sure seems like most of the people reviewing this paper don't really get what it means and immediately jump to concluding it's "racist".
If there are correlates to self-reported race in medical images, machine learning will pick it up. Are you surprised this information is there? Race correlates to genetic history, which is what determines the differences they're seeing.
Concluding from this "AI is racist and this means medical AI is going to be really shitty" is completely the wrong message to take from this. All of this information the ML picks up on is useful data that can be used to improve outcomes for a wide range of people.
Are you suggesting that this diverse team of leading scientists from the biggest institutes in the world, and all of the famous researchers who have agreed with their message on Twitter etc, have somehow jumped to the wrong conclusion, when you yourself find the answer obvious?
What diverse team of leading scientists from the biggest institutes in the world?
Most of the people commenting on this are second grade "ethics researchers". The authors are not ML experts. They clearly had a strong prior belief about the outcomes and were surprised by the outcome, which is fairly obvious: nearly every bit of medical data you can collect on people is ultimately correlated to self-reported race, but in a way that is not immediately perceptible to expert humans! That in its own is the important observation here. Everything else is just authors and commenters making odd statements that are in conflict with reality.
(note: my background is in biology including genetics and image analysis, and I've worked with many of the experts in this field)
3 comments
[ 3.1 ms ] story [ 20.7 ms ] threadIf there are correlates to self-reported race in medical images, machine learning will pick it up. Are you surprised this information is there? Race correlates to genetic history, which is what determines the differences they're seeing.
Concluding from this "AI is racist and this means medical AI is going to be really shitty" is completely the wrong message to take from this. All of this information the ML picks up on is useful data that can be used to improve outcomes for a wide range of people.
Most of the people commenting on this are second grade "ethics researchers". The authors are not ML experts. They clearly had a strong prior belief about the outcomes and were surprised by the outcome, which is fairly obvious: nearly every bit of medical data you can collect on people is ultimately correlated to self-reported race, but in a way that is not immediately perceptible to expert humans! That in its own is the important observation here. Everything else is just authors and commenters making odd statements that are in conflict with reality.
(note: my background is in biology including genetics and image analysis, and I've worked with many of the experts in this field)