Details
Refreshments at 10:30am. Talk begins at 11:00am.
Speaker:
Dr. H. Chad Lane
University of Illinois at Urbana-Champaign
Talk Description:
Learners now have immediate and anonymous access to AI tools that will complete their schoolwork for them. I call this temptation the effortless bypass dilemma, and I argue we should treat it as a motivation problem: GenAI is the first technology capable of broadly eliminating the productive struggle that learning depends on. Designing experiences students choose to invest in is also a technical problem. We need systems that can scrutinize fine-grained learning processes, and models that behave in pedagogically effective ways when trying to support learning. Thanktully, the research field of AI in Education has been building toward both for decades. In the NSF INVITE AI Institute, we are conducting research to accelerate this vision. We center skills that underlie effective learning, including persistence, academic resilience, and collaboration, and we treat them as first-class design targets. I’ll describe our learner modeling work, including theory-driven models of persistence and detectors that separate substantive problem solving from surface-level activity. I’ll close on our commitment to small, local AI, and on a complication: in benchmarking small models as CS tutors, most were compelled to complete the student’s next step despite explicit prompting. Most LLMs are trained to be compliant, not pedagogical. Prompting and fine-tuning only get us so far against that, which is pushing us toward neurosymbolic approaches that put pedagogical control outside the model rather than hoping it emerges from within.