The most recent developments in AI indicate that algorithmic predictions are more accurate than human authorities when comes to calculating the risk of recidivism for offenders, with algorithms clearly outperforming their human counterparts. A software called ‘‘Correctional Offender Management Identification Tools’ (COMPIT) was developed by a private company in the USA to assist human judges during the sentencing stage. 

The aim of COMPIT is to deliver more accurate, unbiased results where their human counterparts fail to do so. While Northpointe Inc. assured their clients COMPIT predictions were in the “good range of predictive accuracy”. 

There were findings in 2016 that African-American offenders assessed by the assessment tool were two times more likely to be given a recidivism high-rate risk than the rest of the population. The risk assessment software is comprised of 137 factors such as age, gender, previous criminal history, and omits race. COMPIT is used to calculate the risk score of recidivism for offenders on a scale of 1 to 10, 1 being less likely, 10 being most likely). 

One particular defendant of African ethnicity was being assessed by the AI tool and subsequently given a much higher rate than his white counterpart who had committed the same crime; they had both stolen approximately $80 worth of supplies from an Office Depot store. 

Algorithmic bias is a common concern when evaluating the ethical implications of AI. Bias can be easily inherited by machines from their creators and can also be a result of unintentionally including proxy information such as postal codes or addresses in training data sets. Discriminatory bias in AI being used in criminal justice needs to be addressed at the root of its developmental stages. The private entities developing algorithms being used in a public driven sector such as the judicial sector should also disclose the datasets being implemented.