Do we react to AI on autopilot just because it seems human, or do we stop and consciously figure out what it actually is?

This explores whether people respond to AI on autopilot, treating it socially just because it shows human-like cues, or whether they knowingly think through what kind of thing they're dealing with. The corpus suggests that framing it as either/or is part of the problem. The two accounts describe different mechanisms, and the field has been mixing them up. Do people respond to machines mindlessly or reflectively? separates the classic 'Media Equation' idea from a second one, 'Media Evocation'. In the Media Equation, minimal cues like a voice or a name trigger automatic social reactions we don't even notice. In Media Evocation, machines provoke reflection because they're hard to categorize: not quite a tool, not quite a person. Both get filed under the same 'Computers Are Social Actors' (CASA) label, which hides the difference.

When researchers actually talk to users, the mindful side shows up more than the old theory predicts. In interviews with owners of Nomi, an in-car AI companion, most people named and dressed the robot and credited it with social agency. They did this consciously, not by reflex (Do people interact mindlessly or mindfully with in-car AI?). The surprising part is that richer voice and facial cues made people *more* reflective, not less. That reverses the classic assumption that convincing cues produce mindless responses. Research on how people mentally model dialogue agents points the same way. Users judge an agent mostly on how competent it seems, then on how human-like it is, then on how flexibly it communicates (How do users mentally model dialogue agent partners?). That looks like evaluating a partner, not reacting on reflex.

The reflective account has a darker side, though. Being mindful doesn't make you accurate. Does AI generate genuine utterances or just text patterns? argues that AI text carries the marks of communication without being a real exchange, and that users do the interpretive work that makes it feel like a conversation. That is a reflective act, and it still produces an illusion. The same tension shows up with consciousness: the features that lead people to see AI as conscious (emotional expression, autonomy, self-reflection, social behavior) are design choices that product teams control (What design features make users perceive AI as conscious?). Suleyman argues the near-term risk is that people *believe* AI is conscious, whether or not it is (Could AI systems fool people into thinking they're conscious?). A separate argument holds that the resulting harms can be studied without settling the consciousness question at all (Do we need to solve consciousness to address AI harms?).

Thinking about the AI doesn't keep people in control either. Leonardi's 'agency loops' describe people who credit AI with its own judgment and then feel less control and responsibility, even though they still hold formal authority (Does attributing agency to AI actually reduce human sense of control?). The useful finding is that the balance can be designed. In one study, AI assistants that asked reflection questions alongside giving advice beat assistants that only gave answers (Do reflection questions help people make better decisions with AI?). Interface choices push in either direction: chat interfaces can add mental effort to simple, goal-directed tasks (Is chat the wrong default interface for AI products?).

So neither model wins outright. Mindlessness explains quick reactions to cues. Reflection better explains sustained relationships with expressive AI, but reflection can still lead people into flattering misreadings. The question worth carrying away is less 'which one is true' than 'which one does this product's design encourage'.