Age Prediction

Age Prediction is the automatic estimation of a person’s age by a computer program. It relies on data that is already generated anyway: a selfie, a voice recording, or behavior within an app. The program does not output an exact date of birth, but rather a guess, such as “about 16 years old” or “probably under 18.” It is therefore a probabilistic statement, not proof. The difference from age verification is important: there, a real document such as an ID card is checked, whereas here the machine guesses based on features. Precisely for this reason, Age Prediction is convenient but never one hundred percent reliable.

Youth protection without an ID copy

Many countries are currently tightening the rules for social networks. In the European Union, the Digital Services Act requires platforms to give minors special protection. Australia has set a minimum age of 16 for social networks. Anyone wanting to implement such rules must somehow know who is young.

The obvious solution would be to require an ID from every user. This is tricky from a data protection standpoint, since a company would then collect millions of ID copies. It also deters adults who want to remain anonymous. Age Prediction is considered a compromise: it is meant to roughly classify age without revealing identity.

For the companies involved, a great deal of money is at stake. Violations of the European platform rules can be punished with fines of up to six percent of global annual revenue. At the same time, entire business models depend on this, since advertising to minors is heavily restricted. Age Prediction is therefore less a technical detail than a matter of legal certainty.

What the software uses to infer age

The most common approach works with facial images. A neural network — a program that learns patterns from many examples — is shown millions of photos with known ages during training. It thereby learns on its own which image features correlate with age. Nobody manually programs rules like “wrinkles mean old.”

In the end, the model usually outputs a number plus an uncertainty. Good systems are off by about one to two years on average. In practice, this is often not precise enough, since everything hinges on the boundary of 18 years. That is why platforms work with buffer zones: anyone estimated to be between 16 and 21 is additionally checked via ID.

Besides faces, providers also use behavioral data. This includes typing speed, times of use, accounts followed, or the language used in comments. A common misconception is that such systems are neutral. In fact, they estimate significantly less accurately for certain skin tones or regions of origin, because the training data was unbalanced.

Selfie checks at TikTok, Instagram, and in the App Store

Age Prediction is most visible on major platforms. When in doubt, Instagram has users record a short video selfie, which a service provider then evaluates. TikTok and YouTube additionally rely on behavioral signals to automatically switch accounts presumed to belong to minors into a protected mode. There, direct messages from strangers or personalized advertising are then restricted.

The technology is also encountered outside of social networks. Self-checkout supermarket registers estimate age via camera when alcohol is purchased and only call staff when in doubt. Games with age ratings and streaming services are testing similar methods. In news reports, the term usually comes up in connection with new youth protection laws.

Anyone reading such reports should pay attention to two terms. “Age Assurance” is the umbrella term for all methods of age verification. Age Prediction is merely the estimating variant of it. Critics point out that biometric data such as facial features are processed for the estimation, which is regulated particularly strictly in Europe.