INSIDE EOSC 07 – Yann Le Franc
Andrew Dubber
Hi, and welcome to Inside EOSC, a podcast all about the inner workings of the European Open Science Cloud. I’m Andrew Dubber. I’m a senior researcher at the Industry Commons Foundation and a consortium member of an EOSC project called LUMEN. And every month I introduce, with no apologies made for the terrible pun, luminaries. People who are central to the LUMEN project and to the wider EOSC ecosystem.
Now today’s guest has been quietly at the center of European research data infrastructure for the best part of two decades. He’s the CEO and founder of eScience Data Factory, a French SME specializing in semantic web technologies and knowledge graphs. He’s the head of secretariat at EUDAT and within LUMEN, he’s the leader of Work Package 2, where his team is building a platform to help researchers make their ontologies FAIR from the start. Yann Le Franc, thanks so much for joining us.
Yann Le Franc
Thanks, Andrew, for having me.
Andrew Dubber
Fantastic. All right. There’s a load to unpack in there, and I’m going to go backwards. Let’s start with EUDAT. What is that?
Yann Le Franc
So EUDAT is a pan-European collaboration between Data and Compute Centre across Europe that aims at providing transnational services, data management services for the research community across Europe.
Andrew Dubber
Okay. And eScience Data Factory?
Yann Le Franc
So eScience Data Factory is a company I created 11 years ago now that works in the realm of data management for scientific data, but not only. We’re also taking care of data from industry and data from public administration. We are specializing in semantic web technology, ontology design, graph technology, and now we are also exploring the connections between AI and knowledge graph to add values for research, but also for industry.
Andrew Dubber
Okay. And finally, just to complete the set, Work Package 2 of LUMEN. What’s that about?
Yann Le Franc
So Work Package 2 aims at developing a platform for ontologists and taxonomists to actually more easily create their ontology and make sure that they are being FAIR from start when they create their ontology or to help them verify existing ontologies.
So I’ve been working in previous projects in the EOSC realm like FAIRsFAIR, where we created or proposed recommendations on how to make what we call semantic artifacts as a sort of overarching terms that encompass ontologies, formal ontologies, or simple taxonomies or code lists, so that actually we can share them more efficiently and work with them more efficiently by using FAIR. And it’s related to another project that just finished, FAIR-IMPACT, in which we are working on issues about mappings, which is how do you build bridges between these ontologies so that actually the different information system relying on these ontologies can work together.
Andrew Dubber
Right. I’m going to rely on the fact that most people who are listening to this know what FAIR means in this context, but we love our acronyms. Findable, accessible, interoperable, reusable. But you mentioned knowledge graphs and ontologies. And maybe not everybody knows exactly what an ontology is or what a knowledge graph does and how those things are useful. Do you want to sort of unpack that a little bit?
Yann Le Franc
So a knowledge graph is another way to represent data and their link compared to classical approaches where we use what we call SQL database, like tabular format, where we can create links between the tables so that you can create complex models. Knowledge graphs are used to actually do the same thing more efficiently, because the basis is a standard from the W3C consortium, which actually creates the web standards like HTML or CSS, and they created RDF, which is a data model that enables you to represent data under the form of a triple, where one side would be the subject, the other side would be an object, and there would be a relation that connects these two that’s called a predicate.
So basically you would have these triples that you can chain them all together to create this sort of tree of knowledge connecting the data together.
Andrew Dubber
Right. The way I’ve had that described to me is in the sentence, the boy kicks the ball.
Yann Le Franc
Exactly.
Andrew Dubber
Kicks is the thing that makes it a triple.
Yann Le Franc
Yeah. So the boy is the subject, the ball would be the object, and then the predicate would be kicks. And so you can create as many relations like this. And in the end, RDF allows you to represent data and the relation in a simple, natural language sentences. And so you can put that into a specialized database that you can query and retrieve the information more easily.
Andrew Dubber
Right. But I guess it gets more complicated and more multidimensional when the ball is owned by the girl and the boy borrowed the ball and the ball is brown and all these other different connections. To me, a graph is something on a two-dimensional sheet of paper with lines connecting bits. But when you start to unpack it like that, it really needs a computational space to be that sort of multidimensional.
Yann Le Franc
Yes. So you can create a multidimensional space. So we understand easily or visualize easily things on a piece of paper. Sometimes you can go 3D, but the graph is not limited in terms of dimensions. So you can create as many relations you want and as many dimensions. And you can even connect different graphs together.
So the first use of a knowledge graph is, for instance, representing social media data or Bob knows Jane and Bob lives in, let’s say, San Francisco and Jane lives in Paris and so on and so forth. So you can create these trees of connections between the people and it can be used for many other things.
Andrew Dubber
Is it useful to AI simply because it describes things that are true and that you can reference that to avoid hallucinations? Or is there something more complex than that?
Yann Le Franc
So I wouldn’t use the word true because a knowledge graph is a database. So if you input wrong data, the knowledge graph is going to be wrong. However, 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, right?
So for instance, we’re having this toy project in the company where we want to accelerate the way we do financial reporting for our European projects. And that requires connecting different bits and pieces of information within the company. And basically, we are starting to build a knowledge graph with that data and connect this to any LLM so that the LLM can answer simple questions like, what is the amount of money that has been spent on that project over that particular period? Let’s say in Work Package 2 in LUMEN, for instance.
But that can apply to very many, many different other contexts. In industry, you can use that for monitoring problems on an assembly line or maybe also address some issues because the data is basically structured and you can connect different sources of information together, which is really hard to do in the current space, right? Because in the companies, you have different information system for pay, for timesheets, and different aspects of the company. And so basically, the graph is a way to connect them together.
Andrew Dubber
I don’t want to go down a rabbit hole too much on this, but I think there’ll be a lot of people very happy with you if you did manage to solve European financial reporting for projects in a way that simplifies things.
LUMEN’s about connecting knowledge across different domains. So we’re talking about social sciences and humanities, but we’re talking about mathematics, and we’re talking about earth system science, and we’re talking about molecular dynamics. When you say you can connect knowledge graphs, is that what you’re talking about? You create knowledge graphs in these different domains, and you find points at which they overlap?
Yann Le Franc
They overlap. So that’s where it’s becoming useful to talk about ontologies. So knowledge graphs are built on RDF, and the main constraints of RDF is there’s no constraint. Basically, what you use to identify the subject and the object, and the predicate is simple URIs. And these URIs are just links that you can follow, but they don’t contain any semantics, any meaning.
And so the ontology is there to help you further structure the graph with meanings. And so one of the big issues is that you need to get an agreement within the communities about the concept they manipulate on an everyday basis in their specific fields, and in many communities they have a lot. I come from the biomedical domain and there you can find tons of ontologies that talks about disease, genes, relations with proteins and so on. In other domain, it’s starting. So people start building these, let’s say semantic models or knowledge models that represents all the concepts that they are manipulating and how they interrelate with each other.
And so for instance, again, coming from biomedical, one of the simplest ontology you can think of is the species classification. It’s a, let’s say, simple ontology. It’s called what we call the taxonomy. But still, it contains the semantics and it contains the logic that actually connects the different species together.
Andrew Dubber
Albeit in Latin.
Yann Le Franc
Yeah, yeah, yeah. And you can do a bit of language, so you’re not forced to do Latin.
Andrew Dubber
That’s a relief. I wouldn’t get very far if it was only Latin, that’s for sure. Is every relationship in a semantic context reducible to a triple, or are some things not so easily simplified?
Yann Le Franc
That’s a very good question. In ontology, so you have different levels of complexity. So you can serialize it as a triple, which means that you can represent for a computer, because the idea is that you model human knowledge for a computer, not just for humans, but you want the computer to be able to act on it because it understands the logic.
And when you go from classification where you have hierarchical relations, you can actually build more complex ontologies using another standard from W3C, which is called OWL, which allow you to create relation on top of which you can have mathematical functions coming from the logic part, the logic side. So it’s a first order logic or a descriptive logic for OWL. There are different favors you can add. So in the end, that’s a way to enrich the simple triple by adding mathematical functions that can be used by the computer to actually infer new knowledge of reclassify things based on the logic you integrate in the model.
Andrew Dubber
So what the ontologies are good for, at least one of the things, is it’s a perfect classifier.
Yann Le Franc
Right.
Andrew Dubber
Because I feel like there are some things that it’s not directly applicable to, or at least not in its most simple form. The example I gave, or that I often give in this context, I worked with a project briefly that was doing knowledge graphs around what I would describe as the rock music family trees, how different musicians are connected to each other. And a predicate that they were really keen on was this idea of has encountered. So the example that was given was Joni Mitchell has encountered David Crosby.
And I thought that doesn’t even scratch the surface of what that relationship was. And so it does, particularly in the realm of social sciences and humanities, it does lend itself to the question of just how much nuance can you get into a system of description such that it is useful to reveal more than just has encountered, but actually all those other things that, and the human relationships and the, humans are complicated. And I think that that’s really what I’m getting at. How much complexity can you get by simply layering triples?
Yann Le Franc
The thing is, when you design these systems, you have to start thinking how you want to use them first, right? And so I think that’s the best approach in the methodology. And there are many people using that. We call that competency questions, right? So if I ask you to build an ontology without giving you a scope or an idea of how far you need to go, you can build an ontology of the universe and spell your life and the life of your sons and so on without never getting to the full complexity of the system.
So you have to constrain that. Now, you can add nuances by adding new relations, right? So you can have the has encountered, but you can do that complexity with the triples. Of course, there would be limitations because triples are really basic. So that’s why you really need to work with this formal logic if you really want to represent things. But then you start entering a real complex nightmare because you need to be an expert in logic to understand how it works so that you can build an ontology that gives you what you want.
So just a thing, an ontology comes from philosophy. It’s not a computer science thing. It’s philosophy things. How do you represent things and how you model things? So again, it can be a daunting task to model the world or even just the realm of rock music because there’s so many things. So you have to, at some point, start simple and start by defining really what are the core competency questions you want this ontology to answer. And that ontology would be used to, for instance, link different data sources.
I remember one of the books I read when I started with Semantic Web is Programming the Semantic Web by Toby Segaran. It’s a very nice reference. And actually, they use Freebase and other database that were available at the time. And when you start working with them, they are all RDF, right? All triples. But then you realize that they don’t represent musicians the same way. So you would have to do these alignments to be able to work with these things, which can be, let’s say, overcome simply by everybody used the same ontology, so that makes it simpler. Or that’s where the work I’ve been doing in FAIR-IMPACT is interesting.
You can map and say that, I don’t know, a composer can be the same thing as a creator, right? And so you would tell the machines that they are synonyms or equivalent or roughly the same because you have, in some context, you can say it’s broader than another concept or narrower than another concept.
Andrew Dubber
Yeah, because different domains quite often mean different things by the same terms. So I guess that would get you into trouble when you start to overlap. So this idea of aligning ontologies, I think, is really interesting in that respect. The nuance thing, I was sitting here as you were talking and thinking about the fact that perhaps I’m biased because I’m interested in the humanity specifically, but perhaps mathematics and molecular dynamics are equally complex and nuanced and all the rest of it. So I guess some of the same things would apply across domains.
Yann Le Franc
Yeah, I agree. So I think that’s the main challenge of science. And one of the reasons is that concepts in science evolve as we discover new things, right? And some of the key theories that we had 20 years ago are void now because we discovered that they were totally wrong. Not because the scientist was wrong, but the technology or the methodology of measuring the thing was not up to the challenge to actually identify more granular relations. And so the initial theory was a nice shortcut. It would fit, but then not necessarily all the time.
And then you gain more understanding. Therefore, you have to evolve these ontologies over time as the knowledge evolves in science. So that’s a challenge for science in general. And humanity is science. So it’s a challenge.
Andrew Dubber
So just to sort of bring us full circle, we talk about the application of knowledge graphs to AIs to make AI more what? It’s not about being necessarily more accurate. What is the benefit to AI or use of AI from ontologies and knowledge graphs?
Yann Le Franc
Well, accurate is one thing, of course. We’ve made some simple tests where basically if you add the graph and an ontology, you get 95, 97 success rate in terms of answers, so you get better answers because they are contextualized, right? Now, these ontologies and the knowledge graphs are here also for, not just for AI, but for humans to search, right? Because then you get these huge trees of knowledge that you can go through and browse and navigate through, and that’s the main point.
Andrew Dubber
Right. To summarize, a triple is essentially the atomic unit of knowledge about how the world works.
Yann Le Franc
Yeah. Yeah.
Andrew Dubber
And the world is the thing that you need to define when you’re starting out.
Yann Le Franc
Yeah. And then you can add more constraints. You can, based on this, you can already go quite far. So the triple brings you quite far.
Andrew Dubber
Right. And so I’m curious, what was your journey towards this? Because you said you started learning about ontologies and knowledge graphs at a particular point. Where had you come from? What was the starting point for all this for you?
Yann Le Franc
So I’m a neuroscientist, by training. I worked in computational neuroscience for a while, so designing models of brain, neural networks. And well, that meant that I needed to use data from colleagues. And that’s how you realize that using data from someone else is extremely tricky because you don’t get all the information and some of the necessary information are missing or loss sometimes. So that’s what brings me to the data part because it’s essential. And then I started working during my second postdoc for the International Neuroinformatics Coordinating Facility based in Sweden.
And I was working on a group trying to actually better organize knowledge around the computational neuroscience model we’re doing. So small models of specific networks in the brain. So I’ve built a small model of the spinal cord for understanding how the spinal cord would deal with pain information coming from peripheral sensors and then send it to the brain. There are colleagues working on large-scale models of cortex network. And so at some point we needed, so there were databases, there are databases to share these models. But then you need to organize the information so that actually they are easy to, much more easily retrieved using keywords that you use on a daily basis, right? For coming from the neuroscience and neurophysiology fields.
So that was my first task, was to build an ontology for computational neuroscience. That’s how I put my finger in there and I realized the possibilities of that technology and how it would be instrumental for understanding the brain, right? Because studying the brain, you start from gene to behavior or even mental disease. And so you have to aggregate data from different scales, both at the spatial scales, but also at the time scale, right? And that, if you don’t have a proper information system with a good way to actually harmonize the way things are named, it’s going to be a daunting task.
Andrew Dubber
Interesting. You mentioned that ontologies come from the realm of philosophy. But I can see a connection, obviously, with neuroscience. And obviously there is maybe an oversimplification to map triples onto nodes and synapses within the brain. But that’s kind of how I visualize things. But in philosophy of mind, you get dualism and you get behaviorism and you get these kinds of approaches to thinking about where consciousness is situated and where knowledge and understanding is situated. Do we need to think about that when we’re thinking about ontologies and knowledge graphs? That’s just the realm of people sitting in high academic towers reading very thick books, or is that something else?
Yann Le Franc
It depends on what you want to do. If you start going for what we call formal ontology, where you get logic, and not just descriptive logic, but sometimes higher logics, then you need to start considering philosophical consideration. When you start defining the concepts, you have somehow to discuss what, I don’t know if it’s philosophy, but you need to actually get a consensus on the definition of a term, right? And so that relates to the philosophy part. I don’t think you need to have knowledge on the different theories of the brain where consciousness sits and so on, you can also have a more practical approach, which is like, okay, I need my ontology to do something, so I will design it to do the things I want it to do first, and then we’ll slowly increase the complexity and the adhesion with the others.
But I had an interesting discussion with an expert in formal ontology and a philosopher, and we came up to the conclusion that it would be impossible to actually get a unique ontology for anything because that would require that all the human beings would agree. And that’s not the case. It’s impossible because we all come with different perspectives. We come with different experience of the world, different viewpoints coming from your cultures or from how your parent raised you. So in the end we all have an ontology in the end, but it’s not the same. It cannot be, right? So one of the hardest parts when you build ontologies for domains, for instance, is to try to get all these different models reaching a consensus, and that takes time and effort.
But again, it depends what you want to do. If you want to do a beautiful ontology that will represent, for instance, neuroscience, well, good luck with that. It’s going to take a lot of time, and you will need a lot of academics, reading thick books, and so on. If you want to build an information system with this ontology, then you can take the more practical slash engineering approach where you try to reach consensus that makes meaningful to what you want to achieve. And that simplifies the work.
Andrew Dubber
When you deal with AI, particularly in the context of a European project, say, like LUMEN, you very quickly get into conversations about ethics. Does the same journey happen when you start talking about ontologies? Do you immediately end up in the realm of ethics, or is it not quite as demanding in that respect?
Yann Le Franc
No, it’s not quite demanding.
Andrew Dubber
Why is that?
Yann Le Franc
It’s just a model. The ethical problem with AI is things like the usage of AI, or the fact that AI is not complying with GDPR. So if you start putting your private information to an AI, it’s going to get fit from…
Andrew Dubber
Yeah, that feels more like compliance than ethics to me. The reason I raised this, because what you’re talking about and the way that you’ve described ontologies, it’s being kind of like the building blocks of how the construction of an understanding of the world is made, how knowledge is managed. And I think if it feels like there’s an ethical dimension to that in terms of whose understandings of the worlds get included in this. It seems like what you’re saying is those conversations don’t come up or.
Yann Le Franc
So one way of actually dealing with who is actually building this ontology is to document the ontology. And that’s where we start going towards the FAIR principles, right? Because the FAIR principles of findable, accessible, interoperable, and reusable rely heavily on what we call metadata, which is the data about the data. So if you take into account the fact that an ontology is a model, it’s a data, right? Then you need to document.
And so you need to actually provide information about who created this ontology, who created a particular concept, who actually provided the definition of the concept, and so on and so forth, which is what we call provenance, so that actually you know exactly how this ontology has been built and who created it so that you can make an enlightened choice if you want to use it or not, or if you agree or disagree with this model.
Andrew Dubber
Interesting. Yeah, we brought something into LUMEN which was JUST data practices, JUST being another acronym for judicious, unbiased, safe, and transparent. Now, no data set is ever going to be those things, but you use data research and data ownership and data management practices that prioritize those as objectives. But annotation is key to that and saying, when you use this data set, know this. And I think that that’s a really interesting way to do it. And you talk about documentation, I think of annotation as a way of providing that context. So it’s not even metadata, but it’s a narrative around the data set that gets used, whether it’s an AI training or whether it’s added to this. So I think that’s a really interesting aspect.
This is something that fascinates me. You’re a scientist, obviously a neuroscientist who has gone heavily into data and knowledge management and so on, working with EU projects and public funding and so on. And you’re an entrepreneur and someone who started their own company and managing a company. That doesn’t seem like it would be a very common thing. Is it something that you come across very often or are you sort of unique in that respect?
Yann Le Franc
I’m not unique. I have met a lot of other colleagues from research, coming from research, building their own company. For me, I think as soon as you start a PhD, you start being an entrepreneur in a way, right? Because you have a project, a goal you want to achieve, and you do everything you need to do in order to achieve that goal. And that actually gives you somehow an entrepreneur mindset. When you are in front of problems, you try to find ways around to resolve them.
And then as you grow as a researcher, you start building your own team, hiring people that will do the work for you. And you will be writing grants and so on and so forth. So it’s like being an entrepreneur in the industry. You would need to find customers and so on and so forth and build a team to do the work and so on.
So I’ve met a lot of entrepreneurs from different countries doing different things. And what is interesting is there are quite few people that started in science, actually, from academia.
Andrew Dubber
Sure. There does seem to be a difference between successfully leading a research team and successfully running a profitable business. There does seem to be a gap, not just in terms of the kinds of practices involved, but in terms of the kinds of psychologies of the people behind them. Obviously I think of you as a researcher and a scientist and so on, but I think of you as a businessman. I think of you as somebody who has run and set up a business that is successful and it’s achieving its objectives in that respect. And to me, that’s, unique is the wrong word, because obviously other people do it, but it strikes me as something you don’t often encounter in the same body.
There are people who are really, really good at research and running scientific teams and solving problems and doing research. And there are people who are really, really good at running businesses and making profits. The running of a business and the running of a research project are not the same thing. They’re quite different skills, I think, but you seem to embody them both. Which one takes priority for you? Which one you think of as your identity, I guess?
Yann Le Franc
Well, I have to say it’s changing, right? We change over time. When I started my business, I studied it like my own private lab. I was just doing research grants. I had very few business contracts and so on. And I realized that there was a gap in my training. I didn’t have the business mindset or let’s say the vernacular, the way to talk to a bank to present your company. And you would not go, if you ask for a money you would not go like you would go for a grant where you show how excellent you are in that particular expertise. There you need to show business indicators, my EBITDA is fantastic and so on and so forth.
So I had to learn this and I went back to school and I started an executive master in a business school in Paris. And that actually gave me a lot of, well, things I knew but never have been able to formalize, right? And management is a field of humanities and social science, right?
And so I’ve learned these things. And now, while I’m getting older, so of course I am still a scientist and I will always be a scientist. I really love writing grants and then having workshops with colleagues starting.
Andrew Dubber
You don’t hear that very often.
Yann Le Franc
Well, I think it’s, well, writing around is pain, but the building of a project is the fun, right?
And so basically, this is something that will always be part of me. But as I’m getting older now, I think I’m moving more towards the management part and the business part so that I can make sure that, well, my company grows and this is yet another challenge.
Growing a company, it’s not that easy. So for me, it’s a change in mindset. So I’m going more toward business and management and that’s reflected with the fact that I’m working for EUDAT as a head of secretariat where I do mostly management and running the organization, trying to optimize process and so on.
Andrew Dubber
Does that find its way over into sustainable business models for LUMEN or even for EOSC?
Yann Le Franc
Yes, because the main difference, I would say, between academics and business people is the business perspective, the business model. I’ve been working with colleagues in academia on business models. It’s a nightmare. They don’t know what a business model is. And it’s because in academia, you don’t need to make profit and people do not like to talk about money. And so basically when you start working with them on these kind of issues, it’s becoming really cumbersome.
And I think the experience I gained by building my own business model and so on is what I can bring to LUMEN. It’s a business perspective. And the good thing is in LUMEN, I’m not the only SME involved. So basically this is something that we can do jointly with other people that have the same perspective so that we can come up with a proper business model, right?
And so, and the same goes for EOSC. This is something that we are pushing in the discussions because this is major. For instance, one of the challenges I view that is we are an organization that provides services across Europe. These services are based on resources that are national resources. And so how do I offer storage space in Germany for, I don’t know, a Spanish researcher? There’s no business model, there’s no contractual framework, and there’s no necessary legal agreement.
So, and this is a key, I think, for the success of EOSC is to start discussing transnational access to resources, because that is going to be the main point. And this is also an issue that colleagues from thematic research infrastructures in the different fields are facing, because their resources are scattered across Europe. But then it’s really complicated to find, can I use your compute cycle here? Well, but you’re not a national researcher, so no, it’s for national researchers. So we are still in the progress of building this Europe of science, right? Because these frameworks do not really exist yet. So there are things that came up, but they are not necessarily finalized and optimal solution. So that’s something that is being discussed, and not just by EUDAT, that at some point we’ll have to discuss about these issues of sustainability of the nodes and providing access to researchers from other countries, and so on and so forth.
Andrew Dubber
That’s putting the European in European Open Science Cloud, I guess. So LUMEN is an EOSC project. We are coming up to one year into it. We’ve got two years to go. 20-something partners, 7 million euros, give or take. What’s been achieved in that first year and what’s the two-thirds left to go, in your opinion?
Yann Le Franc
I think a lot has been achieved already, at least on the perspective of my work package. The task of the work package is really challenging because the idea is to take tools that are used by ontologists and taxonomists that are distributed across the web, built by different persons, and try to orchestrate them into a coherent workflow. And so we have been extensively working on analyzing the tools and laying out the foundation to start the implementation.
And so for us, it’s a big achievement because that’s actually a lot of work. Now the implementation is going to be the next challenge. And for LUMEN as a whole, I think we are making a lot of progress on the different areas. And LUMEN in particular is, for instance, focusing on building an architecture based on the data mesh architecture, which implies starting having discussions about data contracts or metadata to describe the data and so on and harmonizing that across different communities. And I think there, there’s also a lot of progress. Of course, there is also the innovation part in LUMEN that is making progress. We’re here for the first IPL, right? So I think globally we are making a lot of progress on the project for the rest of the two next years.
Andrew Dubber
Fantastic. I’ve got one final question, which probably lies, strictly speaking, outside the realm of this podcast. But I’m curious, because you’re a neuroscientist and you’re working with data, a lot of the conversations I hear about how the human mind works these days use analogies or even metaphors from computer science. The brains are processors. They have this many, it can process this much information. The language that we use to talk about the brain is basically more complicated computers. And I always feel like that almost devalues the human consciousness or human brain or the workings thereof. Is it something that you do, that you think of neuroscience as a branch of computer science, or is there a different language that we could be using to talk about how brains work?
Yann Le Franc
I’m going to ask you a question. Who came first, the brain or the computer?
Andrew Dubber
Oh, that’s quite an easy one to answer.
Yann Le Franc
Who looks like more than, looks like the other.
Andrew Dubber
Right. So we have made computers in our own image, is what you’re saying.
Yann Le Franc
Right. So in my training, I went to do a master in cognitive science, and I had this teacher that started the introductory lecture and said, in the 50s, physicists believed that they solved the matter, the universe and everything. And they said, oh, let’s have a crack at the brain. And from that work, from that idea came out computers, AI. We’re talking about AI. That’s the foundation that’s coming from these guys. Think about Turing, John von Neumann, and all these guys have been the, let’s say, the foundational people. And they started building a computer a bit like what they know about the brain in the end.
So I would turn this way around. Makes sense to talk about processing in the brain and things like that, because computers are somehow a simplification of the brain. Like the AI we’re having now, everybody is into AI. I wouldn’t call that AI. LLMs, that’s my own opinion, have nothing intelligence. They are statistical machines that actually can give you a lot of very interesting behavior, but that does make them really smart or intelligent.
Andrew Dubber
Which brings us back around to philosophy in terms of questions of what is intelligence, what is consciousness.
Yann Le Franc
Exactly. And that’s still an unresolved issue, right? We don’t know. So there are a lot of theories about how to define intelligence. We realize that basically in the end, there are many different types of intelligence in humans. There’s not just the intelligence. It’s not just logics and calculus. It’s also there is intelligence in art. There is intelligence in building things, right? So there are different types of intelligence and still we don’t know how to really define that, because it’s a complex thing.
Andrew Dubber
I was going to end this conversation by asking you if you would ever go back to neuroscience, but it sounds like you never left.
Yann Le Franc
I never left. Well, I don’t do neuroscience anymore, so from time to time I miss it, especially when I bumped into colleagues where we start talking about what they’re doing and I’m like, I miss that. I would love to work on this and this aspect. But in the end, I’m at peace with that because I found a purpose. Data is not just for neuroscience, it’s for science. And science is what is going to save the world with all the big issues. So I’m glad that I work with people in humanities and social science.
I’m discovering things on sociology, archaeology. I work with people on epigraphy. This is amazing. I’m discovering other realms of science, and I’m fascinated by it. I have been working with people doing the models for weather forecast. And they started explaining me the granularity of their model. I was like, wow, thumbstruck. I’m like a kid, discovering new things all the time. And I’m still in contact with neuroscience because I married a neuroscientist. So I never left neuroscience also for that reason.
And I’m still in contact with some colleagues that I meet. For instance, there is a good platform for neuroscientists, EBRAINS, that has been built by some of my former colleagues, which is a very good data management platform for neuroscience data. So I’m still around, right? But in the end, although I studied in neuroscience, I realized as I’m getting older that I’m more a scientist as a whole because I’m curious about everything. And every time I’m discovering or meeting a new scientist, talking about new theory, I’m just like a kid, with, seeing the world for the first time. And I think that gives meaning in my life.
Andrew Dubber
Fantastic. I think we’ve got our pull quote for this episode, science will save the world. Yann Le Franc, it’s been an absolute pleasure. Thanks so much for joining us for the podcast.
Yann Le Franc
Thanks for having me for this very lovely time and discussion. Thank you.
Andrew Dubber
That’s Yann Le Franc, CEO of eScience Data Factory, Head of Secretariat at EUDAT, and Work Package 2 leader on the LUMEN Project. And that’s Inside EOSC, a podcast from the LUMEN Project, which receives funding from the European Union’s Horizon Europe Research and Innovation Programme. I’m Andrew Dubber, and I’m back next month with another EOSC Insider. Now, you can subscribe to Inside EOSC wherever you listen to podcasts. Five-star reviews and shares with colleagues are always very much appreciated. Thank you so much for listening and talk to you soon. Cheers.