AI Advancements Enhance Age Estimation, Panelists Say
Recent innovations in AI have greatly expanded the capabilities of age estimation, said panelists during a webinar hosted by BBB National Programs Thursday. They added that while there are risks associated with the tech, new regulations offer more protections for children’s privacy and data.
Sign up for a free preview to unlock the rest of this article
Export Compliance Daily combines U.S. export control news, foreign border import regulation and policy developments into a single daily information service that reliably informs its trade professional readers about important current issues affecting their operations.
“An age estimation product analyzes a face,” said Charlie Germano, counsel and senior technologist of the Children’s Advertising Review Unit (CARU) at BBB National Programs. “It is trained on data that is not tagged with identities,” so it can only identify age, not a name.
With facial recognition, on the other hand, the “whole goal is to identify who is in the image.” The techniques and data used for each purpose are very different, he added.
For example, age estimation “is not a positive ID, it's just looking at your facial characteristics” to guess someone’s age, Germano said. Identity verification, though, involves a user having to “produce an ID, whether it's government ID, a digital ID, or providing identity via a parental consent.”
Julie Dawson, chief policy and regulatory officer at Yoti, an age-verification provider, said more than one billion people "either don't own a government-issued identity document" or lack access to one, "or they don't feel comfortable using a document-based approach in certain areas,” so companies sought alternative ways to make age inferences, she said.
The facial template method detects a person’s face and conducts a pixel-level analysis to estimate age, she said. Then, the technology sends the result and deletes the photo and facial template.
Facial templates are personal data under COPPA, and other global and state laws also have obligations about what happens with facial templates, Germano said. But once facial mapping data is collected from the photo, “you don't need it anymore," said the CARU counsel: "You can throw the photo away and just use the data from these templates” to determine age.
Germano said a facial template is like having someone’s name or email address, “because without the photo, you can still infer a lot of things about a person.” Additionally, the amended COPPA rule “has more specific language around operators’ obligations for third-party data collection when they collect data on the operators’ behalf,” increasing protections “for the care and safety of this data.”
CARU Director Rukiya Bonner said the amended COPPA rule is important for third-party access to data, because some “companies may not be doing things directly,” and instead “rely on tools from third parties to process this data.”
But some age-estimation technology is better than others. Advancements in AI have made building a tool to guess age based on facial templates easy to do -- but also easy to do badly, Germano said. “You really need to make sure that [you understand] what you're using and how you configure it to suit your business, and your appetite for risk.”
Yoti can estimate within "a year of accuracy for young people around that 16- and 17-age group,” Dawson said.
She also noted that “regulators in each country are looking at this on a risk-based approach,” which can also be culturally specific. Ensuring that estimation methods comply with all global regulations, use data minimization and are transparent are keys to data protection and security, said Dawson: “Risks are evolving,” so regulations must adapt with them.