Missed Opportunities for Human-Centered AI Research: Understanding Stakeholder Collaboration in Mental Health AI Research

DOI

In the mental health domain, where AI technologies will impact their treatment and daily lives, patient engagement can be a key to designing human-centered technologies. CSCW and HCI researchers have delved into various facets of collaboration in AI research; however, previous research neglects the individuals who produce the data and who will be impacted by the resulting technologies, such as patients. This study examines how interdisciplinary researchers and mental health patients who donate their data for AI research collaborate and how we can improve human-centeredness in mental health AI research. We interviewed patient participants, AI researchers, and clinical researchers in a federally funded mental health AI research project. We used the concept of boundary objects to understand stakeholder collaboration. Our findings reveal that the social media data provided by patient participants functioned as boundary objects that facilitated stakeholder collaboration. Although the collaboration appeared to be successful, we argue that building consensus, or understanding each other's perspectives, can improve the human-centeredness of mental health AI research. Based on the findings, we provide suggestions for human-centered mental health AI research, working with data donors as domain experts, making invisible work visible, and privacy implications.