How do users develop different interaction scripts specifically for machines versus humans?

This explores whether people use the same social playbook for AI that they use for other humans, or build a separate set of habits just for machines. The corpus's clearest answer is that they build a separate set. The old assumption — that we mindlessly transfer human-human social rules to computers — gets revised here: extended CASA research finds that through repeated interaction, people actually develop *media-agent-specific* scripts, a second script system that coexists with the human one rather than just borrowing from it Do humans apply human-human scripts to AI interactions?. Longitudinal studies show systematic shifts in how people respond as they accumulate experience with an agent — which means the machine-script is learned, not imported.

What does that second script get built around? One line of work suggests people judge machine partners on partly different axes than human ones. When users mentally model a dialogue agent, perceived competence dominates (roughly half the variance), with human-likeness and conversational flexibility trailing behind How do users mentally model dialogue agent partners?. With a human, competence isn't usually the first thing your social script foregrounds; with a machine, it leads. So the machine-script is organized more around 'will this thing actually do the task' than around the rapport-and-face concerns that structure human talk.

Here's the tension the corpus surfaces, and it's the interesting part: the *interface* keeps pulling users back toward the human script even when the machine doesn't deserve it. Conversational design triggers the language competencies people built over a lifetime of actually communicating — but the AI isn't communicating, it's producing strings, and that mismatch generates failures that feel like user error but are really design error Why do users fail with AI interfaces designed like conversations?. A useful reframe alongside this: the system is better understood as a role-playing character generating character-consistent continuations, so folk-psychology applies to the simulated persona, not the machine underneath Should we treat dialogue agents as role-playing characters?. The user's two scripts are in conflict — the human one the chat box invites, and the machine one experience has taught them.

The friction this produces is concrete. People can't fully articulate what they want, and because AI responds rather than probes, it never helps them mature that intent the way a human partner would Why can't users articulate what they want from AI?. Conversation-analysis work names the missing move precisely: human dialogue is full of *insert-expansions* — clarifying questions before answering — and tool-using agents that skip them silently drift from intent When should AI agents ask users instead of just searching?. In other words, part of why users need a distinct machine-script is that the machine doesn't run the repair-and-clarify subroutines a human interlocutor automatically would.

The quietly surprising takeaway: the script-divergence may be a design problem more than a fact of human nature. When research drops the conversation metaphor entirely — generated task-specific interfaces beat chat in over 70% of cases Do generated interfaces outperform text-based chat for most tasks? — it suggests that much of the awkward 'talking to a machine' script exists only because we dressed the machine up as a conversational partner in the first place. Strip the human costume off, and users may not need a strained second script at all.