AI technology exaggerates age recognition bias in faces more than humans

Artificial intelligence is the future. In fact, it’s already here. One of his latest advancements is using it to automatically estimate a person’s age based on their face. This is the technology used to determine who can enter a bar or view age-restricted content online.

But are there biases in AI processing? I discovered that it not only reproduced the biases, but exaggerated them.

The survey results are scientific report.

There are some well-known human biases and inaccuracies in estimating age from facial appearance. Previous research by this international research team found that people tended to overestimate the age of smiling people compared to those with neutral facial expressions, and that the accuracy of estimation decreased for older faces. Proven.

“The growing interest in age estimation using technology raises the question of how AI compares to human performance and suffers from the same biases,” said senior author Melvyn A. Goodale. says Mr. “Our results show that AI is less accurate and more biased than human observers when it comes to judging a person’s age, despite similar overall patterns of error and bias.” has shown.”

In this study, the AI ​​overestimated the age of smiles even more than human observers, showing that older faces compared to younger age faces, smiles compared to expressionless faces, and male faces compared to showed a sharp drop in accuracy for women. .

“These results suggest that estimation of age from faces is heavily influenced by visual cues rather than by high-level preconceptions,” said lead author Tzvi Ganel, Ben-Gurion, Cognition “The patterns of error and bias we observed may provide insights for designing more effective AI techniques for estimating age from faces. ”

AI performance data were collected from 2020 to 2022 using a representative set of 21 current commercial and non-commercial AI age estimation platforms. The AI’s performance was compared to that of Ben-Gurion’s 30 college students.

“AI tended to exaggerate the aging effect of smiling on the faces of young adults, misestimating their age by as much as two and a half years. It’s not seen in adults, the female face, it was present in the AI ​​system,” said Carmel Sofer of the Ben-Gurion Department of Cognitive Neuroscience.

Based on current findings, researchers can only speculate as to why these biases occur. Possibly due to the set of photos used to train the AI, or perhaps due to a statistical phenomenon called regression to the mean.

In other words, when in doubt, an AI (like a human observer) may choose an age estimate close to the average across the range of faces encountered, leading to an overestimation of the age of young people and an overestimation of their age. It can lead to underestimation. Elderly age.