The Use of AI Chatbots in Domestic Violence and Sexual Assault Contexts

Abstract

In a world where growing numbers of people are turning to AI chatbots for various forms of advice, victim-survivors of domestic violence and sexual assault are also beginning to seek out AI chatbots for emotional support and legal advice. The National Domestic Violence Hotline now offers an AI chatbot “Ruth” — in addition to their existing telephone, virtual chat, and text-support options — to specifically support victim-survivors as they navigate DV/SA situations. Other DV/SA-specific AI chatbots have found favor among victim-survivors for their assistance in gathering evidence, completing legal documents, offering empathetic emotional support, organizing and recording timelines of incidents, and connecting victims to available human resources. As the use of AI chatbots in DV/SA contexts only becomes more widespread, policymakers have a responsibility to investigate 1) how victim-survivors are utilizing these AI chatbots, 2) how competent the AI chatbots are in providing accurate information, and 3) what gaps or issues remain for policymakers to address. This thesis tests both general-purpose chatbots (ChatGPT and Gemini) and DV/SA-specific AI chatbots (Aimee Says, Sophia, Hope Chat AI, and Ruth) to evaluate their effectiveness at providing emotional support and legal assistance in common DV/SA contexts. In comparison to the expected guidance that would be offered by a trained, human DV/SA advocate, this research found that AI chatbots do not currently offer services and information at an equivalent level. In some tests, the AI chatbots in fact offer insufficient information or improper assistance, potentially leaving victim-survivors vulnerable to a number of safety and support gaps. An interview with a frontline DV/SA social worker — from a DV/SA nonprofit organization in New Jersey — is also conducted, summarized, and analyzed. The interviewee discusses how victim-survivors are largely utilizing AI chatbots for extra support in conjunction with formal support from counselors and attorneys — rather than as the sole source of emotional or legal support. However, there remain major legal obstacles (related to confidentiality and privacy concerns) which restrict DV/SA organizations from adopting and encouraging the use of AI chatbots by victim-survivors. Drawing from the results of the AI chatbot experiments and my interviewee’s discoveries, three policy recommendations are made to assist victim-survivors, policymakers, and frontline DV/SA response teams alike in preparing for future AI chatbot use in DV/SA contexts. In order to reduce overall implementation costs, protect victim-survivor’s confidentiality, and serve the greatest number of victim-survivors possible, this thesis suggests a final policy proposal to make changes to the current National Domestic Violence Hotline’s AI Chatbot, “Ruth,” adopting the strengths of the other chatbots and seamlessly incorporating into the websites of DV/SA nonprofit organizations across the country. A general recommendation is also made to train current DV/SA advocates and frontline responders on victim-survivors’ uses of AI chatbots, so that they can be prepared to properly guide victim-survivors when they predictably encounter victim-survivors’ use of AI tools in the future.