Researchers often decide a computer-user is being 'mindless' just by watching politeness — but could any result ever prove them wrong?

This explores whether the long-standing idea that people respond to computers socially without thinking (the 'Computers Are Social Actors', or CASA, tradition) can be put to a real test. The sharpest answer in the corpus is that, as most studies are designed, it can't. Can mindlessness in CASA effects be measured independently? shows that researchers usually watch someone be polite to a computer and then conclude that the politeness must have been mindless. The explanation is read backward from the behavior it is supposed to explain. If every social response counts as proof of mindlessness, no result could disprove it. Lee's fix is simple to state: measure mindlessness separately from the social behavior. Only then can you check whether it actually causes that behavior.

The problem is bigger than method. Do people respond to machines mindlessly or reflectively? argues that the field has merged two different explanations under one label. In the first, people react automatically to minimal cues like a voice or a name. In the second, people react thoughtfully because machines are odd things that don't fit neatly into 'person' or 'object'. Both produce the same visible result: someone treats a machine socially. So watching that behavior can't tell you which explanation is right. You need to know what is happening inside the person's head.

When researchers do ask, the autopilot story starts to break down. In interviews, owners of the Nomi in-car robot described naming it, dressing it, and treating it as a social partner while fully aware of what they were doing (Do people interact mindlessly or mindfully with in-car AI?). Their social treatment was deliberate. A related lateral finding: people likely to cheat choose to report to a machine rather than a human because it feels like a judgment-free zone (Do dishonest people prefer talking to machines?). That is a strategic choice based on what a machine is, which is the opposite of mindless. Suleyman's warning about 'seemingly conscious AI' (Could AI systems fool people into thinking they're conscious?) adds a further twist. As machines get better at seeming human, the line between 'tricked by cues' and 'reasonably convinced' gets harder to draw.

The corpus also points to a way to get the independent measure Lee asks for, though it doesn't make the connection itself. Research on reading cognitive state from gaze, hesitation, and typing speed (Can AI systems read cognitive state from interaction patterns alone?) suggests that how carefully someone is thinking could be tracked during the interaction, separately from whether they say 'thank you' to the machine. That would let you test, rather than assume, whether social responses come with low attention. The same paper warns that these signals can also be used for manipulative profiling.

The corpus has no study that actually runs this kind of separate measurement. What it does show is why one is needed and what such a study would have to separate. The surprising part: 'mindlessness' may have survived for decades not because evidence supported it, but because the way it was studied meant evidence could never count against it.