A controversial study: in one test an artificial intelligence system was trained to identify psychiatric patients with suicidal tendencies
Artificial intelligence could help identify people with suicidal tendencies, to stop them before they can harm themselves as a result of discomfort. The researchers in Carnegie Mellon University (USA) affirm this in a controversial study. They have put in place the most advanced systems of machine learning and image recognition to predict the intentions of those who are overwhelmed by a malaise and want to end their lives .
The Marcel Just team conducted the study of 17 adult patients with proven suicidal tendencies, starting with magnetic resonance imaging (Magnetic Resonance Imaging, MRI) of volunteer brains.
SUICIDE ANNOUNCED
During MRI, the volunteers, eight of whom had already tried to take their own lives, were stimulated with negative words, such as "despair" or "loneliness", positive words (for example "carelessness") and generically complex words ("problems ").
The researchers then isolated the brain areas activated by the various terms and compared them with those activated by the same words on a neurotypical control group, ie without suicidal tendencies.
The images were used to train a machine learning system expert in photographic analysis and recognition: at the end of the process the computer was able to identify with a precision of 91% the patients who showed self-injurious tendencies.
In a second test, the same system, trained only with images of psychiatric patients, managed to correctly identify those who had already attempted suicide in 94% of cases. The most accurate answers are those related to imaging induced in response to the terms "death", "lifeless" and "carefree".
The study was carried out on a small sample and, to be admitted by the same researchers, should be validated on a much higher number of cases before being able to make a diagnostic tool.
However, the research did not fail to arouse perplexity. Several scientists have declared themselves skeptical about the possibility of using such solutions in everyday clinical practice.
"We need very sophisticated and expensive equipment, but also extremely collaborative patients", says Derek Hill (University College, London).
"Each of our behavior has a biological basis, but it is not said that MRI is the right tool to identify it," says Blake Richards (University of Toronto).
To answer at least in part the objections to technology, the Carnegie Mellon team is trying to repeat the same experiment using a common electroencephalogram, which can be done with simpler, portable and less expensive tools.
Short Video From CMU
