By Laurel White
As the use of artificial intelligence (AI) programs continues to grow in the field of education, a UW–Madison faculty member says educators should work to build a keen awareness of the limits of some of the programs to interpret students’ body language and other physical interactions.
Mitchell Nathan, the Vilas Distinguished Achievement Professor in Learning Sciences in the School of Education’s Department of Educational Psychology, says educators are increasingly reliant on AI programs to evaluate data about students. But he notes those programs aren’t built to understand some very important — and very human — signals. He argues the systems’ misunderstandings could lead to incorrect decisions about how and what students are learning, and the resources students need for their development.
Nathan notes the growth and increased appreciation of embodied learning and multimodal learning analytics (MMLA) — data analysis tools which track how students talk, move, use tools, and interact with others — is making it easier for AI systems to monitor and evaluate students based on their nonverbal behaviors, not just what they type or say.
“Managing demands of complexity and speed is leading to growing reliance by education systems on disembodied AI (dAI) programs, which, ironically, are inherently incapable of interpreting students’ embodied interactions,” Nathan writes in a perspective article published in the March 2023 issue of Frontiers in Artificial Intelligence. “This is fueling a potential crisis of complexity.”
Nathan says educators should consider analyzing student data using methods that combine AI and human decision making. For example, so-called “detector-driven interviewing” methods use AI and non-invasive techniques to continually monitor human behavior for cognitive and affective patterns. He argues it is essential that humans evaluate the behaviors detected by these AI monitors.
Nathan urges educators to consider this more nuanced approach when using AI — and soon.
“This needs to change before educational practices become too dependent on dAI systems without proper considerations of ways to address these limitations,” he wrote. “The time is ripe to invest in alternatives such as augmented intelligence systems that cultivate the omnipresence and computational power of dAIs with the embodied meaning making of human interpreters and decision makers (as illustrated by approaches such as detector-driven interviewing) as a means to achieve an appropriate balance between complexity, interpretability, and accountability for allocating education resources to our children.”
Nathan has spent decades researching how people think, teach, and learn, with particular emphasis on the role that language and embodied processes plays in mathematics and engineering learning, teaching, assessment, and the design of educational technologies. Prior to his entry into educational psychology, he earned degrees in mathematics and electrical and computer engineering and worked in private industry developing AI and robotic systems.
Read the full Frontiers in Artificial Intelligence article here.