UNESCO Recommendation on the Ethics of AI PA3a: paras 71, 75 to 77 Data governance for AI: evaluate training data quality continuously, promote open and quality datasets, annotate and share data responsibly

Member States should develop data governance strategies that continually evaluate training data quality, including collection and selection, security and protection, and feedback to learn from mistakes. They should promote open data (reviewing access to information and open government policies, open repositories for publicly funded data and code, and data trusts), promote quality, robust, diverse datasets built on a valid legal basis, encourage annotation standards including disaggregation by gender, adopt a digital commons approach where appropriate with interoperable tools, and encourage companies to share data for research and public benefit in trusted, secure data spaces.