AI systems trained with data that is unrepresentative of the general population are often noticeably worse at providing care for members of excuded minority groups.

Question

Answer

Pro: It is significantly harder to obtain data from groups who have less access to high-tech forms of healthcare, creating significant wealth, geographical and racial diversity issues.

Answer

Pro: Most AI funders and developers are affluent, and come from the dominant ethnic and racial groups of their countries, furthering overrepresenting such people in their eventual AI systems.

Answer

Con: This is a temporary problem: researchers are attempting to find new ways to ensure that biased data sets do not produce harmful end products.

Should human doctors be replaced by automated ones?

AI systems trained with data that is unrepresentative of the general population are often noticeably worse at providing care for members of excuded minority groups.

Pros

Cons

It is significantly harder to obtain data from groups who have less access to high-tech forms of healthcare, creating significant wealth, geographical and racial diversity issues.

Most AI funders and developers are affluent, and come from the dominant ethnic and racial groups of their countries, furthering overrepresenting such people in their eventual AI systems.

This is a temporary problem: researchers are attempting to find new ways to ensure that biased data sets do not produce harmful end products.

Selected claim, opposing its parent

AI systems trained with data that is unrepresentative of the general population are often noticeably worse at providing care for members of excuded minority groups.

Pro claim 1

It is significantly harder to obtain data from groups who have less access to high-tech forms of healthcare, creating significant wealth, geographical and racial diversity issues.

Pro claim 2

Most AI funders and developers are affluent, and come from the dominant ethnic and racial groups of their countries, furthering overrepresenting such people in their eventual AI systems.

Con claim 1

This is a temporary problem: researchers are attempting to find new ways to ensure that biased data sets do not produce harmful end products.