It's scary once you start considering edge cases. What if the machine learning accidentally and unknowingly discovers that people with a certain mental illness

It's scary once you start considering edge cases. What if the machine learning accidentally and unknowingly discovers that people with a certain mental illness are more prone to buying product X, then starts massively promoting product X to people on Facebook who also have the same mental illness because of shared traits? We should be careful with advertising to vulnerable people. This isn't even touching on the capability of nation states of nefarious groups manipulating populations on a large scale like Cambridge Analytica have likely done, which uses the exact same data advertisers use.