Beyond the Chat BoxEssay 07 / 10
Attention and ambient computing
ESSAY 07

Ambient AI Must Earn the Right to Interrupt

An assistant should protect attention as carefully as it provides information.

An assistant that is always available can become an assistant that is always interrupting. Ambient intelligence promises help in the background: noticing a missed step, surfacing a reminder, or preparing information before it is requested. Its value depends not only on what the system notices, but whether it chooses the right moment to speak. Attention is scarce. Every interruption asks a person to leave one task and evaluate another.

Relevance is not enough

A meeting reminder may be useful but disruptive while someone is presenting. A coding suggestion may be correct but poorly timed while a developer traces a difficult bug. A good system needs a model of interruption cost, not simply topic relevance. It should consider urgency, the chosen focus state, and whether information can safely wait.

A consent-aware interruption policy

Users could choose quiet periods, categories of events that may interrupt, urgency thresholds, preferred channels, and whether suggestions should be queued rather than pushed. The assistant should explain why it interrupted and make “not now” distinct from “never.” Accepting one notification should not imply blanket permission for future interruptions.

Sensing should be legible

An ambient system may infer activity from microphones, cameras, calendars, or device state. It should explain what it senses, how that context is used, and how to pause it. A suggestion should not imply certainty. The user should be able to inspect the trigger and correct a mistaken assumption.

Where this can fail

A system can misread activity, expose private information on a shared display, or create anxiety through constant monitoring. It can also train people to expect alerts for every small decision. The design should measure interruption burden and task switching, not only notification engagement.

Questions for research

  1. How do people trade off interruption timing against information value?
  2. Can lightweight feedback teach a useful interruption policy?
  3. What makes a suggestion supportive rather than controlling?
  4. How does ambient assistance affect task switching and autonomy over time?

Design hypothesis

Ambient AI earns trust by protecting attention. Its most intelligent action may sometimes be to wait.

Research starting points

  1. McFarlane (2002), Comparison of Four Primary Methods for Coordinating the Interruption of People in Human-Computer Interaction, Human–Computer Interaction 17(1), 63–139.
  2. Iqbal & Bailey (2008), Effects of Intelligent Notification Management on Users and Their Tasks, CHI 2008, 93–102.
  3. Amershi et al. (2019), CHI. https://doi.org/10.1145/3290605.3300233
← Back to all ten essays