The Harm Assessment Risk Tool (HART) uses information from the local police database, analysing the records to make a decision about whether the person is at low, medium or high risk of offending while on bail.
In the UK, an arrest allows the police to hold a suspect for up to 24 hours before they have to bring a charge or release the person in question, although this window is longer if the charge is more serious (murder), or related to terrorism.
At this time, a custody sergeant will look at the evidence available to present to the Crown Prosecution Service (CPS), and weigh up the options.
For the AI, having access to almost five years worth of offending histories data, accrued between 2008-2012, it has been trained to use probability to calculate likelihood of an individual absconding or reoffending.
For example, suspects with no offending history would be less likely to be classified as high risk, though if they were arrested on suspicion of a very serious crime such as murder, that would have an impact on the output.
It can then offer a suggestion to staff about whether to keep in custody a few more hours, whether to release them on bail before a charge, or whether to remand them in custody.
Although, Sheena Urwin, head of criminal justice at Durham Constabulary, reiterated that HART would not override a custody sergeant’s final decision.
The system was trialed in 2013, for a year, and is now ready to go live, Urwin told the BBC: “I imagine in the next two to three months we’ll probably make it a live tool to support officer’s decision making.”
In the trial, forecasts that a suspect was low risk turned out to be accurate 98% of the time, while high risk was accurate 88% of the time.
This reflects the tool’s built in predisposition – it is designed to be more likely to classify someone as medium or high risk, in order to err on the side of caution and avoid releasing suspects who may commit a crime.
One expert said the tool could be useful, but the risk it could skew decisions should be carefully assessed.