What does a successful AI look like? Thirteen EU, OECD, G20, UNESCO and UN texts screened

Researchers analyse AI governance instruments published by supranational organizations since 2018 and call for moving beyond the broad principles to better coordinate.

Since 2018, supranational institutions have multiplied guidelines, recommendations and then regulations on artificial intelligence. A team including Delaram Golpayegani, Aphra Kerr and Dave Lewis wondered what vision of the "success" of AI these texts convey. Their corpus includes 13 documents published between 2018 and 2025, from bodies of the European Union, the OECD, the G20, UNESCO, the United Nations and the Council of Europe.

The method combines manual thematic analysis and automatic topic modeling, interpreted through the sociology of expectations, which studies how technological promises guide decisions. The authors draw a new ranking grid for the themes associated with the success of AI in these texts. Their conclusion is that the high-level principles are now widely shared, and there is now a need for better coordinated governance approaches between institutions to translate them into practice.