A conversation with Kai Zenner

I sat down for an hour-and-a-half in early August with Kai Zenner, longtime policy advisor to MEP Axel Voss. Brussels had emptied for the summer, and we sported shorts and t-shirts at an outdoor café in a quiet Place du Luxembourg. We talked about his background in EU digital policy, and in particular about the future of AI, in Europe and beyond. A recurring theme was the EU’s need to think globally and strategically — rather than along nationalistic or ideological lines — in order to compete and stay relevant. Here’s a summary of our conversation.

Forays in digital policy

Kai grew up a Star Wars fan in a tech-friendly family. After studying law and political science, he found himself working on digitalisation at the Konrad-Adenauer Stiftung — the political foundation of the European People’s Party’s (EPP’s) biggest member party, Germany’s Christian Democratic Union (CDU) — before meeting MEP Voss and later winning him as a regular speaker for conferences, in particular for an event series called ‘Think Digital’. In 2017, he started working for MEP Voss in the European Parliament, where they quickly became pioneers on AI, monitoring the work of the High-Level Expert Group on AI, organised by the European Commission, before starting themselves to draft AI policies. MEP Voss became rapporteur for a legislative own-initiative report on AI and Liability, then rapporteur of the AI special committee (AIDA) and then shadow rapporteur on the EU AI Act.

Kai describes his boss as a ‘hands-off macro-manager’ who determines the general policy line himself but then empowers his team not only to delve deeply into a topic but also to represent him publicly. Kai was able to spend several months at a time, for instance, before and after the 2019 European Parliamentary election cycle, pouring over long reports by the European Parliamentary Research Service (EPRS), Directorate-General for Internal Policies of the Union (IPOL) and European Commission, and by external actors such as think tanks, trade associations and even law firms. A European Parliament study by the Panel for the Future of Science and Technology (STOA) was especially helpful. Kai looks back on these months in the library as essentially accomplishing what ChapGPT has become so adept at doing instantaneously: collative evidence for effective argumentation.

It was unclear at the time which of several, groundbreaking pieces of EU legislation — notably, the Digital Services Act (DSA), Digital Markets Act (DMA), Data Act, European Chips Act or AI Act — would prove most important. With the advent of ChatGPT, Kai believes the latter may in the end have the biggest policy impact.

In 2021, MEP Voss’s team drafted almost 50 pages of content for the draft report of the AIDA committee in the European Parliament, including a digital policy road map for the European Union. The EPP Group, the largest in the European Parliament, was supportive from the start; no other Group could match it for time and expertise invested, and as a result around 75% of the draft report was quickly agreed as a common baseline, with the rest providing room for political compromise.

The process for the AI Act was similar. Together with fellow experts like MEP Eva Maydell, who has returned as EPP Group coordinator in the powerful Committee for Industry, Energy and Research (ITRE), MEP Voss and his team boldly led the way forward.

AI in the EU: Goals and constraints

Kai sees AI as different from other disruptive tech primarily because of its opacity, complexity, unpredictability and potential autonomy. As a 2020 Commission overview explained, AI represents for the first time a form of non-human intelligence able to compete with, or even outperform, humans in such a way as to incentivise the outsourcing of human decision-making altogether. This understandably worries people, since in a crisis, the allocation of healthcare, for instance, or other critical resources could well be determined by impersonal algorithms. To address these fears, the EU has uniquely insisted on certain safeguards: both ex-ante, for monitoring real-time decisions, as well as ex-post, for auditing or hindsight review. (Admittedly, ex-post course-corrections may help over time, say, in rooting out discrimination in social services — but prove cold comfort to those affected by ‘collateral damage’ following an AI-triggered missile strike.)

European Commission President von der Leyen reiterated in her 18 July 2024 Political Guidelines the EU’s intention to remain a cutting-edge rule-maker for cutting-edge technologies — or, to use a phrase from the European Commission’s ‘Digital Decade’ report, to remain ‘the global policy innovator’. But even putting aside chronic EU challenges like inadequate R&D funding or limited market scalability — not to mention military applications, which crucially remain Member-State competences and as such beyond the scope of the AI Act — the EU must grapple with several problems to stay competitive globally: overregulation, lack of skills and high energy prices.

Indeed, following from this, two quite pointed challenges loom large: access to data and raw computing power. President von der Leyen’s stated goal for a European Health Union, for instance, will require an ability to collect and employ Europeans’ data to a greater extent than is now being done: to feed the AI systems in turn needed to produce more effective medicines and better health outcomes generally. The 2023 Data Governance Act, the 2024 Data Act, the European Commission’s overarching Data Strategy: all these constitute an attempt, together with sectoral data spaces, to build up an EU Data Union. With regard to computing power, the European Commission in January 2024 proposed building up AI factories, essentially for supercomputing, as a way to help SMEs, among others. The Political Guidelines reiterates a plan for an ‘AI Factories initiative’ in the first 100 days of the new Commission mandate.

Unfortunately, Kai acknowledges, these essential building blocks may take years to fully materialise. So while the longer-term horizon offers promising signs for cutting-edge EU industries, the next 4-5 years may be tough: Europe’s companies and developers may struggle to comply with new rules even as new EU efforts to promote innovation or build up infrastructure remain relatively underfunded, in particular by Member States confronted by various other crises. In the meantime, the EU may continue to risk companies’ simply moving elsewhere for the greener pastures of more data, greater computing power, cheaper energy or indeed fewer regulations — or else, as we saw just this past July in Finland, with Silo AI’s being bought by Advanced Micro Devices, Inc., may continue to see European companies taken over by big American firms.

Geopolitics and -economics: The EU role in AI

Despite these challenges, Kai believes the EU has a unique opportunity for global leadership on questions of values and ethics. The November 2023 meeting in which Presidents Biden and Xi agreed to a treaty on AI weapons systems signalled how vital even such rivals as the US and China see the need for common guardrails. This is because, by removing the kinds of human oversight which more than once prevented nuclear holocaust during the Cold War, humankind may face increased risk of catastrophe — not just with missile systems per se but, even more insidiously perhaps, in the form of paralysis to infrastructure like water or hospitals. For even the most advanced, most sophisticated, most internally organised states in the world are not able, comprehensively and simultaneously, to strike a rival’s internal systems (military, cyber, education, media, etc.) — but we can imagine an AI that could. Perhaps our best hope, ironically, Kai argues, is for a new cold-war paradigm in which superpower rivals, each afraid of mutual destruction, enter again a new technological standoff and stalemate.

The EU must be clear, Kai believes, about the stakes of this contest. Despite the attention often given to EU-US disagreements, including over the latest attempts to ensure and protect transatlantic data flows, the reality is that Chinese devices in the EU market send back data for AI training in China — even as the EU has not benefitted from data flowing the other way. AI will increasingly even enable a new threat of digital genocide, whereby powerful actors build such data-set monopolies that whole histories, whether of people groups or eras, become vulnerable to deletion: in effect, the modern, digital equivalent to the burning of ancient libraries.

This is where the AI Act’s embedded ethical values are so important, Kai argues. The EU has said clearly, ‘We need data governance. We need technical documentation.’ The EU’s open-source initiative continues to provide another essential safeguard, which has already pushed some major players to release their models publicly.

As a next step, Kai thinks the EU should work to partner with non-EU companies, countries and institutions interested in creating more trustworthy AI systems. NATO and the OECD are arguably two of the most active and effective international communities in which such efforts might pay off. Israel, too, has a vibrant AI start-up community with which the EU should more closely cooperate. The European Commission’s role would be to vet and publish a list of such partners — whose values and commitments align with EU interests — from which Member States or EU companies could choose to work.

An even more sensitive issue: the EU must also think more strategically about how to maximise its overall advantage. Here ‘European champions’ will play a key role: in defending and promoting EU values and interests globally. No Member State, of course, will ever be able to get 100% of what it wants in protecting its own companies’ revenues or jobs; but every Member State should be able to accept an all-EU strategy whereby everybody wins sometimes and the Union as a whole benefits. As for weapons procurement, so too for cloud providers: Member States are better off sticking together and working as a united EU in the context of other established global frameworks. Neither individual Member States nor even the EU as a whole has the hard power or moral authority to go it alone or insist that everyone else play by its rules.

The EU, ultimately, can build up a competitive edge in targeted areas: not so much in competing with China or the US for the biggest horizonal platforms — business-to-consumer: to replace Amazon, say — since only 5-6 such platforms may be needed, but rather in areas like specialised robotics. Germany’s SMEs, for example, are global leaders in making robots which perform particular surgeries. Strategically, the EU should work to identify US companies offering customers insight and flexibility — and which crucially fill gaps Europe is unable to — and should then develop corresponding niche products. This would be a win-win, as also the European Digital SME Alliance and others trade associations have been pushing for. It would also require putting aside intra-EU rivalries, especially among the biggest Member States, in favour of a smarter EU strategy. Hopefully this can be achieved the easy way, politically: without requiring new crises to make progress.

Hope and change

All this requires trust, both within the EU as well as between the EU and its partners and allies. Such trust could ultimately mitigate against a new push for EU tech sovereignty and opt rather for the guiding value of resilience. The EU has come a long way already, in the past years: from an early insistence that it should be European companies making the web browsers and social media platforms used by Europeans to a recognition that this is simply not realistic. As with 6G now, the EU will need non-European partners — even Huawei, in the end, if the EU can effectively define and control its own most critical infrastructure. But, as ever in the EU, such debates must be pragmatic rather than emotional; and policy must keep up. The new world of AI will not wait.