Understanding the Representation and Representativeness of Age in AI Data Sets

Older adults are severely underrepresented in AI face datasets: Only 5 out of 92 datasets examined explicitly documented inclusion of adults over 65 years old, and just 1 dataset included adults over 85 years old, despite these groups being significant portions of the population.

Age documentation is inconsistent and often missing: Only 24 of the 92 datasets (26%) included any age-related information in their documentation or metadata, and those that did used different methods to record ages, making it difficult to assess or compare age representation across datasets.

Lack of age diversity in training data could harm AI performance for older users: The paper emphasizes that underrepresentation of older adults in training datasets means AI systems—particularly facial recognition and analysis tools—may not work equally well for elderly users, raising fairness and inclusion concerns.

Age has been overlooked in AI fairness discussions: While recent efforts have focused on balancing datasets across race, gender, socioeconomic status, and disability, age has received minimal attention as a demographic dimension requiring proportional representation.

Creating age-representative datasets requires new standards: The authors call for researchers and engineers to establish consistent methods for collecting, documenting, and reporting age information in datasets, recognizing this as an important but underaddressed challenge in building inclusive AI systems.