Facial age estimation raises technical, legal, interdisciplinary questions

The latest Lunch Talk from the European Association of Biometrics focuses on facial age estimation (FAE) – specifically, a “roadmap for technological and legal deployment and development.”

Richard Guest of the University of Southampton explains: “what we really want to do is to identify what we believe are the major interdisciplinary issues in the community, and not just the academic community by any stretch of the imagination; we’re talking stakeholders and end users and industry and governments.”

Guest runs through use cases for age estimation ranging from the familiar (restricting access to pornography) to exploratory (refugee services). He touches on the issue of explainability – a way to explain why deep learning systems make the decisions they make – and variability of facial images, be it in image quality or “facial coverings, beards, makeup, that sort of thing.”

Standards, testing and evaluation get a nod, notably the National Institute of Standards and Technology (NIST) Face Analysis Technology Evaluation (FATE) Age Estimation, and the in-progress ISO/IEC 27566-1 for age assurance systems.

Martin Sas, a doctoral researcher at the Centre for IT and IP law (CiTiP) at KL Leuven, Belgium, takes the reins to take a broad overview of the legal and regulatory framework. He finds a map with many intersecting laws made for different reasons and by various bodies, few of which contain any explicit mention of facial age estimation. Major guidelines like the GDPR, the AI Act and the Digital Services Act have various approaches to proportionality, risk and special categories of data.

Eva Lievens, an Associate Professor of Law and Technology at Ghent University, Belgium, looks at some of the issues around data for training, such as ethical and legal challenges, the tension between privacy and accuracy, and regulatory questions around informed consent.

“From a legal perspective there can be a little bit of a paradox,” Lievens says, “because very often facial age estimation is argued to be a more privacy friendly method, if you compare it to giving your ID a copy of an ID with all kinds of information about yourself. But at the same time in order for the systems to work accurately we need to collect lots of data for the actual training.”

“There are still many legal and ethical questions that need to be solved in terms of transparency, for instance towards the individuals whose images are being used for these kinds of purposes.”

Guest adds that “we have to have an understanding of where data comes from and the legal framework within which the systems actually legally operate.” Synthetic data raises yet another challenge, being still “very much unexplored” in terms of data testing.

“I think testing is absolutely critical in this whole area, just to ensure that people have confidence in the systems,” Guest says. “These are being rolled out with claimed performance, but we need independent assessment.” He also believes there is work to be done on attack detection – notably, determining if someone is a fraudster attacking, or just a guy with a weird beard.

“I think we’ve got to establish a mechanism for detecting when somebody is trying to attack the system, and that’s an ongoing bit of work. This is very much in its infancy, because we’re just trying to get the performance assessment right across age estimation systems. Then we move on to how people might suddenly try and circumvent the outcome. So we’ve got to try and build that into any evaluation that we might want to do on these systems.”

Guest comes back to explainability, and the need for “an easy to understand guide to the technologies.”

“I think it’s going to be absolutely vital as more and more of these systems are used and deployed. Having an easy to use guide for both legal teams within the courts but also juries and the public to understand how the systems actually work, dispel a lot of myths and hopefully gain confidence – I think that’s very very important.”

The big takeaway is that facial age estimation, and the legal levers to regulate it, are still a work in progress: “there’s still research, there’s still development, there’s still deployment questions to be asked.”

Article Topics

biometric age estimation  |  EAB  |  EAB 2025  |  European Association for Biometrics  |  Face Analysis Technology Evaluation (FATE)  |  face biometrics  |  facial age estimation (FAE)  |  ISO 27566-1