The Work to Make Assistance Work: Learning from Videos by Visually Impaired Content Creators

Abstract

While multimodal AI holds strong potential to support the blind and visually impaired (BVI), current systems often fail to provide meaningful assistance. In this paper, we analyse online videos created by BVI content creators to understand how assistance unfolds in practice. Across two studies, we examine videos of BVI creators engaging with systems, peers, and BVI experts (such as surfer, soccer player) in everyday and challenging situations. We show how these actors both succeed and fail in shaping how BVI people project, coordinate, and interpret their worlds. Particularly, interactions among BVI experts revealed how effective assistance unfolds through back-and-forth interaction. From this, we characterise the concept of worldmaking, where assistance is centred on constructing a shared worksite of objects and actions. These findings highlight directions for designing multimodal AI systems that support worldmaking, enabling both AI and BVI people to make the world intelligible together.