21:37 11 May 2017
An artificial intelligence system with five-year training of offending histories data will be used by Durham police when deciding whether or not a suspect should be kept in custody. The system classifies suspects as a low, medium or high risk of offending.
Data for the Harm Assessment Risk Tool (Hart) was taken from Durham police records between 2008 and 2012. The data was then tested during 2013 and the results were monitored over the following years. The system has 98per cent accuracy in identifying low-risk suspects and 88per cent accuracy in identifying high-risk suspects. The Durham system includes data beyond a suspect's offending history - including their postcode and gender, for example.
Prof Cary Coglianese, a political scientist at the University of Pennsylvania who has studied algorithmic decision-making, has warned about the risk of using such systems, saying: "To some extent, what learning models do is bring out into the foreground hidden and tacit assumptions that have been made all along by human beings,"
"These are very tricky [machine learning] models to try and assess the degree to which they are truly discriminatory."