Artificial Intelligence to prevent suicides in prisons
The German federal government of North Rhine-Westphalia has awarded a contract for the research and development of a software using Artificial intelligence to detect and prevent suicides in prisons.
North Rhine-Westphalia (NRW) is launching a research project to develop software to support the existing efforts to detect and prevent suicide attempts in its prison facilities, using AI.
In the last three years the number of suicides in NRW´s prions have declined. So far, prisoners in NRW are screened at entry for potential suicidal intentions or warning signs. Those classified as high risk are then put under constant surveillance. However, given the limited amount of personnel, only those at high risk can be surveilled 24/7. Others are controlled in intervals.
Hence, the software could significantly increase the number of persons under constant surveillance, by this preventing suicide attempts or enabling a quicker response.
The contract for research and development was awarded to a company based in Saxony. There, a test room will be built that meets the requirements of a typical prison room. The knowledge gained there would be backed up by additional tests in the prison (JVA) Dusseldorf in real operation.
During its pilot phase the software is meant to surveil prisoners at risk. By using AI the software will learn and adopt to alarm the authorities about potential suicide attempts. As such, it will analyse movement patterns typical for acts of suicide or the detection of dangerous objects, like knives.
According to the German ministry of justice from 1998 to 2017 almost 1.400 inmates committed suicide in German prisons.
Note: This Article was informed by a Spiegel article about the software
Author: Niklas Hamann
Spiegel (2019). Künstliche Intelligenz soll suizidgefährdete Häftlinge erkennen In: Spiegel Online Available at: https://www.spiegel.de/netzwelt/web/kuenstliche-intelligenz-soll-suizidgefaehrdete-haeftlinge-erkennen-a-1292719.html
AI, NRW, Prison, suicide