Pair-Up

Might combining individual and collaborative learning activities dynamically, as the need arises, be an effective way of personalizing AI-based instruction? 

This question was inspired, in part, by observing teachers use Lumilo: When multiple students were struggling at the same time, they spontaneously teamed up students, so students could help each other.

Challenge

In teaching, it is common to mix activities that students do individually, such as problem-solving practice, with activities they do collaboratively, in small groups, like "Think-Pair-Share" and "Jigsaw." Past research confirms that mixing individual and collaborative learning can lead to better learning outcomes than sticking to just one or the other. 

Although teacher typically such combinations typically unfold according to a teacher's pre-made plan, there may be value in allowing students to move between individual and collaborative activities on-the-fly,  as needed, rather than following a set plan, so everyone gets a dynamic, personalized mix that lets them learn at their own pace. 

Yet how might teachers manage the their classroom sessions during which students frequently move between individual and collaborative learning?

Highlights

• Conducted design-based research with teachers and students to explore how best to combine orchestration efforts and responsibilities between students, teachers, and AI.  

• Carried out a data mining study to study the feasibility of different pairing strategies within AI-based classrooms (e.g., to team up a struggling student with a non-struggling student).

• Created a socio-technical ecosystem that supports dynamic combinations of individual and collaborative learning: This system includes a new orchestration tool for teachers called Pair-Up together with two AI-based  tutoring systems for students, APTA and Lynnette, for group learning and solo learning, respectively. Both support problem-solving practice.

• Ran a pilot study and an evaluation study in classrooms: These studies tested the feasibility of the new ecosystem and compared student learning between pre-planned v. dynamic combinations.

Discoveries

• Teachers like the idea of sharing control with an AI to make managing the class easier. They want the AI to suggest which students should work together but are not keen on AI changing their own decisions.

• Often, there is a balance between wanting to team up students who are different (e.g., in terms of how well they have already mastered the targeted skills) and what's actually possible. This balance can change from class to class, indicating there is a need for customizable grouping rules and parameters.

• Middle-schools teachers and students were able to use the integrated system to orchestrate dynamic transitions between individual and collaborative learning. However, teachers and students wanted  different ways of controling the pairing and learning processes.

• While learning outcomes where comparable between standard (i.e., pre-planned) v. dynamic combinations, the pre-planned condition showed greater collaboration efficiency. On the other hand, teachers and students favored the dynamic approach for its flexibility and pedagogical value. 

Impact

• Insight into the design of an AI-based classroom orchestration tool that makes suggestions but respects the teacher’s autonomy. 

• One of the first empirical studies comparing dynamic versus pre-planned combinations of individual and collaborative learning in classrooms. 

• This study steps closer to the future of highly personalized smart classrooms that can address the unique needs of each student. We envision future classrooms in which the learning activities are constantly adjusted based on what each student needs, including deciding whether they should work alone or with others. 

Funding

• NSF (IIS) 1822861. PI: Vincent Aleven. Co-PIs: Nikol Rummel, Ken Holstein. “Human/AI Co-Orchestration of Dynamically-Differentiated Collaborative Classrooms.”