Towards an adaptive AI scaffold for developing student collaborative problem solving

Background Adaptive scaffolding, designed to assist students based on their needs, has been shown to have promising results in multiple studies. However, most of these designs of adaptations are informed by student choice or rely on aggregated measures of behavioural and performance indicators. Aims We study students' CPS interactions under maximal and minimal scaffolds, and propose a conceptual design of an adaptive scaffold approach that considers the dynamic nature of individual students’ processes during CPS. Sample 78 students aged between 14 and 15 years from a public school worked in triads on a mathematics collaborative problem-solving task through online video conferencing. Methods We examined sequences of students' processes during CPS at the individual level using speech and task logs. The adaptive scaffold design is derived using sequential patterning mining with independent effects logistic regression to obtain a list of sequence-based patterns associated with students’ performance improvements in solving a mathematics problem similar to that in the CPS task. Results Although students with the maximal scaffold had greater performance improvements and were significantly more on-task than students with the minimal scaffold, they engaged in more scripting behaviours. The derived adaptive scaffold design informed conceptual strategies in CPS phases, such as prompting students capable of answering questions to ask questions in the ‘problem identification’ phase. Conclusions This study confirms the need for an adaptive scaffold of CPS for secondary school students and proposes a design that highlights specific strategies for the ‘problem identification’ and ‘ideation, planning and decision making’ phases in CPS.