The courtesy of a clear exit

I picture an open door beside a quiet screen. I want an assistant to make leaving feel that ordinary. When I say stop, the answer should end at once. When I dismiss a suggestion, it should stay out of my way. I should be able to leave a task unfinished without having to justify the choice.

The researchers behind Guidelines for Human-AI Interaction urge designers to make unwanted AI services easy to dismiss or ignore. They also call for easy correction when a system is wrong. I hear a useful distinction in those two ideas. Sometimes I want another attempt. Sometimes I want my attention back.

A field study of families using a voice assistant describes people repeating or rephrasing requests after communication broke down. Some abandoned the request. The authors write that the burden of repairing a breakdown generally falls to the users. I take a lesson from those accounts: when a conversation starts costing more than it gives, the exit should be easy.

I use a simple test. Interrupt an answer halfway through. Dismiss a suggestion. Close the conversation before the task is done. A visible stop control should be there before I need it. Count the taps and extra words needed to make it quiet. Check whether it resumes without an invitation. If something already happened, look for a plain account of what happened.

A good assistant can help me continue. A courteous one also lets me leave without a second request.