Substance abuse interventions could be assisted with AI tool
Image credit: Dreamstime
Intervention programs supporting homeless young people struggling with substance abuse could be aided with artificial intelligence [AI] tools which sort at-risk people into the most supportive groups.
Every year, up to two million young Americans experience homelessness and 30 to 70 per cent of these young people struggle with abuse of alcohol or other drugs.
For young people fighting substance abuse, being surrounded by good company has been found to be crucial to recovering and living a healthy life, while exposure to destructive behaviour can push them towards relapse.
This means that a difficult balance must be struck when organising group intervention programs for young people to open up about their experiences, share coping strategies and make good friends. In these programs, the participants must remain supportive of each other in order to avoid disastrous results; this is known as ‘deviancy training’.
In order to organise participants into the most supportive groups, researchers at the University of Southern California’s Center for AI in Society – working alongside Urban Peak, a non-profit organisation for homeless young people – have developed a machine-learning algorithm which sorts participants into optimal groups.
“We know that substance abuse is highly affected by social influence; in order words, who you are friends with,” said Aida Rahmattalabi, lead author of the study. “In order to improve effectiveness of interventions, you need to know how people will influence each other in a group.”
This algorithm prioritises supportive social connections, while cutting off connections which push participants towards temptation and relapse. It accounts for the existing relationships between members of a group, and each individual’s history of substance abuse. A model of interventions was built, based on behavioural theories and data gathered from homeless young volunteers in Los Angeles.
“Based on this, we have an influence model that explains how likely it is for an individual to adopt negative behaviours or change negative behaviours based on their participation in the group,” said Rahmattalabi. “This helps us predict what happens when we group people into smaller groups.”
Testing has demonstrated that the algorithm performs significantly more effectively than conventional control strategies for forming support groups.
Most surprisingly, the researchers discovered that – despite commonly held views – distributing consistent substance abusers across the groups is not the best way to intervene. Instead, these people’s existing relationships must be taken into account. The study also found that in some cases, carrying out any intervention at all can be detrimental.
“In some cases, we found it’s actually a bad idea to conduct the intervention. For example, if you have many high-risk people in a group, it is better to not connect them with low-risk individuals,” said Rahmattalabi. Rahmattalabi and her colleagues hope to have their AI tool ready to deploy by autumn 2018