds on the long history and promise of computer-based tutors. Computer-based tutors have been developed and deployed for over 60 years with varying degrees of success (Bitzer et al., 1961; Smith & Sherwood, 1976). Computer-based tutors and human tutoring can substantially enhance learning. One review found that human and computer tutors produce a similar learning benefit with effect sizes of 0.75-0.80 (VanLehn, 2011). Despite the long history and important contributions, computer-based tutors are not as common as textbooks and lectures. The advent of LLMs may accelerate the development and deployment of AI tutors.

This case study considers basic elements of AI tutor development that might apply to other AI applications. We assess textbook content by comparing how textbook paragraphs cover content from 50 years of human factors and ergonomics research across nine journals. We also demonstrate how a Human Factors and Ergonomics (HFE) tutor can ask and answer questions and how it can facilitate exploration and mastery of content.
Corresponding Author