But that's the ONLY difference. The rest is fairly comparable from the hallucinations of stuff that never happened, erasure of stuff that did happen from their data set memories, and completely entrusting due process and justice into a system that was fallible enough to be manipulated by a single bad actor.
Being a walking, talking advertisement for Scientology. They target celebrities because they know celebrities are great advertisements. People with parasocial relationships to celebrities are more likely to join Scientology, something which scientology exploits. It's why I've started trying to avoid any media that features prominent Scientologists; I don't want to support that kinda behavior. If they didn't try to exploit it then I wouldn't care too much, but the fact they're aware and try to exploit it makes me very uncomfortable.
Not to pee on anyone's pool but that's something that most law enforcement offices have been doing for a-la-la-long-time: Statistical analysis.
Basically you get a lot of crime data, you create a forecast and voila: you can "predict" where future crimes will happen*.
*Prediction accuracy depends on data accuracy (i.e. as accurate as your cops are legit and people trust them enough to report) and model bias.
Model bias is interesting: we weight some petty crimes like weed peddling heavier than if my company poisons the city water supply.
Why is "AI" included here is because in the last decade, we've become better at deep learning forecasting methods, that improve on traditional statistical analysis.
That said, in real-life scenarios you ought to combine both statistical and ML models with DL ones to reach a valuable forecasting.
My final point is that this is a huge time consuming activity: Anyone promising a system that can produce real-time realistic forecasting is a liar; useful forecasts take months to produce (and crime might have changed once completed).