Hi all, thanks Harry and Mindstorm for inviting me for this talk, it's a pleasure to be here speaking at Epitech and I love to talk about this subject because it's something I love, it's AI.

Work and Creativity in the Age of AI

So what I'm going to talk about, when the old constraints disappear, really. This is something that I love because I have been, I have studied like, I've been a student some, I would say like 18 years ago, I finished my studies. So I think there were lots of constraints during that time.

And now today what I'm going to talk about is like everything has changed. Thanks or because of, I don't know what I would like to say, but let's see.

So what I'm going to talk about is work and creativity in the age of AI, what has become possible and what still matters.

Research at the Intersection of Data, Space, and Time

Just to give you a context, I am an enseigneur - chercheur or within in English I would call it associate professor at CP Lyon and I'm also a chercheur -associé researcher at the Lewis laboratory, the CNRS laboratory.

So I have these two things but I am also going to use the third hat here, sorry, the third hat here, the creator hat, so something that I'm I'm going to talk about some of my personal experiments with the AI, with the generative AI.

So just to give a quick idea about what I'm doing, as you have seen that I'm a researcher. And all my research work revolves around data. So data is a key part.

And then something comes up. It's called space and time. So you have got people coming up with ideas about space. It could be a three -dimensional space, two -dimensional space, et cetera.

And then you have the idea of time. Time, it's very interesting because time brings about lots of interesting information. So one thing that I work with, one thing that I did in my PhD and also in my postdoc is about studying about changes. What does it mean to have changes?

So my talk or my personal research work revolves around urban data science, cultural heritage. change, like you have got storytelling approaches, you have got memoir, and I love the French word for memory, memoir, it's very nice because it talks about something personal. So this is also very important, and this is also one of the things that I do in my research work. I'll talk about this in my upcoming slides.

And finally, I do not know if you know that you can also see three -dimensional views of Lyon as a map, something that you may have not seen. But this is interesting because you could see complex relationships with them. So data space and time together brings interesting knowledge, and that's what we want to explore.

Current Research Projects

So what I do in my research work, I have worked in numerous research projects, but I want to talk about three of my current research projects where I'm involved.

The first one is the French agency, French National Research Agency, funded by them, the ANR, Agence Nationale de la Recherche. And it's a gap project and in this project what we are trying to do is to discover iconographic heritage. What I mean by iconographic heritage is like something like you have your old photographs, old really old photo card postal cards which you may want to visualize you want to see and

understand how things have changed and you also want to link with newer technologies like you may think about social media like instagram etc where you have lots of historical buildings and you want to link how things have changed between uh then i mean i would say it's like between like 15th 16th century or maybe 19th century, etc. And then today. And this is one of my projects and I really love in this because it's about cultural heritage.

The second project is with the Metropole of Lyon. It was funded by La Banque des Territoires and it's about studying the climate result. And this is very interesting because if you have seen, it's a French word for AI, IARB. It's what I would say It's like AI and trees, and here we actually really use AI to study about plantability aspects of Leung, and also study aspects related to heat vulnerability. So this is my second topic.

And the third topic, as everyone is like, right now talking about ChatGPT or Gemini or whatever, I don't know, Claude. And this project, and the third project, Hazorus Automation Creation, we are actually using LLMs to explore how we can link historical data or historical concepts and use LLMs to enrich the relationship between these concepts.

It's very evident for human being to say that table is a material, but for machines to say that automatically is very difficult. So there comes the LLM. So this is the third project.

I will go one by one through, I'm not going to detail, but I'll try my best to talk about these aspects. So, sorry.

From Symbolic AI to Neuro-Symbolic Systems

So where did I start? So I am coming mostly from the symbolic AI background. I did my PhD in doing rules, using rules for representing knowledge,

especially about web services. And I use declarative languages to explore and define this stuff. And this was my first thing.

And now then things that was back in 2011, 2014 and things have changed since then you have this neural network that have come up and you can easily detect stuff like I would say if you want to use some computer vision models you could test them and you could say oh please find me the important things that are present in this photograph. graph. This is very, very common in neural networks.

The second thing which has really changed is the release of LLMs. The attention is all you need, the Google paper which was released some time back and which has revolutionized the textual model world of AI and it's very

interesting because you can ask complex questions and thanks to these statistical associations associations that are present in this LLM, you could find out interesting information.

And now I come to the third part, the FSP, the Foundation des Sciences du Patrimoine work where we actually use neural work, the neural aspects of the work, and the symbolic.

So I always see on the internet, people say, oh, the symbolic AI is gone, nobody's going to use it. And then you come here and you say, no, it's not gone, please. you can use a lot of stuff with a neuro -symbolic AI.

So the symbolic part is not gone, and this is what we are exploring in the FSB TAC project.

Choosing AI Models Responsibly

So one other thing that I want to talk about because this is AI talk and MindStone is about, it's also talking about teaching people, non -technical people about AI.

I really want to talk about AI. AI is a really vast subject. you have different types of concept like symbolic AI or you have the classical before 1990s you have some AI stuff from before this period and so you have to

really ask the question whether you really want to use energy consuming AI models for doing stuff and this is what we are doing in the the ER project for funded funded by the Bank de Territoire we are actually looking at stuff like like what type of data or what type of model or what type of AI should be used? Should we really use time -consuming, energy -consuming models? Or can we make some sort of compromise?

And so the first thing that we do in the project is question the need, the very need that happens, and see if we can reuse what is there. You don't want to create a new model every time you do some stuff.

And so we have got a very interesting model from IGN, the National Geographic Institute of France, and they have created lots of AI models. So what we are actually doing in this project is to look how much of this project works. These models can be reused.

Do we really need to go to the third step, fine -tune the models, or we are okay with the results that we get from the existing models? So this is something that we are asking in terms of our project related to the ER project.

Personal Experiments with AI and Language

Now, let me come to my personal aspects. I also wanted your, I said, work and creativity in the age of AI. So I want to talk about some personal experiments.

Rethinking Programming Languages

So back in 2017 or 2018, I gave a talk in Capital du Libre. It's an open source conference in Toulouse. so I gave a talk there and I said let's talk about multilingual command line why

did I give this talk I just started teaching in 2017 it was bit difficult I was like oh people have difficulty in understanding what's the command CD means what's the command LS means so I was thinking why students have this difficulty in retaining these commands so I quickly understood that

The idea, problem here is that you have these words like directory, change directory, etc. But these are not French words. So how would students keep these mnemonics in their heads? I had this question in my mind, and I came up with some proposition during that period.

I said, let's change differently. Let's make repertoire change, create a fichier, etc., etc. I proposed some ideas. I was not very happy about it.

And then, and that's the part that's interesting, I came to this year. this year lots of agentic AI came up, cloud came up, open AI codecs came up, so many agents have come up, AI agents have come up and said okay let's try my experiment from the past and I wanted to say I want to create a new programming language where people can write in their own languages.

So these are the two examples that I wanted to is very simple I'm not going to talk about because because there are complex examples in the code that if you go to the link, you can find them.

So the idea here is I wanted to explore that if AI is there, can we create a new programming language where people can write in their own languages? So this is one thing. So multilingual is all about that.

Open Data, Open Source, and Multilingual Knowledge

And this is not something new. There are other people who are working about it. I would like to give some experiments

that we are doing right now on open data projects, please do not forget open data and open source.

If you are using AI, it's thanks to the people, the community behind this, so please also consider whenever you use ChatGPT or Gemini, because these are done by community and they feed the data.

Okay, I won't start with the stuff like that. Let's talk about the example.

So everybody knows Wikipedia, so we have seen seen lots of interesting advantages of Wikipedia. One example is the LLM.

The second thing that we forget is that some of the things that we normally see on conversational agents is Wikidata where the information is represented in three columns, like for example, seed developer, Dennis Ritchie, seed influence, C++. So these are boring stuff.

And we, what we want to do is like, okay, there are communities, there are languages where people do not have enough contributors so what can we do what can we how can we create stuff where knowledge can be represented in a manner that could be used by anyone in anywhere in the world in their own language so this is something that we are doing so

on the right left hand side you have got some shared facts and on the right hand side the goal is to generate sentences based on these facts so really this is This is something that is important because knowledge is useless if you don't give this information in the local language. This is something that I'm also actively working on because I'm an active contributor on Wikidata.

AI in Education and Assessment

One thing I also want to talk about, I'm a teacher so I also have to talk about AI and the impact of AI in my classrooms. I think the professors or the teachers from EpiTech may also know the problem.

Of course, AI can produce very good code, but do we know the fundamentals? And this is really very, very complicated for us nowadays because we do not know how to evaluate the students. We have got no clues.

We are testing many different approaches. Should we ban AI? No, that's not a good idea because students should learn how to use AI because that's in future. That's new stuff that's coming up.

So what should be done? We should ensure that the fundamentals are done. So I have done many experiments, like for example, questions and quizzes with the help of AI, developing games with the help of AI, trying to use new ways to evaluate,

maybe going back to the paper stuff, old style, I do not know. Let's see, and we will see what the results can come up.

Creative Coding Beyond AI Slop

So now I can come back to the example that I wanted to talk about, about multilingual.

And I said, everyone is complaining about AI slope, and I really do not like all the AI slope that's on internet because it's not very personal.

So my approach was completely different, I wanted to see if AI could be used to do some creative stuff.

So I wrote all my programs in French language, so you can go and check if I'm telling a lie, and you can check the code that I have written, it's written in French language, and what

I did was to see if I can create patterns, these are fractal patterns, L -systems, those that are those who know about the fractal systems and I said okay let's explore the classical rules that existed long time before I coming back to my initial slide and resurrect them in this

neuro symbolic world and here what you see is is a mix of code that have been generated with the rule systems which I talked initially and the neural neural aspects because the code is generated generated by the AI agents, and I just gave them them.

So here what I'm saying is, you define the rules, you give some probability aspect, and you get some surprises like this generated by the code.

So please do not complain about AI slope. You can do some very, very, very interesting stuff with generative AI.

What Still Matters: Human Creativity and Diversity

So I'm coming to my last slide. I hope I didn't bore you. I gave you some good points about AI. I motivated you to use AI in different contexts,

but I really want to end up with this particular photograph that I took in Lyon. It's a photograph I took on February 13, 2024. I was walking on the quai, Quai du Rhône,

and somehow I was surprised by the color of the sky. And I was just like, wow, it's really good. I need to take the picture of this beautiful sky with the colors.

and what I did I think I took and I'm like it's unbelievable and then can came AI and people were like generating stuff like this but then I my question as a photographer as well is okay you can generate AI picture perfect photographs but did your eyes really see them really see that object because this if you take

a picture so if you generate an AI picture somehow you miss this particular point, this beauty of nature, which was taken, which was captured by me or a person, I decided the particular frame, etc. This will be missing, and this is something that I want everyone to think about, that

AI is not really perfect, there are problems, there are biases, I've already talked about about it in the past, my previous talks in other contexts. So AI could be biased because you could have data that is missing certain languages,

that is missing certain information, so that creates a lot of problem. So these are things that you should always keep in mind when you think about AI.

It's interesting to create lots of stuff, but what I really want to end up is to say that the creative diversity is important. We don't have to generate the same stuff all the time.

There are different ways to create new stuff with AI, and this is what I want you to do it. Please go ahead and use AI, but don't forget that humans are diverse

and our creative brains are also diverse and you can create really interesting stuff. Thank you very much.