When you talk to an AI, are you running on habit or actually thinking about what it is — and can you do both at once?

This explores how autopilot habits and deliberate thinking both show up when people deal with AI, and whether they compete or take turns. The most useful thing the corpus offers is that these are two different mechanisms, not two ends of one scale. Do people respond to machines mindlessly or reflectively? argues the field has lumped them together under the 'Computers Are Social Actors' (CASA) label. The first is the Media Equation, where small social cues like a voice or a name set off automatic, as-if-human responses. The second is Media Evocation, where the machine's unclear status (not quite a tool, not quite a person) makes people stop and reflect on what they're dealing with. They have different causes, so a single interaction can trigger both.

The twist is that the autopilot side isn't just old human habits carried over. Do humans apply human-human scripts to AI interactions? reports longitudinal evidence that people build new scripts made specifically for AI agents through repeated use, and then apply those scripts as mindlessly as the human ones. So reflection doesn't simply give way to habit over time. Moments of reflection may be how the new scripts get built: you work out how to talk to a voice assistant, then you stop thinking about it. Today's deliberate adjustment becomes tomorrow's routine, which is one plausible way the two modes coexist.

What people reflect on is also shaped from outside. How do users mentally model dialogue agent partners? finds users judge dialogue agents mainly on competence, then human-likeness, then flexibility. Those are the dimensions where a mismatch would jolt someone out of autopilot. What design features make users perceive AI as conscious? goes further: features like emotional display, autonomous action, and self-reflective talk reliably push people toward seeing the AI as conscious, and product teams control those features. In other words, the moments that trigger reflection, and the conclusions people reach, are partly design decisions.

A more philosophical thread suggests the reflective work is lopsided. Does AI generate genuine utterances or just text patterns? argues AI output only looks like communication, and humans do all the interpretive work that turns it into a conversation. That reframes the question. Scripts and reflection aren't happening between two parties. Both happen inside the human, who switches between treating the exchange as real and noticing that it's strange. Meanwhile, Can AI systems read cognitive state from interaction patterns alone? shows the AI side can read your hesitation and typing speed. Your autopilot behavior becomes a signal the system can use to time its responses, or to profile you.

The gap: the corpus doesn't have studies that measure, in a single interaction, when people flip from script to reflection and back. The framework is clear, but the moment-to-moment switching is still mostly theory here.