In our previous blog, we asked what EOSC is, and four guests of Inside EOSC [Spotify, Apple Podcasts, Deezer], the podcast produced by the LUMEN project, gave us four different answers. Episodes 5 to 8 take a different route. Host Andrew Dubber talks to two people building LUMEN and two people who helped shape EOSC from CERN. Artificial intelligence (AI) was not the main topic of any of these conversations. Yet it came up in all four.
Julien Homo [Spotify, Apple Podcasts, Deezer], CEO of FoxCub and technical coordinator of LUMEN, sees AI chat tools as a real step forward, because they make exploring knowledge far easier. But they also show bias, sometimes give wrong answers and need to be clearer about their sources. His answer is to pair them with structured knowledge, such as knowledge graphs and ontologies, which can make their answers more precise and easier to trust. LUMEN puts this into practice: it is building a chatbot that other research communities, mathematics included, want to reuse, and a shared way of describing data so that automated tools can spot links across disciplines. For Julien, there is no single best way to represent knowledge, only the right balance between approaches.
“…we need other structures to help this model to be more relevant, to be more precise, to improve the trust.” – Julien Homo
Enrica Porcari [Spotify, Apple Podcasts, Deezer], CERN’s first Chief Information Officer, did not come from physics. A social scientist and former park ranger, she spent much of her career in humanitarian technology, most recently at the UN World Food Programme. One question has guided her throughout: who is this technology for? AI has been used at CERN for many years; what is new, she says, is that it is now in everyone’s hands. Technology can support us, but people must stay in charge of it. The same thinking drives the Open Quantum Institute, piloted at CERN, which is preparing for quantum computing now rather than later. As she points out, nobody expected ChatGPT to arrive so fast. For her, the answer is to shape technology on purpose, not leave it to chance:
“I think we have a responsibility as technology developers, as being in one of the premier European organizations and beyond. I think we have a responsibility to change the narrative, to really give hope to new generations, but not only, that if we are deliberate in our efforts to make technology work for us rather than the other way around, there is a positive future ahead of us.” – Enrica Porcari
Yann Le Franc [Spotify, Apple Podcasts, Deezer], founder and CEO of eScience Data Factory, head of secretariat at EUDAT and leader of LUMEN’s Work Package 2, explains what knowledge graphs and ontologies actually do for AI. A knowledge graph connects pieces of information; an ontology gives those connections an agreed meaning. In simple tests he describes, adding both to an AI tool produced correct answers 95 to 97% of the time. Yet he will not call a knowledge graph “true”: it is a database, and if the data going in is wrong, so is the graph. He is also the clearest sceptic of the four. A neuroscientist by training, he points out that computers were modelled on the brain, not the other way round, and sees today’s AI chat tools as “statistical machines” rather than intelligent ones.
“…what it provides to the AI is structured data. It’s a way to structure the data and to make it easier for the AI to get more context.” – Yann Le Franc
Bob Jones [Spotify, Apple Podcasts, Deezer] Special Envoy to the EOSC Federation and, since the recording, Chair of the new EOSC Node Coordinators Committee, has seen a technology leave the lab before. He worked alongside Tim Berners-Lee at CERN as the World Wide Web took shape, and knew it had gone beyond research when he saw “WWW” on the side of a bus. He does not expect EOSC to follow it there: it is data infrastructure, and most people will never notice it. But it matters for AI. Asked whether AI, having swallowed up the internet, has made responsible data practices a lost cause, he says trustworthiness still matters. By making high-quality, carefully curated data available for reuse across disciplines, EOSC gives AI tools better material to work with.
“So if you put good data in, the chances of you getting more trustworthy results and outputs from the AI tools increases.” – Bob Jones
All speakers seem to coalesce in the same destination: AI still needs good data and people to guide it. That makes the quieter work of EOSC more important, not less: connecting research communities and making well-described data easy to find and reuse. And for all the talk of risk, the mood is optimistic, Enrica and Bob said so themselves.
Until our next blog, feel free to listen to all of these conversations on Inside EOSC [Spotify, Apple Podcasts, Deezer].
This blog was written by Fotis Mystakopoulos, Open Scholarly Communication Officer at OPERAS and task lead of T8.5.
