Yoti pitches facial age estimation for kids 6 to 12 years old

Yoti claims its biometrics-based facial age estimation technology can now deliver accurate age assurance in real-time for children under 13 years of age, potentially setting the company up to meet the online age assurance needs of social media networks and other online service providers.

Julie Dawson, Yoti’s director of Regulatory and Policy, tells Biometric Update in an interview that the facial age verification for younger people could be used in different ways by a broad swatch of online service providers.

Yoti says its technology delivers a Mean Absolute Error rate of 1.3 years for the 6 to 12 years old age group, as well as 1.5 years between 13 and 18. The company turned to professional data collection companies to bring in permissioned data for children too young to use its digital ID app, which is limited to those 13 and older.

The new capability is being launched now because of the improvement in technical capacity, but also changes in the broader market conditions.

Dawson sees a “landmark change in the last months” in the regulatory environment for online interactions involving children. Inappropriate content is just the beginning of the set of worries regulators are attempting to address, and Dawson notes that research indicates 24 children are groomed for manipulation through Facebook each week in the UK.

The UK’s Age Appropriate Design Code is being followed by similar legislation in many other jurisdictions, including Ireland, which is the data protection jurisdiction in Europe for many platforms.

Yoti’s age estimation for people between 13 and 24 had reached a Mean Absolute Error rate of roughly 1.5 years, and by extending its age estimation to younger children, the company hopes to enable platforms that want to provide age-appropriate services to do so.

Dawson also notes that while technology may already be helping these platforms know or at least estimate their users’ ages, Yoti has an independently reviewed technology that can perform the task for any platform, giving regulators a new option.

Yoti has performed over 500 million age checks so far for interactions with age restrictions for the teenagers and people in their early twenties.

In addition to curating content based on age appropriateness and enforcing access control rules based on age, Dawson notes that with many existing regulations applying special status to the collection of data from minors, age estimation could be a compliance tool for online service providers. That service provider could then turn off some standard data collection practices, such as location tracking.

Platforms turning on new capabilities for age appropriate services in one jurisdiction are often considering doing so elsewhere at the same time, Dawson says.

For parts of the world where parental consent is a mechanism of age restriction for online service, age estimation with facial analysis can be used to determine that the person giving permission is indeed an adult.

Stephanie Hare recently expressed one of the alleged downsides of a biometric school meals program in Scotland as training the normalization of using the body to transact.

In this case, Dawson says, the actions being taken by online service providers to make their services age appropriate are driven by necessity. The options for age estimation or verification are limited, however, and since traditional age and identity checks involve the sharing, and often the storage of personal information, Dawson does not have the same concerns around Yoti’s facial age estimation.

“It’s not a one-to-one recognition, it’s not a one-to-many recognition,” Dawson emphasizes. “There’s no unique identification. So on that basis what we’ve found is that the platforms that have done the testing so far are finding that the friction is low, and that consumers in the main are finding it a very straightforward way to go forwards.”

Dawson also notes the importance of offering choice, and explaining how the technology works, for which Yoti produced a video.

The choices available are limited by similar constraints to many digital identity-related challenges. Children generally do not have ID documents, for instance, and if parental consent is to be used, an ID document-based system could also exclude some parents.

With Yoti’s age estimation technology neither identifying an individual, nor storing any information, Dawson says it is “about as privacy-protecting as we can get” while keeping online environments safe for children.

The company’s transparency efforts include its a white paper released this month on age verification, which includes

The company’s technology was approved by the UK’s Age Check certification body in 2019, and has also conducted a bias review with an IEEE expert.

Dawson admits that in theory the system could be spoofed by a determined adolescent with makeup, but says it would require a substantial investment of time and skill with each check, and reiterates that extensive testing and consultation with various customers and independent groups have given all stakeholders confidence in the system’s effectiveness.

Yoti provides a confidence score for age estimation, and Dawson says setting thresholds for those scores will largely fall to regulators.

The company also offers multiple approaches in parallel, allowing a version of step-up age verification.

Service providers tend to be at the point of comparing solutions and approaches to address a well-defined problem, but there are simply fewer options available to estimate the age of younger users. In the case of legal obligations, service providers may want to use a buffer with age estimation, but Dawson points out that for those tailoring instructions or support, it may not be necessary.

The application for Yoti’s younger facial age estimation capability, therefore, extends far beyond keeping children off of Facebook, to a range of safety-related and non-safety use cases.

“It’s enabling organizations to devise an age-appropriate service from soup to nuts,” Dawson sums.

She sees a need for better benchmarking, which is also addressed in the white paper, and for regulators and platforms to be aware of the questions they should be asking, such as whether bias has been observed in an algorithm, and the source of images used.

This post was updated at 9:02am on November 1, 2021, to correct the total number of age checks Yoti has performed.

Article Topics

access control  |  age verification  |  biometric age estimation  |  California Age Appropriate Design Code (CAADC)  |  children  |  facial age estimation (FAE)  |  facial analysis  |  privacy  |  regulation  |  Yoti