In brief: Like other social networks, Instagram keeps grappling with the problem of those under 13 years of age joining the platform. As such, the Meta-owned company is testing a less conventional solution: an age-estimating AI that scans faces.

Instagram didn't start asking new users to input their date of birth until 2019, and it only requests age verification when teens try to edit their birth dates to show they are aged 18 or older. It's partly why 40% of children under 13 use the site and other social networks, according to a study last year.

The current way Instagram carries out age verification is by asking users to send in pictures of identification, such as a driver's license or birth certificate. But it is now testing two additional methods: social vouching and age verification via a video selfie.

The vouching method requires three mutual followers to confirm a person's age. They must also be over 18, not vouching for anyone else at the time, and meet other safeguards Instagram has in place.

The video selfie method is an interesting one. Once someone uploads a clip of themself, Meta shares it with a third-party company called Yoti, which uses machine learning trained on "hundreds of thousands" of pictures to estimate the person's age.

The words "Meta" and "data sharing" don't conjure the most favorable scenarios, but Facebook's parent insists Yoti technology cannot recognize identities and that both it and Yoti delete all information from their servers after the process is complete. You can try it here if you want to be flattered/horrified.

The accuracy of Yoti's tech varies depending on factors such as age ranges, skin tones, and genders---females with darker skin are the least accurate (+/- 3.47 years). The company said its system is 98.91% accurate at identifying 6- to 11-year-olds as under 13---there's no mention of 12-year-olds---and is 99% accurate at guessing if people aged 18 are older or younger than 25, which wouldn't be much use in this situation anyway.