Why would a survey question about 'trusting machines' predict how people rate AI, but not how they rate humans?

This explores why a belief that 'machines are objective and accurate' predicts how people judge AI, but not how they judge humans. The corpus has only one note that speaks to this directly, but it offers a sharp explanation: the pattern may come from how the machine heuristic is measured rather than from how people think.

The key argument is in Does comparing machines to humans actually reduce social responses?. The machine heuristic is described in theory as a mindless shortcut: something that fires automatically when people see a machine cue. But the survey items used to measure it are often explicit comparisons, such as 'machines are more trustworthy than humans.' Agreeing with that statement is a deliberate, reasoned belief about machines as a category, set against humans. Lee points out that this belief should, if anything, push people to treat machines less like people. That runs against the mindless social-response logic the concept is supposed to support.

Follow that reasoning one step further and the result in the question looks less puzzling. (This step is an inference, not something the note states.) If the scale measures 'how much better do you think machines are than people,' it should predict ratings when the source is labeled AI. That is exactly the case the belief is about. There's no reason it would also shape ratings of a human source, because the belief says nothing about humans except that they are the weaker comparison. A measure built around machine-versus-human contrast will naturally light up only on the machine side. So the 'AI-only' finding may be partly a measurement artifact: the scale captures explicit attitudes toward machines, not a general automatic tendency that would color all judgments.

The practical lesson for reading this research: when a construct predicts outcomes in only one condition, check whether the measurement itself has a direction built into it. The corpus doesn't hold the original experiments that found this asymmetry, competing explanations (for example, that people simply don't apply machine-related expectations to human sources), or follow-up work that tested non-comparative versions of the scale. To go further, Lee's critique is the place to start. For now, the collection supports this one explanation more than a settled answer.