AI-Powered Technology-Facilitated Violence Prevention

Technology-facilitated violence (TFV), encompassing cyberbullying, threats, exploitation, and cyberstalking, is an escalating global issue, disproportionately affecting vulnerable populations, particularly disabled people. These individuals face unique challenges, including ableism, social isolation, and limited digital literacy, which make them more susceptible to online harm. Additionally, the use of shared devices with caregivers complicates the detection and prevention of TFV, rendering current technologies inadequate. This study aims to develop an innovative AI system utilizing Contextual AI to detect and address TFV targeting disabled people. By incorporating interdisciplinary expertise, we will model context-sensitive AI solutions that enhance reporting mechanisms and empower disabled people to seek help without fear of retaliation. The project focuses on four main objectives: (1) Context Modeling and Interdisciplinary Collaboration to create disability-inclusive AI designs, (2) Dataset Development to establish a comprehensive, diverse dataset reflecting the experiences of disabled people across their intersecting identities, (3) AI Model Development and Evaluation, applying advanced algorithms like Graph Neural Networks and Large Language Models to detect TFV, and (4) Pilot User Interface Study to assess the AI's effectiveness in real-world settings. This research will contribute to disability justice, creating safer technological environments and improving support systems for disabled people in Canada.

Technology-facilitated violence (TFV), which involves the use of communication technologies to intimidate, control, or harm individuals, has become a critical global issue, with various forms such as cyberbullying, threats, doxxing, exploitation, hacking, and cyberstalking. Vulnerable populations, including disabled people, are particularly at risk due to factors like ableism, social exclusion, isolation, poverty, and limited digital literacy. While disabled people experience the same types of abuse as non-disabled individuals, they also face unique challenges, such as sharing device access with caregivers, which makes current detection technologies ineffective. There is an urgent need for specialized systems to detect TFV targeting disabled people and provide effective prevention and intervention strategies. This interdisciplinary study aims to develop an AI system that uses Contextual AI, technology that understands and responds to specific situational contexts, to recognize and address the unique forms of abuse disabled people face. By leveraging Contextual AI, the project seeks to create safer technological environments, improve reporting mechanisms, and empower disabled people to seek help without fear of dismissal or retaliation, ultimately advancing disability justice and inclusivity in Canada. The primary goal of this project is to develop a specialized and contextual AI system to detect, prevent, and moderate TFV against disabled people. To achieve this, we will focus on four specific objectives:

• Context Modeling and Interdisciplinary Collaboration: We will employ robust context modeling through an interdisciplinary collaboration with disability experts and AI engineers. This will lead to the creation of disability-inclusive AI designs and the development of user-centered, accessible systems for detecting TFV.
• Dataset Development: We will establish a comprehensive dataset explicitly designed for AI training, reflecting the diverse experiences of disabled people, including their intersecting identities (such as gender, race, sexuality, and class). This dataset will be the first of its kind, informed by an extensive literature review (across social sciences and TFV) and expert insights from around the world.
• AI Model Development and Evaluation: We will develop and evaluate contextual AI models for detecting TFV against disabled people using advanced technologies like Graph Neural Networks (GNN), Large Language Models (LLM), and Retrieval Augmented Generation (RAG). These sophisticated AI algorithms have shown promise in improving performance in specific contexts but have not yet been applied to TFV or similar issues, adding novelty to our work.
• Pilot User Interface Study: We will evaluate the performance of the AI model through a pilot human user interface study, involving advocates, end users, and other stakeholders to assess its effectiveness in real-world scenarios.