AI Placebo Study Shows Washing Inflates Performance Expectations

A recent study available on arXiv examines the issue of 'AI washing,' where products are advertised as being AI-enhanced even when they lack essential AI capabilities. In this research, 28 participants engaged in Fitts' Law tasks using a mouse under three different scenarios: no assistance, claimed predictive AI assistance, and claimed biosignal-enhanced AI assistance. The study evaluated objective performance metrics, subjective performance anticipations, perceived workload, and usability perceptions. Findings indicated that participants had significantly higher performance expectations in the placebo AI scenarios compared to the baseline, yet these expectations did not result in any actual differences in either objective or subjective performance. The paper underscores how AI washing can elevate user expectations without enhancing interaction results, revealing a discrepancy between perceived and real AI advantages.