Breaking barriers with AI: Helmholtz AI consultants at the Falling Walls Science Summit 2024
The Falling Walls Science Summit, hosted from November 7 to 9 at the Falling Walls Science House, continues its mission of dismantling barriers in science and driving transformative change. Several Helmholtz AI consultants made an appearance, sharing their expertise and perspectives during multiple thought-provoking sessions. Their discussions highlighted the critical role of artificial intelligence in advancing scientific discovery while addressing its challenges and opportunities across diverse fields.
The Falling Walls Science Summit, held annually since 2009, is a global gathering designed to foster innovation and collaboration across disciplines. The 2024 event, hosted from November 7 to 9 at the Falling Walls Science House, continues its mission of dismantling barriers in science and driving transformative change. The summit assembles a diverse group of participants, including leading researchers, business innovators, policymakers, media professionals, and civil society representatives, all contributing to discussions on the future of science and its societal impact.
Several Helmholtz AI consultants made an appearance at the Falling Walls Science Summit 2024, sharing their expertise and perspectives during several thought-provoking sessions. Their discussions highlighted the critical role of artificial intelligence in advancing scientific discovery while addressing its challenges and opportunities across diverse fields. Below, our Helmholtz AI consultant heads share their reflections on key themes explored at the summit:
Stefan Kesselheim
Helmholtz AI Consultant Head “Information”
"
I had the pleasure of taking part in the circle discussion with Eva Unger from Helmholtz Berlin and Pepe Marquez from FairMat. We discussed the challenges of data, especially in the field of material science. Here, very heterogeneous data is distributed among many labs, and this makes it very difficult to use for foundation model training. FairMat is one of the national research data infrastructures, and has invested greatly into improving the available data, and metadata to scientists all over the world. Generating large-size high-quality datasets remains a big challenge, but with new paradigms and tools, and joint efforts from communities, this will continue to improve."
Marie Piraud
Helmholtz AI Consultant Head “Health”
"
I was part of the session "The Potential of Foundation Models", which was organised in the framework for the Helmholtz Foundation Model Initiative and had a 'Circle discussion' with Loic Lannelongue on the "Sustainability of Foundation Models", with Tobias Schmid as moderator. Sustainability aspects are usually overlooked when it comes to AI. We discussed the environmental impact of AI and foundation models at all stages of their development and application, and in particular their large energy consumption, the currently available tools and the regulations in Europe (or rather the absence of) as well as future trends."
Peter Steinbach
Helmholtz AI Consultant Head “Matter”
"
The session "FM and Data Pipelines" jointly with Wolfgang zu Castell was engaging. The panel quickly focused in on the hallmarks and challenges that current day data aggregation and sensor technologies offer: we have tons of data, but need to get our data management straight. While the former is bluntly put, this tension ranges from data volume (how do we transfer petabytes) to data access (personal rights protection especially in the medical sector), to technology skill (how to ensure, data handling does not get in the way for science), and data quality (does the data offer enough diversity to produce a high-quality FM) and much much more. At Helmholtz, we appear to have good infrastructure in place to tackle most of these challenges outlined above while in other places, we need to make structural or cultural advances. However, both Wolfgang and I concluded on a positive tone that we are ready for Helmholtz Foundation Model Initiative and hope to approach societal challenges by providing good data and training useful models!"