Abstract:

Many applications of AI in education today focus on supporting interactions between an individual and an AI partner in learning. However, AI has been used for decades to support students’ collaborative problem-solving activities through innovative technology-supported learning environments, primarily in controlled environments and higher education and remains underexplored as a site for human-AI teaming. As part of our work as a multidisciplinary AI Institute, we are exploring the potential for human-AI teaming in noisy, face-to-face classrooms in secondary schools to support students’ collaborative problem solving skill, a key durable skill for enabling participation in civic and economic life. In this paper, we present a framework for integrating AI supports for Human-AI teaming into problem-based instructional materials for secondary classrooms.

The framework is grounded in sociocultural learning theory, empirical research on collaborative learning in STEM, and implementation research conducted within our interdisciplinary AI Institute. The framework identifies five functional roles that AI partners can play in supporting productive collaboration, including:
(1) maintaining a shared problem space for the classroom;
(2) representing and tracking systems of ideas;
(3) promoting equitable participation and amplifying lower-status and influence voices;
(4) challenging and extending collective understandings of problems; and
(5) supporting metacognition and reflection.

AI partners are expected to operate at multiple scales, teaming with small groups, the teacher, or the whole class, and can intervene both in real-time and as post-hoc support for whole-class reflection. The framework specifies integration of these AI partner functions within curricular structures or routines that foster students’ epistemic agency in collaborative knowledge building while also supporting the teacher in guiding the direction of learning for the class in a way that reflects students’ own evolving questions and ideas.

We illustrate how the framework has guided our team in interpreting and evaluating implementation data from classrooms where students are using two AI partners, our team was built to support their collaborative learning, evaluating the effects of AI-supported collaboration, and iteratively refining the AI partners. Preliminary insights include the need to create more customizable AI Partners (e.g., by allowing students and teachers to co-create the definitions of collaboration that are used to model their classroom discourse) and to enhance the AI Partners’ capacity to reference dynamic knowledge representations of core curricular ideas, enabling better support of student small-group work by grounding prompts and feedback in shared classroom knowledge. This work contributes to a growing understanding of how natural language processing and AI systems can be ethically and pedagogically integrated to advance human-AI teaming in authentic, collaborative classroom environments.

Keywords:

AI, human-AI teaming, collaborative learning.