At Rise of AI Summit 2020, DFKI deputy site director Aljoscha Burchardt and Bundestag member Mario Brandenburg (FDP) discuss two years spent together on the German parliament's Enquete Commission on Artificial Intelligence, and whether Germany can move past being a policy 'paper tiger.' Burchardt explains he joined the commission almost by accident, sent to a meeting with FDP leader Christian Lindner in place of DFKI's then head, and was later nominated by Brandenburg to the 100-member body split between politicians and outside experts. They describe concrete recommendations, including requiring large platforms to give researchers API access to study hate speech and social bots, and an unresolved, evenly split vote on whether AI technology is neutral. They discuss Germany's shift from sixteen regional AI clusters to five centers of excellence, a pay rule that discouraged DFKI from hiring international researchers, government-backed startup funding, and debates over classroom eye-tracking and platform data access. They close by trading wishes: Burchardt wants the state to become an AI early adopter, Brandenburg wants better long-term monitoring of the commission's recommendations and more direct science communication to the public.
Key takeaways
- Burchardt joined the Bundestag's Enquete Commission on AI after being sent in his former director's place to a meeting with FDP leader Christian Lindner.
- The commission recommended that large platforms give researchers and journalists API access to study hate speech and social bots, without agreeing on who should control that access.
- A commission vote on whether AI is neutral split three parties to three and was left unresolved in the final report.
- Germany's roughly sixteen regional AI research clusters have been consolidated to five centers of excellence, and a pay cap that discouraged international hires at DFKI has been lifted.
- Burchardt calls the loss of explainability in modern machine learning 'a sad thing' compared with earlier, more transparent symbolic AI systems.
Detailed notes
How a DFKI researcher ended up in the Bundestag's AI commission
Burchardt explains he joined the Bundestag's Enquete Commission on Artificial Intelligence almost by accident: DFKI's then head, unable to attend a lunch invitation from FDP leader Christian Lindner to discuss AI, sent Burchardt instead. Brandenburg, then looking to nominate an outside researcher rather than only industry representatives, later pushed to bring Burchardt onto the 100-member commission, split between 50 politicians and 50 nominated experts. Both describe the commission's two years as a process in which politicians' understanding of AI grew from vague impressions to concrete positions, and in which the group concluded that Germany is still in a phase of exploration rather than one ready for heavy regulation.
Platform data access and the limits of a European path
One concrete recommendation, Burchardt says, is that large platforms such as Facebook and Twitter should be required to give researchers and journalists API access to study phenomena such as hate speech and social bots, though the commission never settled who should control that access, debating whether it should sit with a state body, an independent regulator, or an NGO. Brandenburg adds that pushing everything toward 'made in Europe' also has limits in practice: internal debate over the German Corona-Warn-App's data storage showed that even a joint European approach to centralised data was politically unworkable, and the app ended up using decentralised storage instead.
Is AI neutral? A split vote in the commission
The two recall an hour-long, unresolved debate in the commission over whether AI technology is neutral, which ended in a formal vote split three parties to three and was folded into the report as compromise language. Burchardt's own position is that the underlying technology starts out neutral, but loses that neutrality once it is trained on data, because the values embedded in that data carry into the model.
Funding, pay rules and education
Brandenburg describes how a government-backed innovation fund, combining roughly 400 million euros in public money with about 600 million from private investors, emerged only after months of decoupling the idea from broader political framing, aimed at helping startups survive a second funding round. He notes that institutional investors such as pension funds still shy away from the space for political reasons. On research capacity, Burchardt points to a rule, since lifted for DFKI, that had capped AI researchers' pay at public-sector civil-service rates, discouraging international hires; Germany's roughly sixteen regional AI clusters have since been narrowed to five centers of excellence. On education, Brandenburg notes he had computer science lessons in school in 1994, yet many pupils today still finish school without any; the two also discuss, without resolving, whether classroom eye-tracking to flag inattentive students would be acceptable only if the feedback stays anonymous and immediate, never saved or shared beyond the teacher.
Explainability and a European edge in AI
Burchardt, speaking as a scientist, calls the retreat from explainable AI 'a sad thing': earlier symbolic AI systems were well understood but limited, while today's machine learning systems perform far better without researchers fully understanding how. He argues the level of explanation needed depends on the application, from none for a car's brakes to a detailed rationale for a doctor evaluating a treatment recommendation. Brandenburg, who sits on the boards of four B2B AI companies, says those companies focus pragmatically on what works rather than on policy, treat GDPR as a genuine competitive strength for user privacy in Europe, and see multilingual natural language processing, where English-only US models fall short, as another European edge.
Closing: a wish list for German AI policy
Citing an article that counted the words in the commission's report, Burchardt notes 'risk' appears roughly twice as often as 'chance,' reflecting Germany's generally cautious attitude toward new technology. Asked what they would ask of each other, Burchardt wants the state itself to become an early adopter of AI so companies feel pressure to follow, while Brandenburg wants regular follow-up monitoring of whether the commission's recommendations are actually implemented, plus more direct communication from researchers to the public rather than only to politicians, so that AI debates are not shaped only by people who are already convinced.
Quotes
We are not in the age of regulation yet, but we are in the age of exploration and getting to know this technology.
Aljoscha Burchardt
Everything has to be attached to some political sense or goal, and this is absolutely not matching to what's going on in the field.
Mario Brandenburg
This is really disappointing for us as a scientist, because you want to understand how things work.
Aljoscha Burchardt
Questions answered in this talk
What did the German Bundestag's Enquete Commission on AI recommend about platform data access?
According to Aljoscha Burchardt at Rise of AI Summit 2020, the commission recommended that large platforms be required to give researchers and journalists API access to study issues such as hate speech and social bots, though it left open who should control that access.
Is artificial intelligence neutral, according to the discussion between Aljoscha Burchardt and Mario Brandenburg?
At Rise of AI Summit 2020, Aljoscha Burchardt argues that AI technology starts out neutral but loses that neutrality once it is trained on data, since the values in the data carry into the model; a formal vote on the question split the Bundestag commission three parties to three.
How many AI centers of excellence does Germany have, according to Mario Brandenburg?
Mario Brandenburg said at Rise of AI Summit 2020 that Germany originally set up around sixteen regional AI clusters, one per federal state, which have since been narrowed down to five centers of excellence.