Does comparing machines to humans actually reduce social responses?

Eun-Ju Lee argues that the "machine heuristic" — introduced into human-machine communication (HMC) research as a refinement of the CASA paradigm to explain why people respond differently to AI than to human agents — is often operationalized in a way that contradicts the mindlessness logic it is supposed to support. Many studies measure it with items that explicitly compare machines to humans, such as "Machines are more trustworthy than humans" or "When machines perform a task, the results are more objective than when humans perform the same task" (citing S. Lee et al., 2023; Sundar & Kim, 2019). Lee's point: "If machine heuristic hinges on what separates machines from humans, then stronger beliefs in machine heuristic will inhibit, rather than facilitate, social responses to computers, directly challenging the mindless explanation for the CASA paradigm."

The reasoning works through what the construct is supposed to measure. Machine heuristic is meant to be "stereotypes about the operation of machines" (Sundar, 2020) — qualities like being objective, unbiased, accurate, unemotional, and reliable. But once items frame those qualities comparatively against humans, endorsing the heuristic means actively registering machines as unlike humans, which should work against treating a machine socially, not reinforce it. Lee also flags that the construct blurs into "personal beliefs" or "folk theories" (DeVito et al., 2017) about how systems work and what follows from that, and that some operationalizations are double-barreled — bundling a belief about machines with a specific behavioral consequence, as in an item justifying disclosure of private information "because" machines handle data securely, which Lee says comes "dangerously close to a tautology" when that same disclosure is the outcome being predicted. A further structural problem: if machine heuristic is strictly a belief about machines, it should be irrelevant, or inversely related, to how people rate a human agent — yet some studies find it predicts only AI-agent evaluations, a pattern simple mediation models aren't built to explain.

This sharpens the account given in Does machine agency exist on a spectrum rather than binary?, which defines machine heuristic simply as a mental shortcut where users attribute machine characteristics ... when making judgments about an interaction. Lee's critique shows that definition hides a measurement choice with opposite theoretical consequences: a non-comparative item (machines are reliable) is compatible with mindless human-like treatment, while a comparative item (machines are more reliable than humans) is not. Read together with Can mindlessness in CASA effects be measured independently?, the two notes trace why Lee treats neither of HMC's two standard mechanisms — mindlessness or machine heuristic — as adequate on their own, which is the premise for the integrative model the source paper goes on to propose.

The excerpt does not resolve how machine heuristic should be measured, nor does it present the integrative alternative Lee builds toward; it is a conceptual audit of existing operationalizations, not a new instrument or an empirical test. The implication is scoped accordingly: findings that use comparative machine-heuristic items to explain mindless, human-like responses to AI should be read with that internal tension in mind, since the measure may be pulling against the mechanism it is cited to support.