New measure separates bias from human facial-age errors
A reanalysis involving 30 students and a new experiment recruiting 75 adults supported a measure that separates systematic bias from facial-age estimation errors.
Limit: Excluding online participants with unusually large errors could distort the measured error patterns
Edited by Boris Djordjevic · AgingResearch.News
In facial-age judgment experiments, 30 university students (18 women, average age 23.4) and 75 online participants (36 women, average age 29.2) estimated ages from faces. Researchers reanalyzed the students’ earlier data and ran a new online experiment. They excluded five online participants with unusually large errors.
Using these results and simulations, the researchers found that the conventional accuracy measure mixed estimation errors with systematic tendencies to judge faces as younger or older. They introduced a framework with a new measure that separated these biases from errors. With this measure, human errors increased with the age of the face until about 60, then decreased. Tests of ten artificial intelligence models showed a different pattern: their errors continued to increase with the age of the face.
Why it matters
Apparent facial age serves as a marker of biological aging. The framework may help researchers evaluate the tools used to study that marker.
Caveats
The experiments assessed facial-age judgments, not health outcomes or lifespan. The human samples were young on average, and the computer comparison covered ten selected models.
The paper
A new framework for facial age estimation in humans and AIs
Ben-Gurion University of the Negev; Western University
Scientific Reports, May 11, 2026, Open access
