The Observatory of Technological, Economic, and Societal Impacts of Artificial Intelligence (OTESIA) is launching four projects around AI today. The second on this list focuses on the prevention of cyberbullying, and more specifically, how to effectively counter it?
In Europe, more than a third of young people witness hateful remarks on the internet. Among them, no less than 6% of young people are directly affected by cyberbullying. This number has been steadily increasing over the past few years as social networks have become more democratized and established themselves as significant places for interaction. To combat this remote hatred, a software project focused on its prevention via AI (Artificial Intelligence) is being launched by OTESIA. Led by Serena Villata and Elena Cabrio (I3S / Inria team) and Catherine Blaya (Université Côte d’Azur – CAPEF-INSPE), this software is responsible for detecting hateful messages based on natural language analysis. But it’s not just about spotting keywords; it also involves understanding the argumentative structure of the message in its entirety. The software identifies the form of hatred (racist, xenophobic, against a religion) and then seeks to understand its nature (humor, allusion, implicit, context). The goal is to recognize genuine hateful speech and not just simple messages containing insults.
Tools for Victims
Preventing hateful messages is not the only objective of OTESIA’s software. Providing countermeasures for the victims of cyberbullying is also planned as part of this project. It will offer potential responses to the remarks made by the harasser, allowing the bullied person to defend themselves and learn to do so independently. To develop these defenses, interventions will take place between September 2020 and January 2021 in six secondary schools (middle and high schools). Several role-playing scenarios based on harassment situations will be conducted to analyze both the remarks made and the students’ responses. All this data will be processed to make the software more precise and functional. Initial results will be presented in June 2021.
