My first reading of the yet to be published EU Regulation on the European Approach for Artificial Intelligence, based on a leaked version, I find pretty good. A logical approach, laid out in the 92 recitals preceding the articles, based on risk assessment, where erosion of human and citizen rights or risk to key infrastructure and services and product safety is deemed high risk by definition. High risk means more strict conditions, following some of the building blocks of the GDPR, also when it comes to governance and penalties. Those conditions are tied to being allowed to put a product on the market, and are tied to how they perform in practice (not just how they’re intended). I find that an elegant combination, risk assessment based on citizen rights and critical systems, and connected to well-worn mechanisms of market access and market monitoring. It places those conditions on both producers and users, as well as other parties involved along the supply chain. The EU approach to data and AI align well this way it seems, and express the European geopolitical proposition concerning data and AI, centered on civic rights, into codified law. That codification, like the GDPR, is how the EU exports its norms to elsewhere.

The text should be published soon by the EC, and I’ll try a write-up in more detail then.