The Dalle Molle Institute for Artificial Intelligence participates in the Swiss AI Initiative

The Swiss AI Initiative, created by ETHZ, EPFL and CSCS, has set out to make Switzerland an international hub for the development and implementation of transparent and reliable artificial intelligence. The Dalle Molle Institute for Artificial Intelligence IDSIA USI-SUPSI is one of the research centres involved in the project.

At the heart of the Swiss AI Initiative is the Alps supercomputer, which will be operational in the spring of 2024 at the Swiss National Supercomputing Centre (CSCS) in Lugano. The computing power provided by NVIDIA's new Grace Hopper Superchip will make Alps one of the world's most powerful supercomputers in the field of artificial intelligence.

Alps will be used to develop and train new Large Language Models (LLM) capable of competing with those of private industries such as OpenAI (ChatGPT's company), Microsoft, Google and Meta. Based on this common infrastructure, several foundational models of artificial intelligence will also be developed for specific tasks in sustainability and climate change, health, robotics and computer vision, education and science in general.

The Swiss AI Initiative aims to provide Switzerland with a significant competitive edge globally while also developing LLM transparently and openly. This means making both the data used to train these language models and the results obtained from them publicly available while ensuring compliance with legal, ethical, and scientific requirements. Such information is typically not accessible for language models created by private companies.

The Swiss AI Initiative is a collaborative effort between several Swiss academic institutions, including the two Federal Institutes of Technology and IDSIA USI-SUPSI. The initiative also involves the participation of international researchers who have been invited to contribute to the development of a multilingual open-source LLM.

The Dalle Molle Institute for Artificial Intelligence's contribution concerns the development of the shared infrastructure and some specific projects. In particular, with regard to language models focused on text processing, IDSIA will deal with activities aimed at increasing the reliability and transparency of data and reasoning models used and aspects of "alignment" in interaction with human beings. As far as specific projects are concerned, under the leadership of Dr Fabio Rinaldi, IDSIA is focusing its activities on fundamental health models in the working group led by Professor Rätsch (ETHZ) and Professor Salathé (EPFL).

Andrea Rizzoli, Director of IDSIA USI-SUPSI, said: "This is a strategic project that will benefit not only Switzerland but also the entire scientific community through the sharing of methods and results. The LLM sector is currently dominated by private players who share very little information on important aspects, such as the data used for training and the algorithms employed, for commercial reasons. Therefore, it is necessary for public research to catch up and enable responsible development of artificial intelligence."

The IDSIA team currently involved in the Swiss AI Initiative is composed of director Andrea Rizzoli and scientific director Marco Zaffalon, professor Fabio Crestani and researchers Fabio Rinaldi, Marco Forgione, Sandra Mitrovic, Oscar Lithgow Serrano and Joseph Cornelius.