Harvard Reveals AI Cancer Bias: Race, Age Affect Diagnosis Accuracy

• Harvard Medical School research exposes AI bias in cancer detection based on patient race, gender, and age, leading to misdiagnosis risks.
• AI models, unlike human pathologists, identify patient demographics from tissue samples and alter diagnostic results accordingly.
• Study on 29,000 images found AI accuracy varied in 29% of cases, missing lung cancer in African-Americans/men and breast cancer in younger patients.
• Bias stems from incomplete training data, AI using disease prevalence as a shortcut, and detecting subtle, race-specific cellular cues.
• Scientists developed 'Fairpath', a new framework that reduced AI bias by 88% by teaching it to ignore demographic cues and focus solely on disease.

Why It Matters: AI cancer diagnosis shows bias by race/age; Harvard's Fairpath offers a promising solution.