ETH Zurich and EPFL enhance collaboration to boost AI in Switzerland

The coming years will bring tremendous change as AI modelling advances rapidly and expands into new areas, profoundly impacting both society and the economy. Generative AI (GenAI), exemplified by technologies like ChatGPT, has the potential to significantly boost productivity. Nations across the globe are investing heavily in advanced AI technologies and ultra-fast computing infrastructure. Over the next five years, it will be crucial to establish high standards for safe and trustworthy AI. Through SNAI, Switzerland has a unique opportunity to showcase its talent and support public, private, and nonprofit organizations in this endeavor.

ETH Zurich and EPFL are therefore stepping up their efforts to set Swiss AI research at the international forefront. With the approval of ETH Zurich and EPFL executive boards, a new institute has been established: The Swiss National AI Institute (SNAI). SNAI will provide a national perspective on AI-based research, education and innovation. Through coordinated efforts, SNAI will put transparency, open source and trustworthiness at its core. The new institute draws on the expertise from various AI research disciplines as well as  on the two existing AI research centers at ETH and EPFL (ETH AI Center and external page EPFL AI Center).

SNAI will support the development of the Swiss AI initiative, whose funding has been secured recently: the ETH Board, the strategic management and supervisory body of the ETH Domain decided to allocate CHF 20m to the Swiss AI initiative for the years 2025-2028. In addition to funding from the ETH Board, SNAI and its projects will be financed by contributions from ETH Zurich and EPFL as well as third-party funds.

Shaping a Swiss-made GenAI

With SNAI and Swiss AI, there is an opportunity to not only accelerate national digital transition but also to contribute to the global development of this emerging field by building large models with over 50 billion parameters. Developing such advanced models requires access to sophisticated, powerful supercomputers and specialized AI expertise.

Leveraging its integration with the ETH AI Center and the external page EPFL AI Center, the new institute will benefit from the expertise of over 70 AI-focused professors across Switzerland from the get go. Together, they will develop Switzerland's first national foundation model for languages as well as other foundation models to support their research goals. The model will be aligned with Swiss values, such as trustworthiness, open source and transparency, and will be tailored to the needs of Swiss stakeholders.

With estimates suggesting that only a few hundred people worldwide currently possess the necessary AI expertise to develop the largest foundation models, SNAI will focus not only on research but also on training AI specialists for academia and industry to increase the talent pool, benefitting the local economy.

A strong long-term commitment to AI in Switzerland

"By jointly launching SNAI, ETH Zurich and EPFL are making a strong long-term commitment to empowering and promoting AI in Switzerland. We strive to create a research environment that is capable of establishing Switzerland as a location for inclusive, reliable, transparent, and trustworthy AI," says Christian Wolfrum, ETH Vice President for Research. ¡°Young talents in particular will benefit from this diverse research ecosystem and from the latest world-class infrastructure."

"SNAI will address AI opportunities and challenges of national importance that require the collaboration of many researchers across our country. The results of this work will strengthen the competitiveness of Switzerland in AI research and development. Eventually, this will enable us to massively expand our AI expertise in Switzerland and beyond," says Pierre Dillenbourg, EPFL Vice President for Academic Affairs. ¡°Our efforts will pave the way for new applications in favor of Swiss industry and society at large¡± he adds.

SNAI accelerates the momentum of the press release of 4 December 2023) and includes researchers from over 10 academic institutions across Switzerland. It was seeded with a computing power of 10¡¯000¡¯000 GPU hours to develop large-scale AI foundation models (which are the base of modern GenAI) and to deploy them in core areas of Swiss society, such as health care, sustainability, science, education, robotics, and augmented reality. Going forward, the Swiss AI Initiative will be operated by SNAI.

Accordingly, the researchers will open-source their models whenever possible, enabling SMEs and start-ups to use them. With its strong focus on open-source and domain-specific foundation models, SNAI pursues the vision of positioning AI in Switzerland at the international forefront of ethical and legal standards, transparency, and trustworthiness.

¡°SNAI aims to stimulate collaboration across institutions and research areas to jointly explore the role of large-scale AI foundation models and advance them to function robustly and reliably in key societal areas,¡± says Ana Klimovic, ETH Zurich Professor of Computer Science and member of the initiative¡¯s Steering Committee. ¡°As researchers, we place great emphasis on transparency, openness, and trustworthiness. Also, we aim to develop infrastructure to train and serve large foundation models energy-efficiently¡± she adds.

Lean management and organizational structures

To maintain lean organizational and management structures, SNAI operates under a co-leadership model between ETH Zurich and EPFL. Supervisory, management, and research bodies are equally staffed by researchers from both institutions, including the boards of their respective AI centers.  Operational management will be led by Alexander Ilic, Executive Director of the ETH AI Center, and Scarlet Schwiderski-Grosche, Executive Director of the EPFL AI Center, who will serve as SNAI¡¯s Executive Co-Directors.

A powerhouse of AI, supercomputing and data science

During SNAI development, a collaboration with the Swiss National Supercomputing Centre (CSCS), which develops and operates Switzerland high performance computing and data research infrastructure ¡°Alps¡±, and the Swiss Data Science Center (SDSC) have been established. All centers will be tightly interconnected as AI leverages data science methods and requires the powerful infrastructure of high-performance computers. Beyond SNAI, all centers will also continue their own local programs and activities.