The US National Institute of Standards and Technology has examined age estimation software and concluded that it has improved but still needs work.

It is the agency's first report on the topic in ten years. After examining six algorithms, NIST concluded that none clearly outperformed the rest. However, NIST acknowledged the pace of AI development and noted that it would update its evaluation every four to six weeks to reflect technological improvements.

Estimating a person's age based on physical characteristics is becoming increasingly important as regulators seek ways to enforce age restrictions on activities beyond the user simply saying, "Yep, I am over 18," or answering a question about bygone times.

When NIST last examined age estimation algorithms in 2014, the online world was very different. Now, regulators are considering the technology as a gatekeeper for activities with an age restriction, such as accessing mature content online or dipping a toe into social media.

NIST said: "Age estimation has become an enabling technology in age assurance programs recently included in legislation and regulation both inside and outside the United States. These programs aim to permit only those in certain age groups to access social media chat rooms or to buy certain products both online and in the physical world and can be an important part of efforts to protect children online."

The problem is that the six algorithms submitted when NIST made its call for submissions in September 2023 still have plenty of room for improvement, according to Kayee Hanaoka, a NIST computer scientist.

In 2014, NIST's test used a database of approximately six million photos from visa applications. The latest test expands the collection to around 11.5 million photos, augmenting the visa images with FBI mugshots, border crossing images, and immigration application photos.

The images vary in quality, gender, and demographics, and were anonymized.

Worryingly, NIST reported that one of the algorithms in its test did not outperform the best one of 2014. However, overall accuracy was up from ten years ago, although error rates were higher among discerning ages for females than for males. The same finding emerged from the 2014 study. NIST said, "The underlying reasons are unknown."

That said, the mean absolute error for the algorithms has decreased from 4.3 to 3.1 years.

NIST noted: "There is no single standout algorithm, and a given algorithm's accuracy is influenced by image quality, gender, region of birth, the age of the person in the photograph, and interactions among these factors. The algorithms all have their own sensitivities with certain demographic groups; an algorithm that performs well on certain groups can perform poorly on others."

With NIST now adopting a rolling testing program, and new algorithms being accepted, that accuracy will likely improve.

It will need to if regulators - the UK's Ofcom has published proposals for age verifcation - are serious about applying age restrictions using age estimation and verification (AEV) software. ®

https://www.theregister.com//2024/06/03/us_standards_agency_publishes_a/