Trust and Reliance in XAI -- Distinguishing Between Attitudinal and Behavioral Measures

Trust and reliance are not the same thing, and conflating them creates confusion in AI research. The paper argues that researchers often measure how much people actually use an AI system (behavioral reliance) when they claim to be measuring trust (what people believe about the system). This distinction matters because transparency might affect these two outcomes differently.

Making AI more transparent doesn't automatically increase trust, partly because researchers haven't been measuring trust correctly. The paper notes that empirical research shows mixed results on whether explainability improves trust, suggesting that some of this confusion stems from using the wrong measurement approach—tracking behavior instead of attitudes.

Attitudinal measures (what people think and feel) should be used to assess trust, while behavioral measures (what people actually do) should be used to assess reliance. This clearer distinction would help researchers understand whether transparency is actually building confidence in AI systems or simply changing how much people use them.

Properly separating these concepts has real-world implications for responsible AI deployment. Understanding whether users genuinely trust an AI system versus simply relying on it is crucial for identifying potential risks—people might over-rely on a system they don't actually trust, or under-utilize a trustworthy system.