Surfacing Isolated Learners with Outcome-Independent Mediation of Feedback between Teachers and Students Using AI

Researchers developed an AI-powered decision layer that identifies struggling students and prioritized course topics without relying on grades, combining student self-reports, observed learning difficulties, and teacher concerns. Testing in a graduate CS course showed the multi-signal approach achieved 96% accuracy in surfacing at-risk learners and aligned with instructor priorities, demonstrating transparent human-AI collaboration in educational settings.

This research addresses a fundamental challenge in AI-augmented education: converting raw feedback signals into actionable instructional decisions before grades reveal performance gaps. The proposed interpretable decision layer operates as a transparent mechanism that ranks topics requiring intervention by integrating three distinct data sources—prevalence of student learning difficulties, discrepancies between student self-perception and actual performance, and unresolved instructor concerns. This multi-signal approach generated notable results in its pilot study, achieving 96% accuracy in identifying at-risk learners compared to 91% for single-signal methods, while also achieving strong alignment with instructor priorities (Spearman correlation of 0.80).

The work reflects growing recognition that AI systems in educational contexts must balance predictive power with interpretability and human agency. Traditional outcome-based metrics like grades arrive too late to inform timely interventions, creating a gap where early warning systems prove valuable. By operating outcome-independently, this approach sidesteps fairness concerns associated with grade-based algorithms while maintaining credibility through transparent decision records that educators can scrutinize.

For the edtech industry, this represents methodological progress toward trustworthy human-AI collaboration in schools. The correlation between behavioral signals and learning-related constructs—reflective thinking, help-seeking, self-efficacy—suggests that student interaction patterns contain diagnostic information beyond explicit test performance. However, the study remains preliminary, tested on a single course with modest sample sizes, limiting generalization claims. Future research must validate whether these mechanisms scale across diverse institutional contexts, student populations, and subject domains while maintaining educator trust and decision-making authority.

• →Multi-signal integration surfaced 5% more at-risk learners than single-signal approaches, demonstrating synergistic value of combining diverse data sources.
• →The mechanism achieved 0.80 correlation with instructor priorities without using grades, enabling early intervention before performance becomes measurable.
• →Transparent decision records explaining topic rankings support human-AI co-agency by allowing educators to audit and override algorithmic recommendations.
• →Student behavioral signals aligned with learning-related psychological constructs, validating the approach's theoretical foundation beyond operational accuracy metrics.
• →Outcome-independent design avoids perpetuating grade-based biases while enabling timely identification of struggling students requiring support.