At Rise of AI 2023 in Berlin, a panel with Kai Zenner, Jörg Bienert, Jonas Andrulis and Amira Gutmann-Trieb discusses whether Germany and Europe can lead in AI given the pending EU AI Act. Zenner, working on the AI Act negotiations for a European Parliament office, criticizes the Commission's original proposal as a horizontal framework applied too literally across sectors, and describes the Parliament's Article 8 fix giving high-risk obligations a sector-specific interpretation, and its unfinished attempt to close a loophole that let foundation-model providers like Google escape compliance while startups bore the burden, adding that much remains open. Andrulis, founder of Aleph Alpha, argues Europe's risk is hesitation, not technology, since his company already built multimodal and explainable language models but lacks the capital of US competitors like Microsoft-backed rivals. Bienert, president of KI Bundesverband, argues an early Council proposal treated general-purpose AI as automatically high-risk, calling this impractical, and describes a project building a shared European compute platform for research institutions and startups. The panel agrees Europe should build a 'trustworthy AI' brand but warns of repeating GDPR's slow, fragmented enforcement, and audience questions push back on whether 'European values' AI is actually a selling point abroad.
Key takeaways
- Kai Zenner argues the EU AI Act's original horizontal framework tried to cover every AI use case with the same rules, which the Parliament is narrowing sector by sector.
- Jonas Andrulis of Aleph Alpha says his company built multimodal and explainable language models before major US competitors, but must bootstrap without a big-tech backer's capital.
- Jörg Bienert describes a shared European compute platform project meant to give startups and researchers access to compute they could not afford alone.
- Panelists warn that a poorly enforced AI Act could strengthen dominant US companies rather than protect European startups, echoing criticism of GDPR's slow enforcement.
- The panel debates whether 'trustworthy AI made in Europe' can become a real competitive brand, pointing to some US firms that already market GDPR compliance.
Detailed notes
Positioning against an AI Act draft seen as too broad
Kai Zenner, who works on the AI Act negotiations for a European Parliament office, argues the European Commission's proposal was flawed from the start because it tried to build one horizontal framework covering everything AI touches, from decades-old expert systems to foundation models, across transport, health and other sectors. He says the Parliament's new Article 8 addresses this by making high-risk obligations, such as Article 14's human oversight 'stop button', depend on the specific sector rather than applying literally everywhere. He also criticizes the Commission's original text for regulating only 'downstream' actors, meaning a startup building on a foundation model would bear the full compliance burden while the foundation-model provider, such as Google, stayed outside the AI Act.
Startups versus big tech: Aleph Alpha's case for speed over caution
Jonas Andrulis, founder and CEO of Aleph Alpha, says his company built a GPT-2-style German-language model before GPT-3 was released and added multimodality two years before GPT-4, and calls his team the first with explainability, giving it traction with health, legal, finance, government and security clients. He argues Europe's real risk is hesitation: unlike a company backed by Microsoft's capital, Aleph Alpha has to bootstrap, and he worries less about compute availability than about whether German and European partners can move fast enough to adopt the technology.
Building shared infrastructure to compete with the US and China
Jörg Bienert describes a project building a shared compute platform to give research institutions, startups and industry access to expensive compute power, intended as a European 'crystallization point' so the ecosystem can cooperate rather than compete internally while facing competition from the US and China. He also criticizes an early Council proposal that classified general-purpose AI as automatically high-risk as impractical, arguing high-risk obligations should instead attach when such a model is deployed in a high-risk application, and calls for faster, more flexible rules given how quickly the technology changes.
Can 'trustworthy AI made in Europe' be a selling point?
The panel discusses whether 'trustworthy AI made in Europe' could become a competitive brand, similar to how some US companies now market GDPR compliance. Zenner says this depends on the AI Act delivering fast, harmonized enforcement and guidance, warning that GDPR's years-long wait for certification and inconsistent enforcement across member states should not be repeated. Andrulis says Aleph Alpha's enterprise clients, including major banks, care about trustworthiness and auditability regardless of regulation, and that its focus on transparency and control rather than dictating 'correct' outputs suits business customers rather than consumer products.
Q&A highlights
Audience members challenge whether 'ethical AI made in Europe' is a real selling point abroad, and whether the AI Act's late addition of extra rules for generative AI creates dangerous last-minute uncertainty. Panelists respond that legal certainty and enforcement, not the values framing itself, will determine whether SMEs and startups can afford compliance, and warn that a fragmented, ambiguous AI Act would mainly entrench large companies like Google and Microsoft, which can absorb compliance costs and fines that smaller German companies cannot.
Quotes
The starting point of the Commission is just completely wrong.
Kai Zenner
Our strategy has to be collaborative. Our strategy only works if we can empower governments and enterprises to build a strong position themselves.
Jonas Andrulis
Questions answered in this talk
What was wrong with the European Commission's original AI Act proposal, according to the panel?
At Rise of AI 2023, Kai Zenner argues the Commission tried to apply one horizontal framework literally across every AI use case and sector, and initially left foundation-model providers outside the AI Act while leaving downstream startups fully liable for compliance.
Why does Aleph Alpha's founder see hesitation as Europe's biggest AI risk?
Jonas Andrulis says Aleph Alpha already built multimodal and explainable language models ahead of major competitors, but without a big-tech backer's capital it depends on European partners and enterprises moving fast enough to adopt the technology.
Could 'trustworthy AI made in Europe' become a competitive advantage?
Panelists at Rise of AI 2023 say this depends on the AI Act being enforced quickly and consistently across the EU, citing GDPR's slow, fragmented enforcement as the risk to avoid, though one audience member questions whether it is actually a selling point outside Europe.