True and its worth discussing here
OK. It's a myth and the claim is a result of willful ignorance. The more criteria are taken into consideration when evaluating the role of race in the criminal justice system, the more it becomes obvious that there aren't significant unexplainable differences.
A new study, released in Septemeber of 2016, looks at the state of Delaware. Published by the Department of Criminology at the University of Pennsylvania, submitted to the Chief Justice of the Delaware Supreme Court, and the Delaware Access to Justice Commission’s Subcommittee on Fairness in the Adult Criminal Justice System.
Initial Situation:
"African Americans are overrepresented in the criminal justice system of Delaware. African Americans represent
22% of the population of residents in Delaware. Data [...] show that African Americans represent roughly
42% of arrests,
42% of criminal convictions, and
51% of incarceration sentences.
[...]approximately
57% of inmates serving time in the Delaware Department of Correction (DOC)’s facilities
were African American."
What they did:
"This study relies on administrative records to track individuals from their arrest to their corresponding court dispositions, and examines the extent to which observed racial disparities in incarceration sentences and sentence lengths remain after statistically controlling for current case and charge characteristics, contextual factors, and criminal history. This study also details what racial disparities in incarceration commitments and sentences lengths would look like if White defendants had similar characteristics as African American defendants."
Findings:
"The results from this study indicate there are significant disparities in incarceration sentences and sentence lengths for African Americans relative to Whites. These disparities decreased to levels that were practically small for sentences to incarceration in a multivariate regression model that controlled for
current case and charge characteristics, criminal history, and contextual factors (gender, age,county location, public defender,and pretrial detention), as well as a multivariate reweighting method that compared African Americans to similarly situated Whites.
In both comparisons the unexplained differences in the probability of receiving an incarceration sentence were less than 1% (0.5%). The disparities in effective length of incarceration sentences decreased overall to levels (under 40 days) that were no longer statistically significant in the multivariate regression model, as well as the reweighting method that compared African Americans to similarly situated Whites."
http://courts.delaware.gov/supreme/docs/DE_DisparityReport.pdf
Sure that's fair and as one of the people quoted for the article says even the most accurate prediction could produce a socially undesirable result with unbiased data. However, we shouldn't necessarily assume that because these predictions are spit out by an algorithm that they're inherently unbiased. There are people involved at multiple levels here and their human imperfections means the data won;t always be perfect.
Depends on what political philosophies you're subscribing to. If you believe that ethical decision making and justice can only be evaluated on an individual level, something I'd agree to, then somebody's freedom shouldn't depend on correctness and accuracy of predictive modeling.
If you believe, like most people on the left side of the contemporary political scale do, that some sort of common good not only exists but also tops individual rights, it's far less problematic.
Sure, some individuals have to bite the bullet but the prediction will be accurate more often than not, that's how statistics work.
Philosophically, I believe something like this shouldn't be used at all to decide over somebody's future.
From a pragmatic/consequentialist standpoint, the alternative is probably the (un)educated guess of some fat official who graduated in the lower 50% of his class, which depends on what he had for breakfast.
Much of the critique just seems to be a misunderstanding of what 'algorithm', 'bias' or 'correctness' mean.
Private companies are allowed to develop these programs without releasing them to the public. Do the fucking math.
They obviously should be open source. And the article should focus on that.
