Why do people invent a whole new set of social habits for talking to AI, instead of just reusing how they treat other people?

This asks why people who talk to AI and other media agents develop their own separate set of habits and expectations for those conversations, instead of reusing the social rules they follow with other people. The older assumption in this research, known as CASA ('Computers Are Social Actors'), held that people automatically apply human social rules to machines. Newer long-term studies complicate that picture. Through repeated use, people build a second, machine-specific script that sits alongside the human one rather than replacing it Do humans apply human-human scripts to AI interactions?. These scripts form through experience, and that helps explain why they drift away from human ones. Chatbot studies show the social warmth people feel early on fades in a predictable way as the novelty wears off Do chatbot relationships lose their appeal as novelty wears off?. A first-session study captures the human script being tried out. Later sessions capture the specialized script that replaces it.

One reason the scripts split is that the partner behaves differently. Human conversation runs on constant small adjustments: switching register, renegotiating tone, letting the other person change how you talk. Alignment training fixes models into one communicative identity across every context, so users cannot reshape the model's behavior through the kind of back-and-forth they would use with a person Can language models adapt communication style to different contexts?. Once people learn that the move doesn't work, they stop making it. The script adjusts to what the partner can actually respond to. A related framing treats a chatbot as a character the model is role-playing rather than a mind with its own intentions Should we treat dialogue agents as role-playing characters?. On that view, users are learning the habits of a performance, not of a relationship.

Users also judge AI partners by a different scorecard. When researchers measured how people mentally model dialogue agents, competence explained about half of their impressions. Human-likeness and flexibility came well behind How do users mentally model dialogue agent partners?. A machine-specific script built around 'is this thing capable?' is a different object from a human script built around rapport and mutual obligation. Research on conversational alignment shows the parts can be pulled apart. Matching the partner's words helps with efficiency and understanding. Matching their emotion and tone of voice builds warmth and trust Do different types of alignment serve different conversational goals?. Human scripts bundle these together. With machines, people can keep the task-focused parts and drop the relational ones.

The medium shapes the script too. Changing only the communication channel in a human-agent collaboration task shifted trust and awareness of the shared workspace, in patterns that echo earlier findings from human-human teamwork research How do communication modalities shape human-agent collaboration patterns?. This suggests scripts depend on the channel as much as on whether the partner is human. One critic argues that chat itself is often the wrong interface for goal-directed work, because it brings back the cognitive load that clicking and direct manipulation were designed to remove Is chat the wrong default interface for AI products?. If that's right, some media-specific scripts are workarounds people develop for a poorly matched interface.

There is an irony here. Models can now predict human social norms more accurately than any individual person Can AI learn social norms better than humans?. Yet they cannot take part in the community processes that create and revise those norms Can AI predict social norms better than humans?. So the partner understands human scripts from the outside but cannot co-author new ones with you. That may be the deepest reason a separate script emerges: social scripts are negotiated between both sides, and with AI only one side is doing the negotiating. The collection doesn't yet show in detail what goes into the machine-specific script itself, such as which politeness habits survive and which disappear. That remains an open area.