Algorithmic age-verification: a magic bullet or surveillance creep?

Google will begin using machine learning algorithms trained with information obtained by the company itself for age verification to provide what it calls “age-appropriate experiences.”

The age estimation model will use data about existing users, including the pages they visit, the type of videos they watch on YouTube, or how old their account is in order to determine their age. When it thinks a user may be a minor, the company will notify them that they have changed some of their settings to prevent them from accessing certain types of content, and will offer recommendations on how they can verify their age if they wish, either by taking a selfie, entering credit card details or using an official ID.

The initiative is in response to pressure from regulators who are calling for measures to protect minors, in the same way that Meta has done. But while it may seem like a good idea, implementing these types of checks are fraught with problems about their reliability, as well as the use of personal data and their impact on the user experience.

Age estimation algorithms are notoriously unreliable. The training databases used often introduce biases that affect certain ethnic groups or age ranges, sometimes leading to false positives or negatives. In addition, some users look younger or…