I am wondering if we need to reconsider the "within-subjects study" norm in HCI, now that AI biasing effects are so possible. "The AI helped me do X task easier and faster" is increasingly not a very interesting conclusion, when the risk of demand characteristics is so high.
