Face-scanning technology tests could only guess age within an 18-month range in 85 percent of cases. Photo: Supplied/ABC
Children as young as 15 were repeatedly misidentified as being in their 20s and 30s during Government tests of age-checking tools, sowing new doubts over whether the teen social media ban is viable.
ABC News can reveal that face-scanning technology tested on school students this year could only guess their age within an 18-month range in 85 percent of cases.
"It's definitely a problem," said Andrew Hammond, general manager of software consultancy firm KJR, which was tasked with running the trial.
"So far, it's not perfect and it's not getting every child, but does that mean it's no good at all?"
The full results of the age assurance technology trial were not expected to be released until later this year, but preliminary data had experts worried.
"I don't think the ban is viable," said RMIT University information services professor Lisa Given, who had closely analysed the Government's policy.
"Parents are definitely headed for a rude shock, in terms of what this legislation will actually deliver to them."
Under the social media ban, more than 20 million Australians will be required to demonstrate they are 16 or older to log in to most major social media platforms.
The ban is due to take effect in December, but the Government has yet to decide how it will be implemented, amid ongoing questions over whether age-checking technology is up to the job.
The Government's technology trial, which has been running for eight months, was meant to provide some answers, but Professor Given said the public may be disappointed.
"The accuracy level at 85 [percent] is actually quite low and an 18-month range is significant, when you're trying to identify a very particular age grouping," she said.
"We are going to see a messy situation emerging immediately, where people will have what they call false positives, false negatives."
Some students at Canberra's John Paul College, who previewed the technology as part of the Government's trial, were surprised, when their results were up to decades off the mark.
Sixteen-year-old Andy was misidentified as 19, 37, 26, and 23 years old by various face-scanning tools he used.
"I don't think the technology is ready yet to become a full-fledged primary defence system," he said. "It's pretty inconsistent."
Seventeen-year-old Beth was given results ranging from 14-32.
"I usually get told by other people that I don't look 17, I look older, so when it says 14, I thought… that's interesting."
Her results from the other end of the spectrum were unwelcome for different reasons.
"It's a bit insulting, because that's how old my aunty is," she said. "I don't want to look 32 just yet."
Seventeen-year-old Nomi was especially concerned, when one tool mistook her for a 13-year-old.
"I'm almost 18," she said. "If I try to sign up to an app and it tells me 'you're not meeting an age requirement', even though I am, that would be a problem for me."
Ban should not depend on face scanning, says industry
While the face-scanning results from the trial may not seem promising, Hammond said he was confident the ban would still work, because it did not rely exclusively on that tech.
"If the solution to implementing the legislation was just facial age estimation, I'd say, 'Yep, it's probably not good enough'," he said.
"However, it's just one of the tools in the toolkit that could be used."
Age-verification providers are not discouraged by the early results either, arguing that other tech was always going to be necessary as a complement to get precise results.
"You would never rely on age estimation for people who are literally at the age of 16," said Iain Corby from the Age Verification Providers Association, the industry body for age-check companies.
"It was never going to be good enough for that," he said.
One tool mistook Beth, 17, as being 32 years old. Photo: ABC News
Corby said the early data reported by ABC News, showing an accuracy rate within 18 months for only 85 percent of students, is roughly what he expected.
"I think even the best-in-class achieves about a year and a month, on average, above or below your real age."
Among the methods tested were other age-estimation techniques that rely on biological traits, such as voice and hand movements, to guess the age of a user, but those methods struggled with the same accuracy issues and fewer companies offered the service.
Another avenue was guessing a person's age based on their online activity, but that was also imprecise.
Other tools offer a higher degree of certainty by inferring or even verifying a user's age, using data provided by third parties, such as banks, schools or healthcare providers.
The strongest proof is a overnment-issued ID, such as a passport or a driver's license, but the legislation prevented social media companies from insisting on it.
A last-minute amendment to the Bill, when it was passed back in November, meant platforms would be forced to offer users alternative methods to prove their age.
That rule meant many Australians who could not easily provide those more reliable proofs might be forced to rely on less accurate methods, such as face scanning, if they wanted to use social media.
"We do know, generally, that young people are going to be less likely to have a Government-issued ID that would satisfy some form of age verification," Given said.
If facial scanning was on offer, under-16s who wanted to dodge the ban might be tempted to choose it anyway, in the hope they could fool it.
"They might put glasses on, they might put make-up on, different hairstyle, different lighting, just to see if the system is actually able to accurately see that they're underage or over 16," Given said.
How the ban could solve face-scanning inaccuracy
The Government was expected to decide how the ban would work in the coming months, but one possible solution for the shakiness of facial scanning was a cascade-style system, similar to what we've see in bottle shops.
Users might use face-scanning tech as a first hurdle and only be asked for further proof, if their result was within a 5-10-year margin of 16.
"If you're within that margin for error, then you have to go to a second stage and find some other way of confirming that somebody is over the legal age," Corby said.
Even so, everyone agreed it would not be perfect.
"I'm optimistic, having seen the results," Hammond said. "Not necessarily making sure every 16-year-old doesn't get access, but making sure that most 16-year-olds don't get access to social media.
"There's a number of solutions… and they have a level of accuracy. Now, whether the accuracy is good enough is a different question."
Professor Given saw the end of the tech trial as an opportunity to reconsider the ban.
"A responsible decision from Government would be weighing up the evidence in front of them and deciding whether that's actually a robust approach," she said.
In the meantime, public expectations of the policy remained undeterred.
"I think it's a really positive move for our young people," said John Paul College principal Craig Wattam. "I think that limiting their exposure to places that are potentially really dangerous is a really liberating thing."
On the question of the tech's accuracy, he was also optimistic.
"I guess this is the whole purpose of a trial," he said. "I'm confident that by the time we get closer to December… they may well have figured out more accurate ways to verify students' ages."
A spokesperson for Communications Minister Anika Wells told ABC News the Government would be guided by advice from the eSafety Commissioner on how best to implement the ban.
"We know that social media age-restrictions will not be the end-all, be-all solution for harms experienced by young people online, but it's a step in the right direction to keep our kids safer," they said.
- ABC
