Forcing entry to the centre is an interruption
Aliases: interruption cost · resumption lag · annoyance
What it is
When ambient information bypasses the user's current activity and directly demands centre attention and response, it stops being ambient information and becomes an interruption. The criterion is timing and form, not content: the same message delivered at a task boundary in a dismissible form is a service; delivered mid-action in a must-handle form, it is an interruption.
Note that "interruption" is not a synonym for "bad notification". Some information has every reason to enter the centre immediately (a smoke alarm). This card targets non-urgent information arriving with urgent force — the characteristic disease of ambient systems, because they run around the clock, come in numbers, and each stream believes itself "relevant to you".
Why it happens
The cost of an interruption has three components:
- The switch itself. Attention moves from the current task to the interruptor and back; there is a task-switch cost, and the interrupted task's goal state begins to decay, so returning requires reconstruction — visible as resumption lag, the seconds spent after return re-establishing "where was I".
- Timing loss. Interruptions at different moments within a task cost differently: those landing at subtask boundaries are markedly cheaper to recover from than those landing mid-action, where the user must additionally hold a half-finished state in memory.
- Emotional amplification. Being interrupted produces annoyance, which itself degrades performance and interacts with the first two components: the more irritated the user, the slower the reconstruction. In experiments, annoyance scales often prove as sensitive as performance measures.
Ambient systems amplify all three: many devices, always on, each independently judging its own relevance — interruptions individually defensible sum into a standing tax on attention. That is precisely the opposite of calm.
Studying it
- McFarlane (2002, TOCHI) compared four coordination strategies for delivering information to the centre — immediate, negotiated, scheduled, mediated. No global winner, but negotiated interruption (user may defer) was robust across most desktop tasks, establishing that giving users a say over when something enters the centre is a design conclusion, not a fixed property.
- Adamczyk and Bailey (CHI 2004) manipulated the moment of interruption within task execution (subtask boundary versus mid-action) and found boundary interruptions produced faster resumption and lower annoyance — turning "timing" from intuition into a measurable variable. Subsequent work by Iqbal and Bailey used task progress to defer notifications to boundary points.
- Measures: total time and errors on the main task, resumption lag (time from interruption end to first effective action), and annoyance and workload scales (NASA-TLX is common).
One methodological caution: this literature comes almost entirely from desktop office tasks — clearly structured activities, seated users, single screens. Ambient computing involves fragmentary, situated, often mobile activity; subtask-boundary detection and deferral strategies do not transfer cleanly and need revalidation in situ.
Where it stops holding
- Urgent information is excepted. Safety alerts must enter the centre immediately and be impossible to ignore; the interruption cost is justified by the consequence. The design requirement there is "cannot be missed" — the opposite direction from "don't force entry". Classify first, then handle timing.
- Information the user is actively awaiting is excepted. A timer the user set, an incoming call they expect — that is the fulfilment of pull, not an interruption. But "awaiting" is defined by the user; the system may not widen "you actually need to know" on the user's behalf.
- Desktop findings do not transfer directly. See the caution above: ambient task structure differs, and boundary detection and deferral need recalibration.
Applying it
- Never let non-urgent information enter the centre directly: route it through a peripheral transient first; only escalate — still dismissibly — if there is no response within a window and relevance still holds.
- Defer to boundaries: wherever natural gaps in activity are detectable (a completed action, an ended conversation, leaving the room), defer the escalation to the gap; where they are not detectable, lower the form and let it be seen at the user's convenience.
- Grant negotiation: anything elevated to the centre must be dismissible and snoozable, and dismissals should be recorded to adjust that stream's escalation policy.
- How to check: log every forced centre entry with its timing and the user's activity at that moment; label whether it fell in a natural gap; sample-interview the after-the-fact value of these messages. The proportion landing in gaps rising month over month, with "worth the interruption" judgements not falling, means the policy is converging.