> snake oil – [...] predicts where crime is likely to occur in the future.

Speaking strictly in the abstract here: I'm not sure this is as impossible or snake-oily as the quoted bit implies. From a technical perspective, as long as you're willing to put a lot of data in (unknown how much data about this is collected), and have a good pipeline, I'm sure a model can find enough correlations to get some accuracy better than random chance. Tracking movement patterns, events where lots of people gather, events where lots of tourists are present, temperature, employment stats, housing stats, etc. could all lead to some freaky "predictions" that might be way more accurate than a lot of people would guess. You could get a lot of these signals from GSM tracking, tbh.

The problem with this is that even if the implementation is solid (and all the care in the world is taken when curating datasets, ensuring no biases and so on), and even if deployment is solid (say schedule more LE presence in those areas), there will inevitably be situations where things escalate and LE start trusting the system way more than they should, and start harassing bystanders for no reason other than "the AI said so, you must be a criminal". Minority report vibes and so on.

The other problem is that it seems inevitable that some places will develop and deploy these things, and you can be sure that at least some of them will not care about solid implementations, biases and so on (UK is trialing facial recognition in stores, fully private but w/ cooperation from local LE).