40 xAI: eXplainable AI

AI that uses deep learning identifies patterns and correlates answers in a way that is not explicitly programed, creating both higher risks and rewards. In 2016, an exposé by ProPublica was done on the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) algorithm. This algorithm is used by criminal justice officials to determine recidivism rates. This is one of several algorithms used by California, New York, Wisconsin, and parts of Florida. The COMPAS algorithm was found to have significant racial bias, which was not correlated to actual recidivism rates. Using data from Broward County, Florida ProPublica asserted that COMPAS predictions of violent crimes were correct 20 percent of the time, and that blacks were labeled at higher risk almost twice as much as whites. This study was widely criticized, including by the think tank Community Resources for Justice for misinterpretations of the data and the subject matter. Regardless, the problem of bias and trust in AI outcomes still exists.