L’intelligence artificielle peut-elle aider à estimer le risque de récidive dans les comportements violents ?

Abstract

In forensic psychiatry, magistrates raise the question of the existence of a risk of recidivism and dangerousness to psychiatric experts. Follow-up studies in forensic psychiatry showed that the psychiatric elements predictive of recidivism were mainly related to serious mental illnesses, toxic consumption, addictions, high levels of impulsivity, low insight, associated personality disorders, in particular antisocial personality disorders. There are also protective factors, in particular the observance of treatments. Given the complexity of psychiatric and criminological risk factors and protection, can artificial intelligence (AI) help psychiatrists and magistrates to improve the predictivity of recidivism? MethodsSystematic review of the literature on AI applications in the prediction of recidivism in forensic psychiatry, conducted according to PRISMA criteria, using the: “Artificial Intelligence”, “Recidivism”, “Personality Disorder”, “Impulsive” Behavior”, “Alcohol abuse”, “Drug Abuse”, “Schizophrenia”, “Bipolar disorder” on the PubMed, Science Direct, Clinical Trial and Google Scholar databases. ResultsThe vast majority of studies come from legal or computer reviews and very few from medical databases. The studies evaluating the AI in Forensic Psychiatry most often used Machine Learning based on sociodemographic, sociological and criminological data, notably the age of the first offense and the number of previous convictions. To date, there are very few studies evaluating psychiatric parameters, focusing on psychopathic personality disorders. Discussion/conclusionThe applications of the AI in Forensic Psychiatry are still very premature. However, some psychiatric criteria should be more prominent in this field, especially those from Webster's HCR-20 and Hare PCL-R scales. The challenge will also be to find relevant behavioral, psychological and psychiatric keywords to include in AI.