Strand 2
Strand 2: Socio-technical Foundations of Trustworthy AI
Strand 2 explores how to appropriately balance trust between students, teachers, and AI Partners during small-group learning in K-12 classrooms. Our goal is to improve collaborative student learning and engagement by designing trustworthy AI.
Our guiding research question is: "What socio-technical approaches are needed to appropriately calibrate trust in AI during small group collaborative learning in K-12 classrooms?" Building appropriate trust to support uptake and effective use of AI tools is particularly critical when supporting collaborative learning, which involves complex knowledge sharing and negotiations among learners and teachers and creating psychologically safe learning conditions at multiple levels (individual, small group, whole class) to promote engaged participation by all students.
By integrating insights from the learning sciences, human-computer interaction, and team science, we are moving away from generic, "black box" AI. Instead, we are building transparent systems that prioritize reliability, safety, and privacy. To achieve this, our work is organized into two primary themes: Frameworks and Measures for Understanding Trust and Calibrating Trust with "Under-the-Hood" Environments.
The connections between privacy, fairness, safety, and trustworthy AI are underexplored in classrooms, requiring a re-envisioning of what it takes to design trustworthy AI in K-12 schools. Our goal is to build on extensive research on ātrust in AIā in adult populations to re-envision how to design trustworthy AI in K-12 classrooms.
Our goal is to study novel āunder the hoodā AI learning environments for improving studentsā understanding of the inner workings of iSATās AI Partners. We run participatory studies with students and teachers to investigate cognitive, social, and technical factors that shape trust in interactions between students, teachers, and AI. This can ultimately provide students the agency to accept or contest AI inferences, and adapt AI models to their context, with appropriate safeguards in place.