At least the pre cogs could actually see the future, instead of being a tool for confirmation bias such as this news report..... reports
Seriously, if you want to take AI seriously you need s feedback loop. 1sr round yeah use historical data because there's nothing else to use. Then keep training the model and feed it actual outcomes of successful and unsuccessful cases where it was used. It's not rocket science, folks. Reminder, the success rate here was 1%, so if you're not blind you'll pretty much see a giant red flag.
Only THEN it's even worth weeding out discrimination bias which is likely not to happen because it's actually the success criteria. But I digress.