Running summarizing tools on a new article

Abstract: In their 2017 article, "Algorithmic Decision Making and the Cost of Fairness," a group of Stanford researchers led by Sam Corbett-Davies point out that using fairness-mitigated versions of COMPAS for pretrial and posttrial decisions will have an impact on public safety.To estimate this impact, the authors tested the three central types of group fairness metric that we've considered (se lection, precision, and recall) and estimated the increase in violent crime caused by releasing more violent offenders in the name of equality.This was the same historical data that ProPublica collected from Broward County, Florida, so we have rec ords of whether each prisoner did go on to re-offend.All three fairness-mitigated models were compared with a baseline model that only optimized for accuracy.The authors found that each of the three fairness-mitigated models created an increase in violent crime.In addition, the authors found increases in the rates of low-risk prisoners who were incorrectly detained. Fairness versus Accuracy Parity metricViolent crime Increase Low-risk detention increase Equal se lection 9% 17%Equal precision 7% 14%Equal recall 4% 10%These findings are consistent with what we know about group fairness metrics.Equalizing the rates of White and non-White prisoners for these metrics will most likely come at the cost of releasing more "higher-risk" members of the disadvantaged groups (FPs) and imprisoning more "lower risk" members of the advantaged group (FNs), compared with the blind model.