Researchers develop new training technique that aims to make AI systems less socially biased

FairDeDup reduces social biases in AI by removing redundant data while incorporating controllable, human-defined dimensions of diversity to mitigate biases4. This approach enables cost-effective and accurate AI training while promoting fairness by allowing people to define what is fair in their specific setting, rather than relying on biased internet data or large-scale datasets4.

FairDeDup addresses biases related to occupation, race, gender, age, geography, and culture during AI training2. It mitigates these biases by removing redundant data and incorporating controllable, human-defined dimensions of diversity, leading to more fair and socially just AI systems12.