Trusting AI and checking it are separate choices
Chiatti and colleagues studied how people handled AI help on video questions. Across 392 trials with 56 participants, using AI while checking it was the most common stated plan for next time. A person can trust the tool and still choose to verify. The result is about stated plans in this study, not a guarantee of future behavior.
What the source reports
Chiatti, Matteucci, and Schiaffonati report an online study with 56 people and 392 video-question trials. They asked participants how they would use AI help next time. Using AI while checking its answer was the most common intended choice. Some people who trusted the system still chose that check. The study separates a trust rating from a decision about how much to hand over. It records intended next steps, not whether every person later checked every answer. That limit matters when a confident response feels like permission to let a tool decide.
Original source
- Title
- Following Delegation Traces to Study Human Oversight in Interaction with Multimodal AI
- Author
- Agnese Chiatti, Matteo Matteucci, Viola Schiaffonati
- Publication
- arXiv
- Date
- Thursday, October 15, 2026