Europe won’t dictate the terms on Global AI

Two days ago, the European Parliament came to an internal consensus on a draft version of the European Union (EU) Artificial Intelligence (AI) Act. Following this development, horse-trading will intensify among the European institutions to settle on a final legal text. This new phase of consensus building between the EU Parliament and the EU Council (with the EU Commission shuttling in-between) – the so-called trilogues – is an elaborate policymaking dance rivalled only by the elegant denseness of the Roman Curia’s Praedicate Evangelium.

The EU’s risk-based approach to AI regulation.
Source: EU Commission

The hope of the proponents of the EU AI Act is that it will become a global blueprint for how to realise the full benefits of AI without losing sight, and eventually guard, of its potential downside risks. The tendency of other parts of the world to follow EU thinking on how to regulate complex technical areas has led to a term: “Brussels Effect“.

Scenarios of how the EU AI Act will influence global corporate conduct.
Source: Siegmann & Anderljung (2022)

No context has experienced the Brussels Effect in so concentrated a manner as internet privacy and data protection. The EU’s General Data Protection Regulation (GDPR) is widely acknowledged to have shaped the compliance culture of companies and government agencies far beyond Europe.

Impact of the EU GDPR on the profits and sales of a sample of companies from 61 countries and 34 industries worldwide.
Source: Chen et al (2022), Oxford Martin School

The question on everyone’s mind then is, would the EU AI Act do the same?

As things stand today, many popular AI systems don’t measure up to the proposed EU standards. Source: Stanford CRFM

Reports that Parliamentarians in Ghana might initiate legislation on AI, following extensive stakeholder engagement in various African countries funded by Germany’s GIZ, outlines one obvious path through which the EU’s influence in such matters spread: donor relations and international development aid.

My more radical friends in the Civil Society movement promptly dismiss such channels as just another neocolonialist ruse. The more charitable ones reference dated theories of “normative imperialism“.

A more nuanced view of the Brussels Effect however focuses rather on its utility, the practical problems it solves for those societies around the world choosing to delegate regulatory thinking and standards setting to Europe. As pragmatists like to say, there is no genius in reinventing the wheel. If Europe has the will and wherewithal to do the heavy-lifting of tackling such complex, and protean, novel challenges such as how to tame AI for the rest of the planet, why protest?

In a preprint short essay, I explore this theme in some depth.

As is now widely acknowledged in the literature, the EU’s success in exporting and diffusing GDPR was a special case born out of exceptional circumstances, and not the sign of a general, not to talk of universal, trend.

GDPR capitalised on broad longstanding conventional wisdom about the desirability of more privacy and better data protection to build its momentum. The benefits of lax and loose privacy and data protection policies have rarely been collated into a definite counter-proposal against which GDPR had to contend. There is limited contention about the importance of good privacy and data protection rules and what they look like. The EU only needed to give heft to projects and processes scattered in bits and pieces around the world, and supported by a fairly expansive bedrock of popular and elite sentiments.

My argument, in the essay referenced above, is that emerging technologies with potentially vast benefits and a profoundly uncertain trajectory present unsurmountable barriers to the Brussels Effect.

I predict that not only will countries and companies hesitate in taking strong cues from the EU regarding how to design and respond to compliance mechanisms, but that we will also see distinct efforts to dilute the EU’s moral influence in the AI ethics & regulation domain.

Global approaches will splinter along various axes through, as India is doing, calls for genuinely multilateralist solutions, which will typically slow down any serious action as the technology races forward; and, as many others are doing, resistance movements against the EU’s signature approach of use-case classifications and proscriptions.

1 Comment

  1. I agrée with most of your ideas. GDPR was easier to implement because the pieces were already there. Global sentiments on data privacy and allied matters were generally aligned and the monetization models are fairly mature now.

    I will throw my hat in the ring and predict that new axes of collaboration will form based on relative positions in the AI dominance race.

    Why I think this new attempt by the EU may not fail as badly is that without firm centres around which to permanently coalesce, multilateralism may end up like pan africanism. The concept may be powerful but the actual effect may be limited. Even if BRICS for example were to become a viable alternative to the current order, how likely would it be that China and India agree on AI standards? This may allow the EU effort to stand, even if on wobbly legs.

    I’m new to your site but thank you for such a thought provoking piece.

Leave a Reply