Cyberbullying : the limits of human means
In this age of digital technology and virtual communication, cyberbullying has become a major societal problem. These remote attacks are virtual in name only. Their psychological impact is very real, and their consequences can be irreparable, particularly for teenagers who are the primary victims (according to an American study 1, 20% of young victims of cyberbullying have already considered suicide in a sample of 2,000 middle school students).
Unlike face-to-face bullying, which operates in a limited physical and temporal environment, online bullying knows no boundaries. It can occur anywhere and anytime, making it much more difficult to combat. Hidden behind their screens, anonymous bullies feel protected, while victims find themselves overexposed and more vulnerable. Moreover, as social networks and communication platforms continue to multiply, it is impossible for humans to monitor, and control cyberbullies in real-time. These bullies use a wide range of techniques to attack their victims (insults, mockery, threats, grooming, defamation, outing, etc.).
The contributions of artificial intelligence
Developments in artificial intelligence, particularly in Machine Learning and natural language processing, have opened up a whole host of possibilities for curbing or at least detecting cyberbullying early on, and thus protecting people online, particularly the younger users.
Machine learning involves building models, based on statistical techniques or neural networks, that detect specific elements among a large amount of data and classify them into precise categories, previously defined by humans.
Thus, one of the major challenges for AI in the fight against cyberbullying is to successfully create models that automatically and in real-time identify signs of bullying from unstructured data (texts, images, videos), and alert parents to detected problems so they can take timely action if necessary.
