EU regulation of Artificial intelligence in the shadow of global interdependence

To make regulatory interdependence tractable, RegulAite develops a dedicated theoretical framework that links it to the central regulatory challenges AI poses. This framework elucidates when and how it matters for regulation at home--in this case in the EU--what other jurisdictions do. And it clarifies what, given this regulatory interdependence, effective forms of cross-border regulatory cooperation are.

This framework concentrates on two central dimensions of regulatory interdependence. The first one is the degree of inter-jurisdictional cooperation necessary to achieve regulatory goals. The second one covers the economic (opportunity) costs of exceeding other jurisdictions’ standards. Where regulatory concerns fall along the two axes depends on their specific characteristics.

The project iteratively applies the outlined framework to EU AI diplomacy to order and understand how regulatory interdependence shapes the latter, and to use empirical data from the field to refine the framework itself. This cyclical integration of deductive and inductive elements—the analytical dimension—then allows prospective policy analysis of the options the EU faces in its AI diplomacy. The iterative character means that the project both assesses and refines propositions, instead of testing hypotheses in a unidirectional fashion.

RegulAite has been designed with clear knowledge utilization in mind: building theory and analysis to aid forward-looking policy debates and public deliberation. It pragmatically adapts methodological approaches common in the social sciences.

Mapping and theorizing regulatory interdependence in AI

In the first project phase, the individual regulatory concerns constitute the units to be analysed and compared: for example, how is regulatory interdependence different for potentially discriminatory AI scoring systems than for automated disinformation through digital media?

Because most regulatory concerns have a non-economic character, strictly quantitative cost-benefit analyses are of limited use. Even where economic (opportunity) costs are a key worry, these are impossible to quantify reliably in absolute terms. Relative comparisons and qualitative judgments are therefore essential for locating regulatory issues along the two axes of the regulatory interdependence framework—the economic costs of exceeding regulatory standards elsewhere, and the degree of inter-jurisdictional cooperation necessary to achieve regulatory goals.

Our empirical mapping of current policy dynamics draws on policy documents, reports from the organizations concerned with AI governance, and news reports—all material that is online. That said, understanding the character, stakes and policy dynamics surrounding regulatory concerns requires deep familiarity with the policy field. Reflecting RegulAite’s integrated inductive and deductive approach, we will use semi-structured interviews to inquire about policy dynamics systematically—assessing the strength of our propositions—while leaving space for unexpected insights.

Prospective policy analysis

The prospective dimension of the project takes inspiration from the Delphi method, developed in the 1960s to furnish policymakers with scientifically grounded future scenarios to aid decision making. It develops competing scenarios, and then spells out their characteristics and assesses their plausibility with the help of experts. In contrast to methods bent on finding and explaining patterns in historical data, this future-oriented approach is a natural fit for emerging AI policy and diplomacy.

Our scenarios will follow the alternative options for EU AI diplomacy in its four axes—more or less (selective) bilateral engagement, multilateral standard development, mutual recognition of local rules, and so on. For their assessment, we adapt the policy Delphi method pragmatically. The fully-fledged version uses a rigid expert interview protocol, formulated ex ante, to collect answers to a specific policy question—say, the expected economic benefits of a new dam. Instead, we will elicit qualitative assessments of encompassing scenarios, asking experts to articulate their main characteristics and highlight factors that, in their eyes, would make them more or less likely.