Second meeting of the DS&AI CoP

Published on: 27.03.2024
Author: Alex Lavrynets

The second official meeting of the federal administration's Data Science and Artificial Intelligence (DS&AI) community of practice took place in Neuchâtel on 20 March 2024. Discussions focused on generative artificial intelligence (GenAI) and large language models (LLMs).

The event

The second meeting of the DS&AI CoP was a success. The event brought together nearly a hundred members from the Swiss federal administration working in the fields of data science and artificial intelligence (AI). Participants were able to present their innovative data science and AI projects and share their technical and practical expertise. The main topics of the meeting were generative AI and large language models. The importance of these topics was highlighted at the first meeting of the DS&AI CoP on 14 November 2023. The official DS&AI CoP meetings are organised regularly by the Data Science Competence Center (DSCC) as part of the implementation of the Swiss federal data science strategy.

The programme

In cooperation with the Competence Network for Artificial Intelligence (CNAI), the DSCC outlined the scope of the meeting, emphasising the particular challenges involved in the use of large language models within the federal administration, from both a technical and societal point of view.

DS&AI CoP members provided the cornerstones of the programme with their valuable contributions.

The keynote speech came from the Federal Office for Defence Procurement (armasuisse), Science and Technology S+T, presenting the latest technical advances in large language models. This presentation reviewed the intrinsic dangers and limitations of these powerful models, which can be limited in certain contexts and often opaque.
The Federal Chancellery presented the fact sheet on the use of generative AI tools in the Federal Administration.
A representative from the legal hub attached to the CNAI presented some of the legal challenges associated with the use of AI in the federal administration.
An innovative projected based on large language models was presented by the Swiss National Bank (SNB).
The Federal Court's IT services presented several ongoing projects based on large language models and contributing to technical independence, digitalisation and digital sovereignty.
Nils Güggi, head of the Federal Supervisory Authority for Foundations (FSAF), presented the "Esi - the FSAF chatbot" project. The project consisted in developing a chatbot capable, thanks to AI, of answering customers' general enquiries on the basis of sources specified by the FSAF. The chatbot, currently in production, can be tested online.

The meeting confirmed how important it is to share experiences and good practices within the federal administration and was rounded off by a buffet dinner providing ample opportunity to network.

Context

Generative AI powered by large language models

According to the CNAI terminology v2.0, generative AI is defined as follows:

‘Generative AI’ is a broad term that refers to AI systems that are trained on large amounts of data from the physical and virtual world in order to generate data themselves (e.g. texts, imagery, sound recordings, videos, simulations, and codes). They are often multimodal, with input and/or output in one or several modalities (e.g. text, image or video).

Large language models like GPT-4 or Llama have billions of parameters and represent a sub-category of generative AI, which is based on natural language.

When using generative AI systems within the federal administration, the AI Guidelines and the Confederation's code of practice must be respected for a human-centred and trustworthy data science (and AI). Proper compliance with the principles defined is intended (among other things) to ensure public trust, the protection of individual rights and alignment with democratic values in a state governed by the rule of law.