Background: AI-generated images can support or impede health communication efforts and influence perceptions of health-related topics, making it important to evaluate how these images portray topics like palliative care (PC) and hospice care (HC), which are characterized by limited public understanding (e.g., regarding where these services are delivered) and misperceptions (e.g., widespread conflation of these services). Objectives: To examine characteristics of AI-generated PC and HC images. Design: Three generative AI tools (ChatGPT, DreamStudio, and Midjourney) were used to generate 40 images each for the prompts “a photograph of palliative care” and “a photograph of hospice care” ( N = 240). Images were coded for features such as the setting and the characteristics of people in the image (patients, providers, and caregivers). Code frequencies and percentages were calculated using STATA. Differences were assessed across prompts and AI tools. Results: PC images more often depicted medical settings (PC = 37.5%, HC = 27.5%) and medical equipment (PC = 67.5%, HC = 50.0%). DreamStudio images were the most medicalized. When included, patients were generally presented as White (PC = 85.0%, HC = 91.0%), female (PC = 81.0%, HC = 75.3%), and older (PC = 88.0%, HC = 94.4%) and rarely had negative affect (PC = 5.0%, HC = 2.2%) or looked ill (PC = 4.0%, HC = 2.2%). Providers were more racially diverse (PC = 41.1%, HC = 48.4% non-White), but nearly all were women (PC = 95.9%, HC = 96.9% female). Providers displayed supportive touch frequently (PC = 72.6%, HC = 71.9%) but were rarely shown engaging in medical tasks (PC = 12.3%, HC = 6.3%). Caregivers were infrequently included in either PC or HC images. Conclusions: The few substantial differences identified between PC and HC images could reinforce conflation between these services. Some image features could also create additional misperceptions (e.g., regarding the role of caregivers).
Gaysynsky et al. (Fri,) studied this question.