How AI is used in looking after someone from abroad, and what it should never do

What the evidence says

• The largest randomised trial so far (981 people, four weeks, 300,000 messages, run by MIT and OpenAI, 2025) found no effect of chatbot use on loneliness overall, and that heavier use went with more emotional dependence and worse wellbeing. Trust in the chatbot predicted dependence.
• A survey of 1,006 student users of a companion app found 63% reported at least one positive outcome, but the sample was self-selected and 90% were lonely to begin with.
• Products marketed to older people report strong results in their own research (one companion robot claims 96% of users report improved wellbeing), and a Japanese therapeutic seal robot has been used in dementia care since 2003. Independent evaluations are thin. Treat vendor figures as vendor figures.

What regulators require

• The World Health Organization's 2024 guidance on large multimodal models in health warns of false, inaccurate, biased or incomplete outputs and of automation bias, and asks governments to require independent audits and developers to build for well-defined tasks with demonstrated accuracy and with patients and clinicians involved from the start.
• In the UK, the MHRA regulates software as a medical device by its intended purpose. A reminder app or a general wellbeing tool sits outside; software intended to diagnose, monitor or treat a condition sits inside and must be registered. The line for a check-in product is whether it starts interpreting a condition.
• In Nigeria, the Nigeria Data Protection Act 2023 gives people the right to access, correct and erase their data, to object to processing, and to be protected from decisions made solely by automated means, enforced by the Nigeria Data Protection Commission. Health data is sensitive data under it.

Where AI actually helps a family abroad

• Answering the question at 2am. What does this reading mean, what did the nurse say last month, what should we ask at the appointment tomorrow. A model reading the member's own record answers better than a search engine reading nobody's.
• Keeping one record. Visit updates, prescriptions, readings and hospital letters in one place, in the member's name, readable by everyone the member has invited, in any time zone.
• Noticing. Three visits in a row where blood pressure drifted up, a medicine that was in the cupboard in June and not in August, a mood the nurse described the same way twice. A pattern across updates is what a family reading each one on its own cannot see.
• Translating. Between the clinical note and the family, and between the family's English and a parent's Yoruba or Igbo, so that the same information reaches everyone in words they use.

The lines Kinship draws

• Kemi, the AI in the Kinship app, answers questions and helps the family and the member find their way around the record. Kemi does not diagnose and does not prescribe; the doctor the member talks to under Kinship Well is a doctor.
• The visit is done by a qualified nurse, in the house, and the update is written by that nurse. Kemi helps people read it; Kemi did not write it.
• Every visit update is reviewed by the member's named Kinkeeper, a person, before the family relies on it. Anything that needs a decision goes to a person.
• The member controls the data. What Kemi can see is what the member has agreed to share, and the member can see and change that at any time, which is the same right the Nigerian Act and UK law give them.
• Nothing is designed to keep someone talking to Kemi. The measure of success is a family that needs the app less, because the person in Nigeria is better looked after.

Questions people ask

Related

• What is a Kinkeeper?
• Keeping one health record for a relative abroad