Designing Human-centered Artificial Intelligence to Assist with Domestic Abuse Recovery: Mitigating Technology Enabled Coercive Control

Human-centered artificial intelligence (HAI) is at the forefront of current AI research efforts. Although several known challenges related to the use of AI are discussed in the present academic literature, the authors propose that AI has the potential to revolutionize many aspects of healthcare and social services, particularly for vulnerable populations. However, it is important to carefully consider the unique needs of vulnerable populations throughout the HAI design process as well as the ethical implications of using AI with these populations, and to build ethical HAI tools that will be thoroughly evaluated for their safety and security properties. Our current research investigates the unique recovery challenges and needs of domestic abuse survivors with continued vulnerability to technology enabled coercive control (TECC) attacks. One challenge of designing HAI for such an adversarial context, is what information to provide the HAI with so that the HAI can detect, interpret, or predict potential cognitive attack tactics used in coercive control. Another challenge in this context is determining what role the HAI should adopt, such as training versus decision aid. To explore potential design solutions to address these challenges, the authors aimed to derive safe, ethical, and situationally aware requirements within a potential clinician-patient training use case to enhance awareness of TECC-enacted cognitive attacks. The present paper describes the development of a preliminary set of HAI system requirements for the design of an instance-based learning (IBL) HAI model, that is contextualized for the complex adversarial landscape presented by TECC.