AI-powered age verification methods have raised questions over potential bias and privacy impacts. — AFP
LONDON: As governments crack down on online platforms from social networks to porn sites, business is booming for one sector offering AI age checks based on selfies.
The technology promises speed and accuracy in implementing laws like Australia's social media ban for under-16s – in force from December 10.
As a user, verifying your age with the technology is a breeze.
All it requires is taking a head-on selfie with a phone or computer camera, and an automated decision is made within a minute.
"We estimated your age is 18 or older," read a pop-up window on the Roblox gaming site.
At startup Yoti's spacious London office, mannequin heads – some fitted with wigs or masks – are lined up along one of the windowsills for testing.
Their AI is not fooled by them. "We can't be sure that the image was of a real face," said one test platform installed on a smartphone.
Over time, "the algorithm got very good at looking at patterns and working out, 'this face with these patterns looks like a 17-year-old or a 28-year-old'... it just got very good at estimating age", said CEO Robin Tombs.
Data privacy fears
Today, the startup runs around one million age checks a day, including for big-name clients like Meta, TikTok, Sony and Pinterest, Tombs said.
Yoti began turning a profit this year after booking £20mil (US$26mil/RM107.41mil) in revenue in the 12 months to March.
The company forecasts a 50% increase in sales in its current financial year.
Competitors are also doing well, with 34 companies like Persona, Kids Web Services, K-id or VerifyMy tallied among the members of the Age Verification Providers Association (AVPA).
In 2021, the AVPA forecast annual revenue of almost US$10bil (RM41.31bil) for the sector across the 37 countries belonging to the OECD club of wealthy nations for the 2031-36 period, although it has not issued any more recent outlooks.
AVPA director Iain Corby is wary of predicting exactly what lies ahead.
"There's so much that can happen. Regulation is moving very fast, technology is moving very fast," he told AFP.
AI-powered age verification methods have raised questions over potential bias and privacy impacts.
The tools are "potentially a bit intrusive or dangerous to people's private life," said Olivier Blazy, a cybersecurity professor at Polytechnique, the prestigious French engineering school.
But it also depends on the information that sites share with their third-party age verification provider, he added.
"I think the ecosystem is tilted towards AI at the moment, but there may be a pull-back in the coming years to better protect users' private life," Blazy suggested.
Make-up tricks
Blazy pointed out that the systems still have weak points, noting that "just with normal makeup, it's relatively easy to make yourself look older or younger".
Others have pointed to biases in the age-verification algorithms, which can be less accurate in their estimates when presented with non-white faces.
One report from an independent Australian body found that "underrepresentation of Indigenous populations remains a challenge" – albeit one "that vendors are beginning to address".
Yoti boss Robin Tombs acknowledges that less data was available to train the company's model for some age groups or skin colours.
But he insists their tool can "check to see whether we think people are using false accessories or they are using makeup and other such things".
On privacy, he said Yoti's tool immediately removes all data after carrying out its analysis.
Platforms that deploy the selfie analysis tools from companies like Yoti can adjust settings for the evaluations, leaving some margin of error.
For example, sites barring under-18s often require the system to judge that visitors are over 21.
Those in the grey zone may find themselves asked to fall back on a more traditional verification method – such as pulling an official ID out of their wallet to snap or scan. – AFP
