compas unhinged an algorithm used by big portions of our justice system to predict whether someone will be a repeat offender, influencing their sentencing. the more likely to repeat offend the higher the sentence. julia dressel, a computer science student at dartmouth, wrote a thesis to analyze the effectiveness of the algorithm. she wanted to test whether or not it was more effective than humans so she went to the website mechanical turks and enlisted actual humans to do the same thing as COMPAS. '...on average the workers were correct in 63 percent of the cases... ...the performance was NOT significantly worse than the performance of the algorithm. these two methods, human and algorithm, were indistinguishable. this is a sobering result for advocates of using algorithms in sentencing. despite all the complexity of the algorithm - the comprehensive data collection; the long interviews with offenders; the use of principal component analysis and regression models; the writing of 150-page operational manuals and all the time take to train judges in using the algorithm - it produces results that are no better than those achieved by a bunch of people randomly recruited from the internet...amateurs beat the algorithm. julia told me that she was motivated to conduct the study because she is frightened by the many ways that technology reinforces oppression. 'humans are quick to assume that technology is objective and fair, so cases where technology isn't fair ar the most dangerous,' she told me.' - david sumpter 180723 what's it to you? who go blather from