> In the year ending March 2026, more than 6,400 migrants claiming to be children were age assessed at the border, with 43% found to be adults, according to Home Office data.
Whatever method the border force used to determine this, I cannot imagine how AI is going to be more accurate.
It's less about accuracy and more about who gets the blame when mistakes happen. And computers are excellent scapegoats for this kind of situations because when mistakes happen, you can just throw your hands up and say, well, the computer sucks, and no actual human will bear any consequences. It's not like the computer is going to fight back or anything.
Since this is the UK, see also the entire post office scandal. It was often blamed on a faulty accounting software developed by Fujitsu, even though the software put no one in jail. Human prosecutors did. But of course the British state apparatus will not admit to that, so all we hear is this story about software errors that conveniently ignores any human involvement in the process.
Actually the humans come up all the time. The problem was that as always somehow highly paid executives conveniently don't know anything and had no idea anything was happening. Several Post Office executives testified that they had no idea they were ordering people to be prosecuted, they were just completely incompetent and signed whatever was put in front of them, while being paid a huge sum of money. If they received specific documentation telling them Horizon was busted and mustn't be relied on they mislaid it, and oops, forgot to take any action as a result.
Likewise politicians supervising those executives somehow conveniently didn't ask any questions, forgot what they'd been told and generally had no idea what was happening.
Some of the crimes so often committed by executives who walk free needs to delete Mens Rea so that when executives say they had no idea the prosecutor doesn't even skip a beat because it doesn't matter. For comparison in UK criminal law if you have sex with a ten year old, and you try to argue† you thought they were of age and also you had their consent, the prosecutors will move on because you haven't actually defended yourself at all, sex with the ten year old was rape by definition, the fact is the crime, what you believed about it was irrelevant, you're done.
† If you have defence counsel they'll strongly urge you not to try this because it can't work
Of course I don't believe them. But the criminal standard isn't "balance of evidence" it is "beyond a reasonable doubt" and I can't say that I have no doubt she's lying.
Eliminating Mens Rea would solve the problem. If they don't want to go to jail they can try being more competent, or, I expect, they can try not being crooks and what do you know all the crime they supposedly "weren't responsible for" magically stops. Huh.
The process, per the article, is that a border agent makes a determination, and if it is not what the applicant claims, then a social worker takes over to make a final determination.
That is a massive time sink for social workers, and the appeal of having an automated system is pretty obvious. Considering that it is already all largely guesswork, I'm not really sure that "more accurate" is even an acceptance criteria for them right now- they'd probably be very happy with "mostly the same accuracy".
Of course, the social workers are opposing being taken out of the loop, but I can't imagine that there isn't already plenty of work for them elsewhere in the UK.
"a computer can never be held accountable, therefore a computer must never make a [legal] decision"
in my observation: when humans are automated out of a process due to the human element being inconvenient, the perceived efficiency gains are often because wronged individuals have less recourse in the automated system.
There is a widespread belief (on which I am personally undecided and not interested in attempts to sway me one way or the other) in the UK that asylum seekers already have too much access to recourse. Making it easier and cheaper to reject asylum applications within the requirements of international human rights law is probably an implicit aim in this project.
> In another example, a Vietnamese national was initially given the benefit of the doubt at the first triage that took place in the waiting area. The CIO and social worker commented on his “soft face”, which they said was consistent with his claimed age of 17. However, his “developed shoulders” and “huge hands” cast doubt for them, as did a “tiny bit of stubble” that they noticed when they asked him to raise his chin. The CIO and social worker told inspectors afterwards that Vietnamese young people were typically difficult to assess because they “did not have the same ageing process”, and “did not show signs of ageing”. When asked where the evidence for this was, they said that it was knowledge gained through their own experience. The social worker said, “It is just genetics”, but was unable to support this with evidence.
If I had to choose between being judged by an AI model and being judged based on ad hoc stereotypes of what my race's shoulders and hands typically look like, I'd definitely pick the AI.
I understand that "no" isn't the answer you're looking for but I'm not sure what else to say in response. A computer system is the opposite of an ad hoc stereotype; it can be directly tested for problems and those problems can be corrected if found.
It's a problem when people use this kind of system to circumvent the question of "do we have to make this judgment at all". We shouldn't, for example, predict from someone's photo how likely they are to commit a crime, so we're rightly skeptical of people who try to argue about system X or system Y might better predict it.
But as the source article covers, the UK's asylum laws require it to make this age judgment, because child migrants are entitled to special programs separate from adult migrants on account of their vulnerable status.
I am. I would expect an age classification model to be trained on a dataset containing pictures of people with known ages and their known age.
It's true that the model might develop the same strange belief that large hands prove a Vietnamese person is not 17, if the training data is biased in that way. There's no perfect solution.
Purely curious if there's better ways, like I know X Rays can get us pretty close for children and adolescents, but that might not be the best way to do it, I wonder if there's other alternatives that are low cost.
I think the only plausible argument for AI here is not "it knows age better than humans," but "it might be more consistent than ad hoc visual judgments by different officers"
it depends, i'll concede that AI is unfortunately incredibly good at pattern matching, if you give it 100 billion pictures of a 13 year old, and 100 billion pictures of a "not 13 year old", you'd be surprised how accurate the pattern matching can get
Why can’t it be more accurate? Sure just doing it on facial features will have some limit on accuracy but it may be less biased and more accurate than human judgment (not always but it’s possible). I think you could then confirm any finding of falsified age using more expensive techniques - like analyzing medical imaging, dental records in particular. That can be the “confirmation” step to improve the overall accuracy of the process.
Twenty years ago, I spent a brief period observing Chicagoland's immigrant halfway houses.
The majority of the "juvenile" males were adults claiming such, and the detention facilities couldn't really do anything more than let thirty year olds co-mingle with actual children (e.g: actual teenagers don't have full dentition (already with years of decay on their third molars).
Definitely didn't seem right, and we definitely didn't let females (of any age) alone inside.
The point of this is unambiguously to use technology as an accountability sink. You can't have a human eyeball age and call that a "process." You want a machine to point to instead. Its accuracy or lack thereof is immaterial to anyone who would seriously suggest using it.
As ever, this is the real risk of "AI"; not the technology itself so much as the technology-as-social-construct. A machine oracle we can abdicate decisions to with a facade of neutrality.
In this case, the facade is painting over the underlying motivation which is to reject asylum claims. You could imagine a world in which it is instead used to scan and fast-track claims through an automated and unaccountable process, but the form of the deployment has baked-in the outcome and interests of the powerful. Don't be surprised if there's another automated AI system that totally-pinky-promise-for-sure validates that rich tourists aren't terrorists so they can walk through security unmolested and another system that uses AI to flag "suspicious behavior" for the proles. The outcome is baked in, the AI just provides plausibility and legitimacy.
>the facade is painting over the underlying motivation which is to reject asylum claims.
That feels like an unfair read. Asylum seekers claiming to be younger than they are is a known reality, and it makes total sense that a system would need to guard against that. Are you suggesting we just take every person who shows up at the border at their word?
> Are you suggesting we just take every person who shows up at the border at their word?
That's not what I said. It's all about the framing. Rejecting is automated, accepting is slow and deliberate. This is reflective of a preference for rejection but presented as neutral and pragmatic. If you cared about false negatives as much as false positives you wouldn't fast-track or preference either branch. The observation that this is the case is value-neutral.
As for my personal opinion, I consider freedom of movement a human right and accepting asylum seekers a moral obligation. Both can be implemented poorly, but it's our duty to try to do them right and allocate these systems the resources they need to succeed and to hold incompetent or malicious leaders responsible for failing to operate them effectively.
How would you train a system like this? Presumably you are going to need quite a few subjects of each ethnicity. Where are you going to find these people? You can't use asylum seekers, as you can't be sure about their real age.
I'm sure the AI is not biased and has a diverse training data set accurately representing the war-torn and impoverished countries these asylum seekers are coming from.
Reminds me of the story of this dentist in Sweden who was fired for raising concerns about how 80% of migrant children he was treating were actually adults, based on their teeth:
The funnier part is that people are being let it without the gov crosschecking their identity with the government they are from and just being like "yeah you look ok, come in".
> In the year ending March 2026, more than 6,400 migrants claiming to be children were age assessed at the border, with 43% found to be adults, according to Home Office data.
Whatever method the border force used to determine this, I cannot imagine how AI is going to be more accurate.
It's less about accuracy and more about who gets the blame when mistakes happen. And computers are excellent scapegoats for this kind of situations because when mistakes happen, you can just throw your hands up and say, well, the computer sucks, and no actual human will bear any consequences. It's not like the computer is going to fight back or anything.
Since this is the UK, see also the entire post office scandal. It was often blamed on a faulty accounting software developed by Fujitsu, even though the software put no one in jail. Human prosecutors did. But of course the British state apparatus will not admit to that, so all we hear is this story about software errors that conveniently ignores any human involvement in the process.
Actually the humans come up all the time. The problem was that as always somehow highly paid executives conveniently don't know anything and had no idea anything was happening. Several Post Office executives testified that they had no idea they were ordering people to be prosecuted, they were just completely incompetent and signed whatever was put in front of them, while being paid a huge sum of money. If they received specific documentation telling them Horizon was busted and mustn't be relied on they mislaid it, and oops, forgot to take any action as a result.
Likewise politicians supervising those executives somehow conveniently didn't ask any questions, forgot what they'd been told and generally had no idea what was happening.
Some of the crimes so often committed by executives who walk free needs to delete Mens Rea so that when executives say they had no idea the prosecutor doesn't even skip a beat because it doesn't matter. For comparison in UK criminal law if you have sex with a ten year old, and you try to argue† you thought they were of age and also you had their consent, the prosecutors will move on because you haven't actually defended yourself at all, sex with the ten year old was rape by definition, the fact is the crime, what you believed about it was irrelevant, you're done.
† If you have defence counsel they'll strongly urge you not to try this because it can't work
And you believe it?
Of course I don't believe them. But the criminal standard isn't "balance of evidence" it is "beyond a reasonable doubt" and I can't say that I have no doubt she's lying.
Eliminating Mens Rea would solve the problem. If they don't want to go to jail they can try being more competent, or, I expect, they can try not being crooks and what do you know all the crime they supposedly "weren't responsible for" magically stops. Huh.
The process, per the article, is that a border agent makes a determination, and if it is not what the applicant claims, then a social worker takes over to make a final determination.
That is a massive time sink for social workers, and the appeal of having an automated system is pretty obvious. Considering that it is already all largely guesswork, I'm not really sure that "more accurate" is even an acceptance criteria for them right now- they'd probably be very happy with "mostly the same accuracy".
Of course, the social workers are opposing being taken out of the loop, but I can't imagine that there isn't already plenty of work for them elsewhere in the UK.
"a computer can never be held accountable, therefore a computer must never make a [legal] decision"
in my observation: when humans are automated out of a process due to the human element being inconvenient, the perceived efficiency gains are often because wronged individuals have less recourse in the automated system.
There is a widespread belief (on which I am personally undecided and not interested in attempts to sway me one way or the other) in the UK that asylum seekers already have too much access to recourse. Making it easier and cheaper to reject asylum applications within the requirements of international human rights law is probably an implicit aim in this project.
When you put the computer in charge, it tends to say no
https://www.youtube.com/watch?v=x0YGZPycMEU
It might be helpful to be a bit more specific about what's being automated. One story from the underlying report (https://assets.publishing.service.gov.uk/media/687f6f9dfdc19...):
> In another example, a Vietnamese national was initially given the benefit of the doubt at the first triage that took place in the waiting area. The CIO and social worker commented on his “soft face”, which they said was consistent with his claimed age of 17. However, his “developed shoulders” and “huge hands” cast doubt for them, as did a “tiny bit of stubble” that they noticed when they asked him to raise his chin. The CIO and social worker told inspectors afterwards that Vietnamese young people were typically difficult to assess because they “did not have the same ageing process”, and “did not show signs of ageing”. When asked where the evidence for this was, they said that it was knowledge gained through their own experience. The social worker said, “It is just genetics”, but was unable to support this with evidence.
If I had to choose between being judged by an AI model and being judged based on ad hoc stereotypes of what my race's shoulders and hands typically look like, I'd definitely pick the AI.
Isn't the AI going to use the same ad hoc stereotypes?
I understand that "no" isn't the answer you're looking for but I'm not sure what else to say in response. A computer system is the opposite of an ad hoc stereotype; it can be directly tested for problems and those problems can be corrected if found.
It's a problem when people use this kind of system to circumvent the question of "do we have to make this judgment at all". We shouldn't, for example, predict from someone's photo how likely they are to commit a crime, so we're rightly skeptical of people who try to argue about system X or system Y might better predict it.
But as the source article covers, the UK's asylum laws require it to make this age judgment, because child migrants are entitled to special programs separate from adult migrants on account of their vulnerable status.
Are you familiar with how AI gets trained?
I am. I would expect an age classification model to be trained on a dataset containing pictures of people with known ages and their known age.
It's true that the model might develop the same strange belief that large hands prove a Vietnamese person is not 17, if the training data is biased in that way. There's no perfect solution.
Purely curious if there's better ways, like I know X Rays can get us pretty close for children and adolescents, but that might not be the best way to do it, I wonder if there's other alternatives that are low cost.
I think the only plausible argument for AI here is not "it knows age better than humans," but "it might be more consistent than ad hoc visual judgments by different officers"
Fixed it for you:
I cannot imagine how AI is going to be more INaccurate.
it depends, i'll concede that AI is unfortunately incredibly good at pattern matching, if you give it 100 billion pictures of a 13 year old, and 100 billion pictures of a "not 13 year old", you'd be surprised how accurate the pattern matching can get
6,400 is maybe a tiny fraction of the total. Maybe AI will allow them to have way more breadth?
Why can’t it be more accurate? Sure just doing it on facial features will have some limit on accuracy but it may be less biased and more accurate than human judgment (not always but it’s possible). I think you could then confirm any finding of falsified age using more expensive techniques - like analyzing medical imaging, dental records in particular. That can be the “confirmation” step to improve the overall accuracy of the process.
Twenty years ago, I spent a brief period observing Chicagoland's immigrant halfway houses.
The majority of the "juvenile" males were adults claiming such, and the detention facilities couldn't really do anything more than let thirty year olds co-mingle with actual children (e.g: actual teenagers don't have full dentition (already with years of decay on their third molars).
Definitely didn't seem right, and we definitely didn't let females (of any age) alone inside.
The point of this is unambiguously to use technology as an accountability sink. You can't have a human eyeball age and call that a "process." You want a machine to point to instead. Its accuracy or lack thereof is immaterial to anyone who would seriously suggest using it.
As ever, this is the real risk of "AI"; not the technology itself so much as the technology-as-social-construct. A machine oracle we can abdicate decisions to with a facade of neutrality.
In this case, the facade is painting over the underlying motivation which is to reject asylum claims. You could imagine a world in which it is instead used to scan and fast-track claims through an automated and unaccountable process, but the form of the deployment has baked-in the outcome and interests of the powerful. Don't be surprised if there's another automated AI system that totally-pinky-promise-for-sure validates that rich tourists aren't terrorists so they can walk through security unmolested and another system that uses AI to flag "suspicious behavior" for the proles. The outcome is baked in, the AI just provides plausibility and legitimacy.
> The point of this is unambiguously to use technology as an accountability sink.
Seems like another measuring device, like a breathalyzer or radar gun, and should be held in court to the same (hopefully high) standards.
>the facade is painting over the underlying motivation which is to reject asylum claims.
That feels like an unfair read. Asylum seekers claiming to be younger than they are is a known reality, and it makes total sense that a system would need to guard against that. Are you suggesting we just take every person who shows up at the border at their word?
> Are you suggesting we just take every person who shows up at the border at their word?
That's not what I said. It's all about the framing. Rejecting is automated, accepting is slow and deliberate. This is reflective of a preference for rejection but presented as neutral and pragmatic. If you cared about false negatives as much as false positives you wouldn't fast-track or preference either branch. The observation that this is the case is value-neutral.
As for my personal opinion, I consider freedom of movement a human right and accepting asylum seekers a moral obligation. Both can be implemented poorly, but it's our duty to try to do them right and allocate these systems the resources they need to succeed and to hold incompetent or malicious leaders responsible for failing to operate them effectively.
I think "accountability sink" is the right phrase here
How would you train a system like this? Presumably you are going to need quite a few subjects of each ethnicity. Where are you going to find these people? You can't use asylum seekers, as you can't be sure about their real age.
Australia has a torrid history of misjudged age by wrist bone tests (fused? Infused?)
Massive payouts for misstatements attributing adulthood to kids from illegal Indonesian fishing boats and jailing them in adult prisons.
I'm sure that won't be misused at all.
I'm sure the AI is not biased and has a diverse training data set accurately representing the war-torn and impoverished countries these asylum seekers are coming from.
Reminds me of the story of this dentist in Sweden who was fired for raising concerns about how 80% of migrant children he was treating were actually adults, based on their teeth:
https://www.dental-tribune.com/news/dental-hygienist-fired-f...
It is similar, in the sense that pseudoscience is being used to potentially ruin people’s lives.
Nothing magically happens to teeth when someone turns 18.
Wisdom teeth develop during teen years and may erupt as early as 16, as late as 25, or not at all.
This website never ceases to downvote easily verifiable facts.
The only wishful thinking here is assuming people won't lie about their age and then you end up with "children" competing in school-level competitions
> may erupt as early as 16, as late as 25
Cool so assuming the average age to be 20.5 this means someone with wisdom teeth is more likely to be of age than not.
Put this fact together with other factors like wrist x-ray estimation and obviously looking like an adult and maybe we can have some common sense
I didn’t say there was no age fraud. I just said this method of determining age is nonsense, which it is.
So out of 100 claimed children, 40 of them are actually children and you'd want to reject them because of the law of big numbers?
The funnier part is that people are being let it without the gov crosschecking their identity with the government they are from and just being like "yeah you look ok, come in".
This is a long-solved problem and can be sone cheaply without AI: https://pmc.ncbi.nlm.nih.gov/articles/PMC11122822/
So much time, money, energy wasted on being cruel. It’s shameful.
[dead]