AI is not particularly capable today of choosing between different visions. On the other hand, it is capable of suggesting different visions. But afterward, we keep our own critical judgment. It is an additional tool, an additional point of view, it is also something that opens up a lot because when you ask it for possibilities, for next steps, for ideas, it has plenty. However, telling you which one is the best, that is something it generally cannot do. But we do know that what it suggests is sourced, based on facts, and based on many references. There are two ways to use AI. There is the lazy way, which is "do my homework for me," and then there is the active way of using AI as a tool for learning. Personally, I am obviously in the camp of using it as much as possible as a tool that can teach me many things. And so, that is positive in any case because it is a source of additional knowledge that should fuel our reflections.
Would you accept being governed by an AI? Imagine an AI becoming the president's most listened-to advisor. In "Robot 1st," Fred Mazella imagines an AI that enters the heart of power to inform public decisions based on facts, data, and what works elsewhere. So, but from the moment it advises better, faster, and with more information, a question becomes hard to avoid: Where does advice end? Where does power begin? How far, sorry, are we ready to let AI into our lives? Welcome to "AI Unfiltered" with Fred Mazella. I remind you that you are the founder of BlaBlaCar and DIFT. You also participated in the development of France Digitale, you launched the Prix des Citations, and now, you are an author. And yes, with your novel, "Robot 1st." I am the first to be surprised.
I am delighted to have you with us. I am delighted to see Isis again, who currently handles AI adoption at BPI France. We are going to, well, talk about this work. Uh, the first question I asked myself is how you think we humans would accept power managed by an AI? But
I think it is a form of rationality. It's from the moment we have something in front of us that is, uh, quite strong, objective, rational, sourced, has arguments, is capable of a certain logic, and manages to convince us, well, we are capable of getting carried away. Yeah. Hm.
There would still be implications for our reality. Oh yes.
Which in fact creates a certain amount of reluctance, as is the case today.
Yes. Uh, afterwards, it all depends on the measures taken, but since it is, uh, an artificial intelligence, it is therefore necessarily not human, and it will imagine things that we aren't capable of imagining as well. But I find that the interesting thing—and that's why, well, I like to use this pretext of AI—is to look at our society differently. That is to say, if you want, for the novel, I could have taken an extraterrestrial and put it at the head of the country. That would have led to the same type, could have told the same story. Yeah. Because, in fact, it's a way of looking at society that is different, saying: "Wait, if the entity that somehow exercised power wasn't the same and wasn't human, what would it think in terms of organizing common goods?" Perhaps it would think much more about the common good than a human who carries an ego that is sometimes... yeah.
And who has their own biases and who necessarily has access to much less information than an artificial intelligence. In "Robot Premier," it's Cobot.
Yeah, his name is Cobot. It's the, it's the collaborative robot.
Exactly, that's it; it starts as an assistant to power. So, it compares, it analyzes, it provides info to the president, and then, gradually, the line between the advisor and the decision-maker narrows. Even, well, without completely spoiling the novel, what happens is that at one point, since it's the head of state, we organize a... First-round presidential debate that is human versus Cobot, which...
Sounds kind of appealing. No, it's not human versus Cobot; it's just that there are all the humans, all the candidates, and then Cobot just does fact-checking. So it's in the middle, and when there's something that's false, it says so. It says, "Oh no, that's not true." So, if you want, live, you can no longer have candidates saying false things that can't be immediately corrected. And so that
provided it has the right sources to check.
Yes, obviously, it all depends on what we put into it, but indeed, the training matters enormously. The dataset it's given is very important. But it also allows for a different perspective on the fact that you can obviously have opinions, but they still need to be based on facts.
Yeah. Uh, and so it gives, uh, it gives a different perception. What happens is that the candidate who initially adopted, uh, artificial intelligence did so because they were struggling in the polls. They thought, "Hey, this will give me a little bit of visibility." And then, uh, little by little, as all the other candidates engage in their favorite game of tearing each other down, uh, and we witness, well, quite simply almost a collective suicide of the candidates as they keep attacking each other and blaming one another for things. Uh, well, the last man standing, ultimately, that's the candidate with their AI, because nobody could really attack them since it was true and it was factual. And so, by default, they end up being the candidate who makes it to the second round, and then they manage to get from the second round to the presidency.
It's almost angelic, isn't it, to think that facts are enough today.
Maybe. Yeah. Yeah. No, sure, of course. No, but I think there's a part of me—I'm an entrepreneur, I'm an optimist. Uh, I...
there's that side to it. And so then, the way the AI actually comes to power isn't so much by a slide, but because it continues to fact-check, and it also fact-checks, uh, the president. So it contradicts the president at times, when they say something, and then right behind them, the artificial intelligence says, "Oh no, that's not it." Uh, it corrects them, and so the president eventually gets fed up, and so one day they say, "I've had enough of this machine, of this know-it-all, so we're getting rid of it." So he holds a referendum, asking everyone: are you in favor of getting rid of this robot, since, after all, it was me who was elected? And then, well, the population basically says no. They vote no by 52.4% in his referendum. So, it's the same scores, if you will, as General de Gaulle in '69, with a "no" at 52.4%. Which means I have him leave, uh
the president
the president says, "I've had enough, I'm off to Colombey-les-Deux-Églises," and, uh,
and we don't know if he'll come back yet.
We don't know yet if he'll return. We're at chapter 25, we don't know.
Yeah. So, we're going to tell the process. Just a word from me, and I see they also say you want to go there. But "Robot I" sounds a bit like an emperor, or even a dictator.
Yes, but that's because it's the name the press gives him. Meaning, he calls himself Kobo. He never called himself "Robot I," if you will, but at one point, when he actually becomes president—well, when the artificial intelligence becomes president—uh, the press, if you will, talks about "robocracy," because, you see, it's a new... it's... it's no longer a democracy, it's... well, you see, it's something else entirely. It's managed by a robot. So it's called "robocracy," and then, bit by bit, they end up calling him "Robot I" because they don't really know what to call him, and so it's the name the press gave him. It's not his own name. So much for the backstory.
We started to touch on this a bit at the beginning of the show, but how do we continue to form our own opinions? How do we keep our free will when facing a machine that seems to have all the answers, knows everything, and can do everything?
Well, no, because it is an excellent tool for providing us with all the information we need to decide on future matters. We have the right to keep our own opinions and, obviously, our own visions. Today, AI is not particularly equipped to make choices between different visions. However, it is capable of suggesting different visions. But after that, we keep our own critical thinking. That's why it's very important to continue as much as possible to develop critical thinking in our young people, in everyone, in the entire population. It's an additional tool, an additional point of view, and it's also something that opens many doors because when you ask it for possibilities, for follow-ups, or for ideas, it has plenty. However, telling you which one is the best, it usually doesn't know how to do that very well. It also doesn't really dare to say, "I prefer this option or that option." But on the other hand, we know that what it suggests is sourced, based on facts, and based on plenty of references.
But, but I see it around me, we often ask the question, "What do you think?" Whether it's Claude or ChatGPT, we explain a situation to it, "What do you think?" It will give us an answer; how do I actually manage to keep my...
Wait, you ask it "What do you think?" You don't ask it "What do I think?"
What does... what does it think?
Of course. No, but you ask the AI, "What do you think?" Wait, there are people who ask it how they are doing, you know. And there...
there you go. But you can say, "What do you think?" And even when you ask someone "What do you think?", it doesn't mean you're going to think the same thing. You have the right to keep your own, your own idea. And then the AI might think something, and you have the right to think something else. However, it can explain why it thinks that way; it can provide arguments. Now, it turns out that quite often, its arguments are fairly difficult to challenge. So...
Fred, with Isis, we talked a lot about AI and therefore about processes, because AI is changing processes. I'd like us to talk about the creative process. Now, we have a Goncourt contender who got bashed under the pretext that he might have used AI—you never know if one can be certain. For you, it's assumed.
Oh, for me, for sure. Yeah. Uh, well, in his case, he topped the sales charts and I think he was a Renaudot contender. So...
but I'd really like to talk about how you systemized the creation of this work and to what extent—I want to understand what you put into it and also if you were ever surprised by the options for your characters or the plot twists that were going to happen in the novel.
Yeah. So, it creates some pretty amazing things. Well, first of all, in terms of quantity of prompts, I must have sent about three to five times the volume of prompts compared
to the book. So, if you like, it's still a, um, um, well, continuous work, where you... Sometimes, for certain passages, you, you do 25 prompts on them because, actually, you're not satisfied and you look for synonyms, you look... You see, that's a tool that allows you to go look for information. Well, basically, you see, if I'm talking about synonyms, either I use AI to find the best word for me in a certain place by saying, this is what I want to express, it's more like this, can you find me the word that fits well. Or I take my synonym dictionary and then I, you see, but it amounts to the same thing, actually.
So, in terms of percentage, could you tell me what was produced by the AI and what was produced by you?
It's impossible, it's truly written with it. It's not a question of percentage, it's a question of writing. I write. So, generally, I can, um, write down my intention. So, for example, a paragraph, I don't know, about that, I'll, I'll type it, I'll write it, I don't know, over 20 lines, saying, here is the intention for this little passage, which should be about 300 to 400 words. Um, and I send it to it, and I say, how would you write it? Boom, it can suggest something in how it writes. Then, I take it back, I take back the text, I rework it, I re-inject it saying: "Here, I've modified this, that, that." And then there, it can tell me: "Careful, that's something you've already talked about in such and such a place." So it also makes connections, compared to, you see, because the book, well, in volume, it's about... well, the 40 chapters, it will be almost 500 pages. So you see, when you want to cover all the subjects and not necessarily repeat yourself, it's good that it also points out passages that are similar.
When, when you talk with authors, you realize that, well, I'll summarize a bit, but globally you have two creative approaches. You have those who let themselves be carried by their own writing without knowing where they're going, and you have those who first build a narrative arc, all the character details, and who follow it and slightly adapt it within the prompts you've imagined, because on top of that, you release chapters regularly, so we don't have the ending yet. I don't know if there will be an ending, actually. But it's open-ended.
But did you prompt by first projecting the narrative arc, or are you just letting yourself go?
No, I wrote the screenplay first. So I first wrote the, well, a plot outline, right. That is still, uh... Yes. Meaning it can still be adapted. I can graft things onto it. Uh, actually, why is it so important to write the screenplay first before starting the actual writing? It's because when you're going to write, you're going to plant, like Tom Thumb, lots of seeds to reuse later. You see, you plant seeds in chapters 1, 2, 3, 4, 5, 6, 12, uh, to be able to use them in chapters 18, 22. You see, I have 40 chapters, you have to build the...
Yeah. So you plant lots of things. Uh, so for me, it's very important to already have the screenplay in mind. I know what's going to happen at the end, and so as I tell the story, I plant things that might seem trivial, and some will lead to surprising things later on.
You're promising us a final climax.
Well, yeah, it's a sort of, it makes you think. It's more of a takeoff.
Are you planning to make available or show us a little bit of the prompts you used?
I could, but I don't know, I mean, it's heavy, it's intense, it's a lot, you know. I mean, it's, it's personal.
Also, yeah, no, the thing is, what's difficult is that in the prompts, how should I say, I'd have to extract them because, you know, these are conversations you have with the AI. So sometimes there are things that concern "Robot Premier" and then there are lots of things that don't concern "Robot Premier" at all. So I have to extract the prompts that concern the novel. But
what do you prompt with for "Robot Premier"?
uh, which writes quite well, which has good suggestions.
However, sometimes I, well, yeah, very often it's me who has the idea, obviously, and very often even the ideas it suggests. It's funny because it suggests things, but I'll say to myself, "Yeah, but not quite," and I'll change it, I'll generally improve it, you know, when there's an idea worth using. After that, 9 times out of 10, it's stuff it suggests that I'll never use. So, it opens up possibilities, you know. And in the same way, when I obviously want to push my idea, I push it and I also challenge it. I ask it, "Does this hold up?" "What do you think?" And
and if we move away from Robot Premier, what has been your boldest move involving AI?
Ah, I don't know. First off, you have to use it. Well, beyond imagining that an AI would be President of the Republic, which you could say is a bit bold, you know. I mean, you see, it wasn't exactly obvious. There's the aspect that I also developed a tool called Fred 24, it's my digital twin that answers like me because I trained it with 20 years of content. So everything I've said and written was used to train it, and...
It's a good thing you mean what you say.
That could be problematic.
Ah, if I don't say what I think or if I think what I don't say...
