AI LEGO: Scaffolding Cross-Functional Collaboration in Industrial Responsible AI Practices during Early Design Stages

Early-stage harm identification requires dedicated tools, not generic project management software: The research found that standard tools like JIRA and Google Docs fail to support meaningful cross-functional collaboration on responsible AI concerns, often requiring "significant extra effort to align understanding" between technical and non-technical team members.

Technical teams rarely communicate design choices in ways non-technical roles can meaningfully evaluate: A key finding from interviews with 8 industry practitioners revealed that "technical design choices are rarely handed off in ways that support meaningful engagement by non-technical roles," creating a critical gap in harm evaluation across teams.

Visual, shared structures are essential for bridging the technical-non-technical divide: The study identified that "collaborative workflows lack shared, visual structures to support mutual understanding," suggesting that structured, visual representations of AI development decisions can help teams work together more effectively on responsible AI concerns.

Non-technical practitioners need systematic scaffolding to evaluate potential harms: The research found that non-technical team members are "left without scaffolds for systematic harm evaluation," indicating they need structured guidance and frameworks—not just access to information—to contribute meaningfully to responsible AI practices.

Boundary object theory can operationalize cross-functional RAI collaboration: The authors developed AI LEGO, a prototype using boundary object theory, where technical roles draft development plans through interactive blocks while non-technical roles engage with those same blocks through role-specific interfaces, demonstrating a practical approach to supporting joint harm identification.