A coalition of more than a hundred refugee children’s organisations has said controversial plans to use AI to assess the age of young asylum seekers could lead to more children wrongly ending up in adult prisons or detention centres.
The warning follows a Home Office announcement on Friday of a contract to roll out AI facial age estimation technology on young asylum seekers whose age is disputed.
A report from the Refugee and Migrant Children’s Consortium seen by the Guardian before its release in June raises the alarm about the risks of using such technology on young people who do not fit the norms of others in their age group.
The consortium, whose member organisations work to promote and protect the rights of refugee and migrant children, say that as a result of trauma, under-nutrition and the harrowing journeys the young people have undertaken to reach safety means AI assessment is complex.
The report – Benchmarks and Borders: the use of facial age estimation to assess the age of unaccompanied young people seeking asylum – does not rule out the use of AI altogether, but cautions against relying on it and says it should not be a substitute for comprehensive age assessments carried out by social workers.
It urges the Home Office to use AI in an advisory rather than determinative capacity, with a range of safeguards built in including access to an appropriate adult, legal advice and a right to challenge decisions.
It urges government not to replace the human errors made in some age–assessment cases with machine error.
Age assessments of this group of young people is complex, especially as the majority of lone child asylum seekers coming to the UK are aged 16 or 17. According to Home Office data, young asylum seekers are more than twice as likely to be recorded as children in assessments by social workers than those carried out by immigration officers at the border, with more than two-thirds assessed to be minors.
The Home Office’s announcement places significant emphasis on adults making “fake claims”, pretending to be children and attempting to “game the system”, but it does also acknowledge the need to safeguard minors.
The minister for border security and asylum, Alex Norris, said: “For too long, adult migrants making false age claims have exploited the system and diverted vital support away from children at risk.
“That is why we are rolling out AI technology to put a stop to this, ensuring those who game the system are identified, detained and removed without delay, and those who deserve support and protection are given it.”
Final decisions will continue to be made by immigration officers, and the Home Office said the technology would undergo rigorous testing, evaluation and assurance before national rollout.
The co-chair of the Refugee and Migrant Children’s Consortium, Kamena Dorling, said: “The government’s proposals are deeply concerning. AI cannot account for the factors that can significantly affect a young person’s appearance after fleeing conflict and persecution and undertaking dangerous journeys, including trauma, malnutrition, and exhaustion.
“Existing evidence also shows that AI faces the same problems with bias and inaccuracy as human decision-making, with similar patterns of errors.”
Kama Petruczenko, a senior policy analyst at the Refugee Council and member of the consortium, said: “The government’s own figures already show that hundreds of children are being wrongly treated as adults following flawed visual assessments at the border, with devastating consequences for their safety and wellbeing.
“AI and facial age estimation technology are not a simple or risk-free answer to these longstanding problems. Poor image quality and bias in datasets can also affect accuracy.
“There is a real danger that this technology creates a false sense of certainty in decisions that are already extremely difficult to get right. If flawed assessments are simply automated, more children could end up wrongly placed in adult accommodation, detention centres or even prisons.”
The Home Office says AI will estimate an individual’s age in seconds by analysing facial photographs already taken of small-boat arrivals at Dover. A contract for the work worth £322,000 over three years has been awarded to Akhter Computers Ltd, under which the technology will be further tested and developed before being rolled out in 2027.
