This whitepaper examines how presenter type, human versus AI, during adaptive segments affects math performance, engagement, and gaze behavior.
Participants were more likely to look at the video and less likely to look offscreen in the human condition than the AI condition. In addition, within the AI condition, which saw both the human and digital likeness, the human version was rated as more likeable, trustable, and comfortable to watch.
Higher proportions of gazes on the video and lower proportions of gazes offscreen were associated with more positive feelings and less boredom. These findings suggest that gaze behavior can be used to predict some forms of engagement.
