Seeing Like a Toolkit: How Toolkits Envision the Work of AI Ethics

Toolkits encode hidden assumptions: The 27 AI ethics toolkits analyzed in this study aren't neutral—they embed specific assumptions about what ethical work should look like and who should do it, meaning their design choices actively shape how organizations approach AI ethics.

Gap between vision and support: There is a significant mismatch between what toolkits imagine ethical work should entail and the practical guidance they actually provide, leaving users without clear direction on implementation.

Power dynamics are overlooked: The toolkits notably lack guidance on navigating labor, organizational, and institutional power dynamics—critical real-world factors that determine whether ethical recommendations can actually be implemented in practice.

Unclear responsibility assignment: The toolkits don't adequately clarify who should be responsible for performing ethical work, which can lead to confusion about accountability and diffusion of responsibility across organizations.

Implications for toolkit design: The research identifies specific gaps that should inform future development of AI ethics toolkits, suggesting that more effective tools need to address organizational realities and power structures, not just technical or procedural considerations.