On Monday January 26, the Information Systems Department (DSI) organized an event for staff on campus entitled "Let's get talking about AI" to share experiences and use cases of various artificial intelligence tools at the Institut Pasteur. The aim was to explore current knowledge of these digital tools, their strengths and weaknesses, what they can offer and where they fall short. The event was also an opportunity to focus on some of the key principles upheld by the Institut Pasteur, such as confidentiality and sovereignty, and to provide reassurance.

Implementation of the Institut Pasteur's strategy on generative artificial intelligence and reassurance

o    Adopting AI to reduce the digital divide

The main goal of those involved in implementing the use of AI at the Institut Pasteur is to democratize access to these new tools for everyone on campus. The initial aim is to provide as many staff members as possible with opportunities to understand, use and become proficient in AI tools, so that we can avoid a divide opening up between those who know how to use these technologies effectively and those who don't. Generative AI represents a revolution in accessibility as it makes AI available via natural language, without the need for technical skills. It is crucial that we all have the opportunity to use it.

o    Confidentiality and sovereignty: a hybrid architecture

Given the uncertain geopolitical situation and the need for confidentiality, the DSI has developed a hybrid architecture for the AIs hosted within the Institut Pasteur's infrastructure, based on three key principles:

• Confidentiality: Sensitive data remain in house. The LibreChat infrastructure uses local models so that your data never leave the Institut Pasteur. Depending on their needs, users can also choose to use LibreChat with cloud models that do not offer this guarantee.

• Sovereignty: Independence in light of current geopolitical upheavals and maximum service continuity.

• Optimized costs: A pay-per-use model for occasional use, with costs of intensive use recouped via a local infrastructure.

• Reassurance: The use of local AI models on LibreChat is entirely confidential. The Institut Pasteur infrastructure is currently also used by various other academic institutions (the Paris Brain Institute, Inria and the Imagine Institute) precisely because it guarantees confidentiality, which is a prerequisite for these institutes. Users can only access LibreChat if they complete prior training that covers the confidentiality issues related to cloud models and the way in which locally hosted models can offer a solution. The Institut Pasteur's approach therefore involves supporting and raising awareness among the entire community.

Find out more about the Librechat plateform

Find out more about the uses of generative AI (in French)

An approach based on use cases with high added value

One aim of the event on January 26 was to identify AI use cases that generate the most added value for research and support services, by looking at exploratory use cases developed with AI tools (Reviewer 2, Grants to Sponsorship, Press Kit, Whiteboard, etc.). This pragmatic approach can help us determine specific ways in which AI can change our practices.

Image

The Information Systems Department would like to thank you for coming along to this morning on artificial intelligence.

More than 200 of you attended the various talks and shared your views. 

For those who were unable to join us in person, videos of the presentations are available here.
Building a user community

Those involved in implementing AI on campus would like to set up a community based around the use of AI as a way of suggesting and sharing uses and encouraging mutual learning and open innovation. We can all create specialist assistants (research, HR, finance, communications, etc.).