EU regulation of Artificial intelligence in the shadow of global interdependence


Rapid innovation in AI has a worldwide impact -- touching societies around the globe, in both good ways and bad. The global nature of the AI transformation means that jurisdictions cannot govern it on small regulatory islands; instead, they have to appreciate that attempts to regulate AI, too, reverberate around the world. Project I asks: how do AI rules elsewhere in the world shape and affect the effectiveness of those in the EU?

The shape and effectiveness of European AI governance depend heavily on developments in the rest of the world. The USA and China are the two leading AI powers worldwide: they are home to the largest firms and research centres and generate most AI investment. AI-powered digital services can often be supplied from abroad through fibreoptic cables, potentially blunting the bite of European rules. And mismatches between EU AI rules and those elsewhere could create digital boundaries and fragmentation.

At the same time, the EU pursues an AI agenda that is different from what is found in many other places: it's emphasis on "trustworthy AI" seeks to curtail corporate and governmental intrusion into citizens' lives with AI tools - more so than most other countries.

Regulatory ambitions differ around the world, and in many ways, the EU finds itself at the restrictive end of the spectrum. Given AI's global character, the question therefore is: what impact do regulatory frameworks elsewhere in the world have on EU AI rules? In other words, what regulatory interdependence characterizes EU AI policy?

Regulatory interdependence can take a range of forms. More lax rules elsewhere in the world - say, about the quality or character of training data sets - can mean that EU rules lack bite or will be hard to enforce. A strict EU regime may also disadvantage European firms through hampering their development or raising their operating costs compared to competitors. And mismatched rules may simply mean that algorithms developed in one jurisdiction cannot be used in another.

None of these effects - regulatory ineffectiveness, competitive disadvantage, and market fragmentation - are necessary outcomes of EU rules being different from those elsewhere. In some AI applications, regulatory divergence will have a big impact, meaning that regulatory interdependence is high. In others, the consequences of divergence are bound to be much lower, with no real price tag attached.

The key is therefore to understand just what regulatory interdependence in AI looks like - how it matters to Europe what other regulatory powers do - and how this interdependence is variegated across different AI systems and applications. This knowledge is essential to develop smart regulatory policy, which internalizes the global technological, commercial and regulatory force field within which the EU operates here. In some instances, the EU may be well-advised make concessions to other regulatory powers in the name of cross-border rule harmonization; in others, it would be better for the EU to stick to its guns. The aim of this subproject is to theorize and map the regulatory interdependence the EU confronts in AI, in order to help European societies to maximize their sovereignty over how AI is developed and governed in the Union.

go to project IIIIIIIVV