AI Safety and Security at the Pint of Science Festival: focus on energy efficiency and aviation
May 26, 2025
AI Safety and Security at the Pint of Science Festival: focus on energy efficiency and aviation
Pint Of Science Festival 2025
At the international Pint of Science festival, scientists in over 500 cities around the world report on their research in a relaxed atmosphere and offer exciting insights into the future.
The DLR Institute for AI Safety and Security gave two presentations at the international Pint of Science Festival in Ulm and Cologne in May 2025
In Ulm, Hans-Martin Rieser shed light on the enormous energy consumption of AI systems and discussed possible solutions for more efficient technologies
In Cologne, Charles Berro posed the critical question of whether Large Language Models are safe enough for use in air traffic control
Both presentations emphasised the social relevance of AI safety and security research and met with a positive response from the audience
Focus on AI's hunger for energy
In Ulm, where the Pint of Science Festival took place for the first time, Hans-Martin Rieser focussed on the pressing question of the energy consumption of artificial intelligence. Under the title "Hungry AI - can it also be efficient?", he shed light on a problem that is becoming increasingly relevant as AI applications become more widespread: The immense energy requirements of modern AI systems. The dimensions are impressive: a single image generated with AI tools such as Stable Diffusion or Midjourney consumes as much energy as ten hours of work on a laptop. Rieser not only explained the background to this high energy requirement to the audience, but also discussed possible solutions and the role of quantum technologies in the development of more efficient AI systems.
AI security in critical infrastructures
At the same time, Charles Berro presented another, no less important facet of AI safety and security research in Cologne. His presentation "Large Language Models in Air Traffic Control" dealt with the critical question of whether large language models such as ChatGPT or DeepSeek are safe enough to be used in air traffic control. This question goes to the heart of the work of the DLR Institute for AI Safety: while AI systems are already widely used in everyday life and perform impressively, significantly higher safety standards must be met for use in critical infrastructures such as aviation. Berro guided the visitors through the complex challenges that arise when integrating AI into safety-critical systems.
Communicating science in an understandable way
This year, the international Pint of Science Festival provided a platform for scientists from 27 countries around the world to present their latest research findings, free from dry theory and complicated technical terms. In a relaxed atmosphere, visitors were able to talk directly to the researchers and gain an understanding of complex scientific topics. The positive response to the presentations emphasised the social relevance of the DLR Institute for AI Safety and Security's research work. The institute works daily to develop methods and technologies for the safe and reliable use of AI, combining basic research with application-oriented developments while taking ethical, legal, and social issues into account.
