
AI in family care should reduce noise first: combine repeated signals, expose uncertainty and route relevant change to the right person.
It should not replace clinicians or family decisions, and it should not collect unrelated data simply to appear more intelligent.
Organise, explain and connect before making decisions
- 01
Find change: compare with the person’s routine rather than a universal threshold.
- 02
Explain: show the signal behind an alert and distinguish risk from missing data or an offline device.
- 03
Prioritise: put routine status in summaries and escalate only what needs action.
- 04
Connect: send information to someone able to respond and record what happened next.
High-consequence decisions require people
Diagnosing disease, deciding whether emergency care is required, consenting to data use for someone else, or giving a confident answer with insufficient information should not be automated by a consumer system. Outputs should say whether they are observations, inferences or confirmed facts.
More output is not the same as more help
Useful AI reduces repeated checking, irrelevant alerts and app switching. If family members spend more time verifying the system, errors remain unexplained or the older person feels watched, the feature has not completed its job.
A list of data not collected can be more important than another feature
Product definition can exclude unrelated conversation, exact long-term movement histories and raw audio or video unless a task truly requires them. Model features, family status and support logs should have separate controls.
Model updates may alter alerts. Families should see which functions changed, reconfirm important permissions and be able to return to a more conservative setting.
BEIIU keeps an AI feature only when it reduces repeated checking by relatives, unnecessary interruption for the older person or avoidable confirmation work for staff. A new summary that changes no next action should not occupy the centre of the experience.
Sources
- 01
- 02
- 03
- 04
