> a talk by researcher Kristian Lum a few years back I think you mean this research, discussing Oakland: "Setting the record straight on predictive policing an

> a talk by researcher Kristian Lum a few years back I think you mean this research, discussing Oakland: "Setting the record straight on predictive policing and race" - K Lum, W Isaac, 2018 [https://theappeal.org/setting-the-record-straight-on-predictive-policing-and-race-fe588b457ca2/ https://theappeal.org/setting-the-record-straight-on-predict...]. Her full bibliography is at [1] and X/Twitter is @KLdivergence > drug crimes, but she pointed out that if you look at medical data (where people tend to be fairly honest about their drug usage) pretty much everyone in the metro area under study (SF) used drugs Right. Second, without knowing their definition of "drug use" or "drug crime", that could mean anything from minor stuff like smoking/buying/possessing marijuana by under-18s, possessing or transporting >1 oz of marijuana, all the way up to possession or sale of large quantities of meth or fentanyl. Third, and what would be obvious to anyone familiar with the Bay Area 2020-22, if predictive policing used "drug crime" (convictions? or arrests?) as opposed to "drug use", then when the then-SF DA stopped prosecuting possession of personal-use levels of meth, SF police tend to reduce or stop arresting for it. So the arrest, prosecution or conviction data from SF would differ sharply to Oakland or pre-2020 SF or 2023 SF or San Mateo County. All the above factors combined seems like a huge combination of "data drift", "feature drift", "label drift", "model drift". [0]: https://hrdag.org/people/kristian-lum-phd/ https://hrdag.org/people/kristian-lum-phd/ [1]: https://scholar.google.com/scholar?q=%22kristian+lum%22 https://scholar.google.com/scholar?q=%22kristian+lum%22