AI/ML, lesson 1.0; do not introduce bias elements FFS!
Otherwise: neat example of all that can go wrong with these applications if you aren't careful enough.
Back in the 80s they tried to build a model that would be given a patientβs entire medical record and predict whether they had cancer.
It was surprisingly accurate, but it turns out that it was really just selecting for patients seeing oncologists.
It's surprising how they forget that the machine will learn a pattern and if it "case a" always shows a "b" pattern, then it will likely show any case as "a" just because it's biased by "b". Even when "case a" isn't involved D:
And that's why you need specialists!
I can't imagine how bad it would be in some areas if they simply do it without care. Without wanting to be intrusive Mika, hope you are doing better these days π!
