Expanding Explainability: Towards Social Transparency in AI systems

Current AI explanations ignore social context: The paper finds that most Explainable AI (XAI) approaches focus narrowly on the algorithm itself, overlooking the fact that in human conversations, explanations are naturally shaped by social and organizational context. This gap means AI explanations often fail to match how people actually need information presented in real-world settings.

Social Transparency can improve trust calibration: Based on interviews with 29 AI users and practitioners, the research suggests that incorporating socio-organizational context into AI explanations—what the authors call "Social Transparency"—can help people develop more appropriate levels of trust in AI systems, rather than either over-trusting or under-trusting them.

Better explanations support organizational decision-making: The framework developed in this research indicates that socially-situated explanations can facilitate collective action within organizations and improve the quality of decisions made with AI systems, suggesting benefits beyond individual users to entire teams and institutions.

Holistic explainability requires multiple levels of design: The paper identifies three interconnected levels where Social Transparency matters—technical, decision-making, and organizational—indicating that responsible AI requires attention to how explanations function across these different layers, not just at the technical level.

Human-centered XAI needs expansion beyond algorithms: This work contributes to a broader shift in AI design philosophy, arguing that truly human-centered explainability must account for the social and organizational environments where AI systems operate, expanding the design space beyond traditional technical approaches.