AI may help predict which patients with psychiatric conditions are likely to commit crimes.
The vast majority of patients with psychiatric conditions are not violent or dangerous at all, but there are a few that will go on to commit a crime sometime after they leave the doctor’s clinic.
A group of Canadian researchers from McMaster University have developed a machine learning algorithm that can pick out the bad apples (but have sadly failed to name the system RoboCop).
Their AI system could predict which patients with psychiatric conditions would commit a sexual offence with a sensitivity of 82% and specificity of 60% using 36 different clinical, historical, and sociodemographic variables.
Violent and non-violent crimes could be predicted with a similar accuracy by the AI but with fewer variables, the researchers reported in the Journal of Psychiatric Research.
The gold-standard way to determine the likelihood of criminal offences among a patient cohort is use of actuarial risk estimates.
But these old school tools can’t individually predict the type of criminal offence a patient will subsequently commit, they can only assess the general likelihood of crime occurring in a group sample.
“The current results suggest that machine learning models can show greater accuracy than gold-standard risk assessment tools,” the researchers said.
“However, unlike existing risk tools, this approach allows for the prediction of cases at an individual level, which is more clinically useful.”
If you see crimes before they happen, alert felicity@medicalrepublic.com.au.
