Research on the influencing mechanism of blind or visually impaired persons’ evaluation on generative AI in visual tasks

Introduction. Generative AI (GAI) has shown significant potential in assisting blind or visually impaired (BVI) Persons in visual tasks. However, existing evaluations of GAI tend to focus on technical performance, overlooking the specific usage contexts and experiences of BVI users. Method. This study conducted action research and semi-structured interviews with 19 BVI persons, to explore their evaluations of GAI in visual tasks and the influencing mechanism of their evaluations. Analysis. Following grounded theory, we identified 16 categories, and corresponding 5 core categories, as well as their relationships. Results. The findings indicate that BVI persons primarily evaluate GAI based on three criteria: accessibility, credibility, and interactivity. Their evaluation is influenced by four main factors: system, information, BVI user, and context. Notably, both BVI user and contextual factors moderate the influence of the system and information on user evaluation. Conclusions. This study develops a model that explains the influence mechanism behind the evaluation on GAI by BVI persons in visual tasks. It not only broadens the scope of human-AI interaction research by incorporating diverse user types and task contexts, but also provides an empirical foundation for developing human-centered GAI.