An exploratory behavioral and electroencephalographic study of artificial intelligence-assisted learning modes in high school students

As artificial intelligence becomes a staple in educational settings, understanding how different interaction styles affect student cognition is critical. This study moves beyond the binary "AI vs. no-AI" research framework by categorizing AI interactions into three distinct modes: Tutor (Socratic guidance), Collaborator (shared effort), and Solver (direct answer provision). The researchers aimed to determine if these modes elicit different behavioral responses and neural signatures in high school students.

Researchers recruited 48 high school students (ages 14–18) to complete two parallel-structure quizzes involving math, reading, and problem-solving tasks. Participants rotated through the three AI interaction modes while wearing a consumer-grade EEG headband to monitor frontal brain activity. Behavioral metrics—Initiation, Processing, and Stress—were recorded using a standardized observation rubric, and statistical analyses (including Friedman tests and repeated-measures ANOVA) were applied to evaluate differences across the modes.

• Behavioral Differences: The study found statistically significant differences in student behavior across the three AI modes. Specifically, the modes influenced how students initiated tasks, processed information, and experienced stress.
• EEG Observations: While the researchers identified descriptive patterns in frontal EEG spectral activity (such as variations in Alpha, Beta, and Theta bands) across the different AI modes, these findings did not reach statistical significance.
• Carryover Effects: The study explored whether AI interaction leaves a "carryover" effect on neural activity during subsequent independent work, providing a framework for future research into the lingering cognitive impacts of AI-assisted learning.

This research provides a foundational, replicable framework for studying human-AI interaction in classrooms. By demonstrating that different AI modes elicit distinct behavioral responses, the study suggests that the design of educational AI tools should be carefully tailored to support specific cognitive goals rather than treating all AI assistance as a uniform experience. It highlights the need for more nuanced pedagogical approaches to AI integration.