Data Hazards: An open-source vocabulary of ethical hazards for data-intensive projects

Abstract

Understanding the potential for downstream harms from data-intensive technologies requires strong collaboration across disciplines and with the public. Having shared vocabularies of concerns reduces the communication barriers inherent in this work. The Data Hazards project (datahazards.com) contains an open-source, controlled vocabulary of 11 hazards associated with data science work, presented as ‘labels’. Each label has (i) an icon, (ii) a description, (iii) examples, and, crucially, (iv) suggested safety precautions. A reflective discussion format and resources have also been developed. These have been created over three years with feedback from interdisciplinary contributors, and their use evaluated by participants (N=47). The labels include concerns often out-of-scope for ethics committees, like environmental impact. The resources can be used as a structure for interdisciplinary harms discovery work, for communicating hazards, collecting public input or in educational settings. Future versions of the project will develop through feedback from open-source contributions, methodological research and outreach.