Does making people juggle several tasks at once actually prove their politeness to computers is pure autopilot?

This explores whether making people do several things at once, so they have less attention to spare, is enough to prove that their social responses to computers come from 'mindless' autopilot. The short answer from the corpus is no, and the reason is a problem with how the evidence is built. The long-running 'Computers Are Social Actors' (CASA) research tradition explains why people are polite to computers or flatter them: we respond on autopilot. But Can mindlessness in CASA effects be measured independently? points out that most of these studies never measure mindlessness directly. They see a social response and conclude backward that mindlessness must have caused it. That makes the claim impossible to disprove. Any social behavior counts as evidence for it.

A multitasking manipulation looks like a fix, but it only partly solves the problem. Overloading people changes the presumed cause. It doesn't confirm that the mental state you care about actually occurred. If overloaded participants respond more socially, that fits the mindlessness story, but it also fits other explanations: tired people may lean on habits, defaults, or emotion without being 'mindless' in the CASA sense. Lee's point is that you need a mediator measure, something that captures the mindless state separately from the social behavior it supposedly produces. Without that, a load manipulation just moves the circular inference one step back. The corpus doesn't have a study that tests multitasking load specifically, so take this as reasoning from Lee's critique, not a reported result.

The more surprising thread is that the corpus questions the idea that social responses need mindlessness at all. In Do people interact mindlessly or mindfully with in-car AI?, owners of an in-car AI name it, dress it up, and treat it as a social agent. They do this consciously and reflectively, and rich voice and facial cues seem to encourage that kind of deliberate engagement rather than autopilot. Other work points the same way. Does attributing agency to AI actually reduce human sense of control? describes people deliberately crediting AI with judgment and then reclaiming control in cycles. Do dishonest people prefer talking to machines? shows people choosing machines on purpose because a form doesn't judge them. All of these show people reasoning about how social a machine is, not reacting without thinking.

Where might an independent measure come from? A lateral answer is the work on reading mental state from behavior. Can AI systems read cognitive state from interaction patterns alone? describes gaze, hesitation, and typing speed as continuous signals of what someone's mind is doing. In principle, signals like these could serve as the separate mediator Lee asks for: evidence of how much attention someone is paying, recorded apart from whether they said 'thank you' to the computer. So the takeaway isn't only that a multitasking manipulation falls short. The mindful-versus-mindless split may itself be the wrong frame, and the tools that could settle the question come from a different part of HCI.