Beyond GIGO: how "predictive policing" launders racism, corruption and bias to make them seem empirical

Originally published at:


If this predictive tool were turned on just the police, the conclusions would match those based on data that had been gathered randomly.

Think about it.


The author is right in that this will exacerbate bad policy - policy which well predates the technology.
You see this in Detroit suburbs (Warren, especially, which has four tight little fully-manned precincts on the Detroit border and then four more sprawling precincts covering the rest of the city): Dense concentration of police on the city border means that there are more cops who spot more people and find more crime, thus confirming that there’s more crime on the Detroit border, committed by Detroiters, which requires more cops…Begin again, Finnegan.



The IRS has a similar problem for deciding which taxpayers to audit. They need some way of establishing what are rates of tax compliance and what are good predictors of non-compliance and what audits will give the best ROI. If they took the same GIGO approach that some police are taking with “predictive policing” and did something like “predictive auditing”, they would be looking at results of audits to predict results of audits, and that would be a poor strategy.
Instead what they do is they periodically perform intensive audits on a random sample of taxpayers. It’s not a widely advertized program, but it’s real, and here’s a paper about it. It seems like a pretty good strategy for creating a set of valid training data, at least for IRS audits.
I’m not exactly sure how you would translate this into the world of policng, but the key is, you use true random sampling to learn what the real patterns are. Maybe one example would be, you pull over a random sample of cars (truly random, selected by computer) and search them. That would give an accurate training data set if you wanted to know which cars to search. The problem is that you can’t just pull cars over at random. But maybe there is some way of taking this idea of using perfectly random samplings to get a valid training set, and do it in a way that’s possible within 4th amendment protections.


Thank you for furthering the conversation and linking to evidence–which to me (working in patient safety around predicting incidents) is an interesting new discovery. This is the kind of comment I love and appreciate on bbs.


These programs create a feedback loop: areas where crime have been rampant will always be the areas that crime is rampant, at least until those areas are turned into mini-police states where nobody will even leave their houses, let alone jaywalk.

So again, what is the purpose of police forces, to protect and serve, or to be puritanical drill sergeants who exist to make life hell for everyone and make their jobs seem utterly beyond reproach?

Municipalities are always increasing police funding and presence, yet crime rates are falling. At what point do city governments say “ok, enough”? Can they?

The data needn’t be “dirty” to be considered questionable for making decisions. If you look at probability based on groups classified according to race, gender, i.e. profiling, you’ll seemingly get better results, however, you’ve lost all concern for the rights of individuals who happen to be caught up in one your problem groups.

Check out “Hello World” by Hannah Fry, and “Weapons of Math Destruction” by Kathy O’Neil for a look at problematic algorithms.

1 Like


The reduction of funding to the IRS has really undermined that. I read an article last year that pointed out the agency needs fewer resources to focus on compliance issues with wealthy people - because they make up a smaller percentage of tax filers. They are also more likely to be able to pay penalties or fines. That leads to a better ROI. Instead, we have an agency with fewer resources targeting the wealthy and corporations less frequently.

1 Like

A mini-police state may reduce some crime but in the process it will increase the likelihood people are arrested for things that didn’t used to be crimes.

This morning, in downtown Seattle, half asleep I jaywalked when there was no traffic visible in either direction. About 10 feet from the other sidewalk I realized the guy on a bike, obscured by a lamp post, was a cop. About a second after I saw him he noticed me and that I was jaywalking. And nothing happened.

If I was not white and lived in an area of “enhanced” policing I would have been tackled by multiple cops and fined for jaywalking and probably arrested for resisting arrest because I flinched as I was punched, and tased.


This topic was automatically closed after 5 days. New replies are no longer allowed.