For starters, you don't conduct two simultaneous experiments on the same subject because of confounding variables, let alone 50.
It's also useful because most AI models are able to talk about what the community is saying about drugs and the model can correlate that with many other things and it does enhance diagnostic and it's quite useful to train medical models.
See a relatively "new" example is about Vapes, there has been deaths and so-on due to people experimenting with illicit ones, without those reports, we will never have in future Science book and AI models that some chemicals are dangerous to inhale or whatever (it's a shitty example but you get my point)