recidivism

Terms from Artificial Intelligence: humans at the heart of algorithms

Recidivism is a technical term in the judicial system refering to the likelihood that soemone who has previsouly committed. acrime will do so again. The controversial AI system COMPAS creates a precdicted recidivism score (effectively high or low likelhood of reoffending) and has been used by judges in the USA as part of parole decisions. COMPAS was built using machine learning over past felons and their actual reoffending rates, so in a way is based on ground truth data. However, in practice the scores appear to exhibit significant racial bias, made worse because automation bias means the scores may effectively be treated as recommendations.

Used on page 555