Last year our research team built a new version of ChatGPT that better understands time. By looking at millions of anonymous health records, it can predict who might get diseases like cancer or heart disease, and roughly when this might happen. These early warnings could help doctors choose the best time and person who should get screening tests, preventative medicines like statins, or limited-supply vaccines to new epidemics. But the tool works less well for people already in poorer health: ethnic minorities, low income families and those who live outside big towns. Because most training data come from richer, white groups, the predictions for others are less accurate. If hospitals use the tool today, health gaps will widen.
We can tackle this. We hold secure, linked health and social data on every adult over 40 in Estonia, Finland and Denmark, and we will soon add Sweden. With these four national databases we will:
- Ask our upgraded ChatGPT to predict four real problems (breast cancer, colorectal cancer, heart attacks and brain strokes, and death from COVID 19) in each country, and measure where it gets less accurate.
- Test and benchmark five new technical tricks that claim to make AI fairer, checking whether they improve fairness. We will check accuracy if someone happens to be older, female, not born in the country where they live, with no university education, with less than average income, currently unemployed, unmarried, with a disability, or living in the countryside.
- Run an online study with 1,500 volunteers (375 per country). They will compare health scenarios, for example a tool that helps many people but is unfair to a few, versus a second tool that helps fewer yet treats everyone more equally. Their choices will show how much unfairness the public will accept.
These steps will create the first "fairness map" for medical AI: a clear guide to which social groups and diseases face the biggest risk of unfair treatment, and which fixes work the best. By sharing the map, we will steer developers and health leaders toward tools that are fair enough, show how to fix those that aren't, and stop those that still don't meet the fairness mark. In short, we will turn vague worries about biased AI into clear numbers and action, so future health technology helps everyone, not just the lucky few.
