Engagement Intensity as a Learner-Modeling Signal for Adaptive AI Ethics Instruction

Usage frequency is the strongest predictor of AI ethics readiness: Among the three intake measures tested, self-reported LLM usage frequency showed associations with all five baseline perception outcomes measured, outperforming both self-rated familiarity (three outcomes) and prior AI education (zero outcomes). This suggests that actual hands-on experience with AI tools matters more than formal training when assessing where trainees stand on AI ethics issues.

Prior coursework alone doesn't predict ethics perception: Surprisingly, prior AI education showed no significant associations with any of the five key perception outcomes studied. This finding challenges the assumption that traditional classroom instruction on AI topics adequately prepares researchers to think critically about AI ethics, suggesting that different types of learning experiences may be needed.

Non-linear patterns matter more than simple gradients: Rather than showing a smooth progression across all measures, the data revealed "threshold-like patterns" particularly for training interest and trust in accuracy—meaning there may be critical tipping points in experience level rather than uniform, incremental changes in perception as people use AI more.

Simple behavioral signals can enable personalized ethics training: The research demonstrates that lightweight intake surveys asking about actual LLM usage can effectively identify which trainees need different levels or types of AI ethics instruction, enabling more efficient and targeted educational approaches without requiring extensive prior coursework.

Self-rated familiarity serves as a useful secondary indicator: While usage frequency was most reliable, self-assessed LLM familiarity still correlated with three of five outcomes, making it a practical backup measure when detailed usage data isn't available.