CaTE Data Curation for Trustworthy AI

Data curation is a critical phase for trustworthiness: The report identifies the data curation phase as a key opportunity for development teams to actively promote trustworthiness in AI systems, suggesting that trustworthy AI cannot be achieved through model design alone.

A structured, sequential approach exists: The authors provide a series of core steps that data scientists and development teams can follow during data curation, with documented strengths, weaknesses, and preconditions for each step, making trustworthiness a systematic rather than ad-hoc process.

Multiple pathways accommodate different contexts: The report traces parallel paths where alternatives exist in the data curation process, recognizing that different AI projects may have different constraints and requirements while still pursuing trustworthiness.

Practical tools are available and documented: The guidance includes implementations using open-source software tools, making the recommended practices accessible to development teams rather than purely theoretical.

Trustworthiness requires intentional definition and planning: The report begins by defining both "data curation" and "trustworthiness" explicitly, emphasizing that teams must clarify what trustworthiness means for their specific AI system before implementing curation practices.