Big cities used to be really dangerous but crime was put under control by methods that are now under attack. Data-driven predictions of where crime will happen allows police to dedicate resources to locations where they will be most effective at preventing crime. Data-driven decision making is what drives every sector in society. It belongs in policing. This ban is a mistake. Unfortunately, the consequences of the ban won't be experienced until those who voted the rule into effect are long gone from office.
Being able to predict where/when murder happens leads to a more efficient and effective distribution of law enforcement. Other cities have done this with non-violent crime/traffic violations/etc., and crime has gone down. So prediction = prevention.
I think that this is an important point, and regret not discussing this more in the paper.
Predictive policing can absolutely be bad, and can be especially problematic when the algorithms are designed (or used) to target individuals or groups of people for investigations.
They can be racist, biased, and are open to misinterpretation by those who end up using them. They can also lack transparency, and are often trained on skewed datasets that can result in self-perpetuation of reported crime patterns.
I don't mean to suggest that this work is entirely without bias. However, in this particular paper we describe a method that uses point of interest data exclusively as predictor of violent crime levels. We don't consider income, education, age, race or any other socio-economic factors in the model. We simply determine the linear combination of different point of interest densities which most closely reproduces the crime density map for a city -- i.e. we don't weight one venue more strongly than another because of its location.
Police departments and governments have a lot to do when it comes ensuring fair, equal policing. Academics/researchers need to be mindful that the work and analyses they do in a lab can have significant impacts on people's lives. However, we tried to be sympathetic to these issues.
The paper mentions interventions around alcohol-serving venues -- these are things like staggered closing times for pubs (avoiding everyone in a city being kicked out from their pub at the same time), restrictions around using glass bottles/glasses, and security requirements at certain times of the day. These interventions have been shown to work in pubs/clubs, and have measurably reduced violence related hospital admissions. Our hope was that this analysis might encourage councils to consider spreading such interventions/policies to other locations, beyond those that serve alcohol, as doing so may have similarly beneficial effects.
With this, police will soon stop intervening to stop crimes from happening, and will instead focus their attention on after-the-fact data gathering and reporting. I can't blame them, but quality of life for law abiding citizens is likely to suffer. In high crime areas, I can understand police considering certain neighborhoods as "no go" zones. Why take such a risk with your own life to stop a crime against somebody else?
Wasn't it known in the 70s that something like 90% of crime occurs on a select streets? Not sure we need AI to make this prediction and achieve this accuracy.
> More police presence means more recorded crimes.
The article in the headline implies the opposite: statistical inference of future crime location -> more cops in said location -> no crime in said location.
It should be fairly easy to back out an out of sample data set for this - ie predicted crime location where the extra cops didn't get deployed and see if that lines up with empirical observations.
And if they accurately predict where crime will occur, they will reduce crime in those neighborhoods. It's likely that they would be able to get good results.
Right, and the secondary effects? If you live near where some people got robbed, you now have cops stopping you for no reason, just because you look like everybody else in that (now marked as high crime) neighborhood.
I have no idea what the optimal way to plan police patrols is, but algorithms predicting crime used to guide patrol planning? That's just dystopian.
Also, if there is higher prevalence of crime in certain areas, it makes sense to do more policing there. Now it is illegal to detect higher prevalence of crime?
The game Watch Dogs 2 is actually quite on target regarding this. In the game, the TPTB puts a non-scientific bias into predictive algorithms for certain neighbourhoods making the population of those areas pay more in insurance and be targeted by police a lot more often.
It's a dangerous path to take as suddenly you can be labeled a criminal just for living in the wrong place or even worse if the police put their bias into the algorithms.
This doesn't seem to make sense. By more accurately predicting where crimes will occur, the police departments can reduce the amount of patrols needed.
People want to be data driven until the data goes against their feelings. You see the same in SF, the narrative was that crime has spiked during COVID and after seeing stats that it didn't, the new narrative is that the perception of crime is still important, the data didn't matter
I still encounter people who think a map of crime is an actual map of crime, like SimCity, and not a map of policing. They really do believe it reflects reality. Most times they snap out of it when you point out it's impossible to map crime without collected data, and then it's a short hop to realize policing is how you get that data. And any intellectually honest person will recognize the biases inherent to that data.
I would be surprised if there are more predictive variables than “places where crime has happened repeatedly over time” that are significant enough or uncorrelated and noncausal to ever be more useful than patrol planners do today using standard statistics.
reply