WASHINGTON AI NETWORK PODCAST Host Tammy Haddad EPISODE 25 Special Guest: ANNE BOUVEROT France’s Special Envoy for AI July 19, 2024

Tammy Haddad: [00:00:00] Welcome to the Washington AI Network podcast. I'm Tammy Haddad, the founder of the Washington AI Network, where we bring together official Washington, DC insiders, and AI experts who are challenging, debating, and just trying to figure out the rules of the road for artificial intelligence. It hasn't even been a year since the first AI Safety Summit in London, and as France prepares to host its own AI Safety Summit in February, French President Macron appointed Anne Bouverot as Special Envoy to lead what they're calling an AI Action Summit in Paris. Tammy Haddad: Anne is an engineer, a PhD in AI, a business leader and educator. She's in Washington this week, and we're so pleased she could join us at the Washington AI Network. Anne Bouverot, thank you so much for joining us. Anne Bouverot: Thank you. Thank you for having me, Tammy. Tammy Haddad: You have done so many things. I don't know where to start. Why don't you tell us now what you're working on in AI? Anne Bouverot: Yes, well, I'm working on fascinating things first, because AI, there's really [00:01:00] so much happening and so many things. The most recent thing I did is I co-chaired a report for the French government on recommendations for what to do, What France should do on AI and we delivered that to President Macron in March, back in March. And I'm very pleased to say that some of the recommendations are already being implemented and hopefully more to come. But as a result of that, and because the UK had organized this Bletchley Park AI safety summit back in November Of 20 of last year of 2023, we volunteered. We France volunteered to host the next summit. And yes, we were calling it the AI action summit, and it will take place in February[a] of next year. Tammy Haddad: And President Macron has been focused on AI for a long time. Anne Bouverot: Absolutely. This is actually the report we just delivered, is the second such report. There was one very early on in 2018, chaired by Cedric Villani, one of our field medals academics, very well known, [00:02:00] and therefore we were tasked with updating that years after that and after the, uh, the rise of generative AI and, and all the changes. Tammy Haddad: Well, we in the U. S. are very familiar with Mistral. But can you bring us up to date on the depth and breadth of the AI work in France? Anne Bouverot: So actually, we're very lucky, I think, to have a thriving ecosystem in France. And in AI, it really starts with research. So, we have a great community of researchers. A lot of them actually go and work in the U. Anne Bouverot: S., but some of them come back. So, Yann LeCun, for example, at Meta is, is French, and there's many, uh, many others. Joëlle Barral at Google is French and she's a French woman. Arthur Mensch from Mistral and a few others worked in U. S. companies and then decided to come back to France and to create Mistral as an open source model. And there's many other. companies. There's the Franco German Helsing. There's Hugging Face was created by Clement Delangue, a [00:03:00] French guy. So yeah, there's a really thriving ecosystem of academia and startups and innovators in AI. Tammy Haddad: So I have to take you back when ChatGPT came out. Yes. Were you surprised? Anne Bouverot: I think everybody was surprised, including OpenAI. Anne Bouverot: I mean, yeah, yeah, this, of course, people knew that the work around transformers and language models that was progressing, but I don't think anybody really expected, uh, the level of interest of take up. I mean, even OpenAI has spoken publicly about this. Mira Morati has said like on day two, she called Microsoft and said, I need more compute because there's so many people using it. Anne Bouverot: So, yeah, I think it was, it was a bit of a surprise for many people. Tammy Haddad: She also said that most of the people in the company didn't know anything about it. Either. So, can we talk about the difference between the U. S. and France? We'll get to the EU act [00:04:00] on, you know, in general, but let's start with France and the U.S. in terms of technology companies and what you're trying to accomplish. Anne Bouverot: Well, clearly, I mean, the U. S. has a lot of great, large, medium, and small tech companies and a very vibrant ecosystem of investors, VCs, growth capital, and thriving academia. We really believe that such an ecosystem is key to the development of any technology. And so some of the things that we've been trying to do in France since a number of years is to have these types of communities. The Villani report in 2018 recommended to put some strong focus and funding into AI research and to create AI clusters, which has been the case and we've recommended to continue and to expand in this direction. And there's been, since a number of years, a push on creating the ecosystem for startups, Station F, and other incubators and funding and things like Choose France. Are you familiar with Choose France? This is a once a year event in Versailles, we're lucky to have Versailles, where companies from around the world are invited in relation with investments or projects that they develop in France. And they can meet with the startup ecosystem and that has been really, really beneficial for innovation. Tammy Haddad: Did you worry when Meta announced? I think a week ago and Apple earlier that they're not going to roll out some of their AI products in the EU because of the new legislation. Does that worry you? Anne Bouverot: I think it's actually not really new, Google has been doing that for Gemini as well. Anne Bouverot: It's not that they're not rolling out, it's they're saying it will take us a bit more time to make sure we take into account the regulation. And really the idea of having regulation at EU wide level is to actually make it simpler so that they don't have to adapt to [00:06:00] regulation in Germany, in France, in Spain, et cetera, et cetera. Tammy Haddad: Yeah, but don't you feel like you're sometimes dragging the rest of the EU along because you guys are so aggressive about innovation? Anne Bouverot: We're trying to, yes, inspire the rest of the EU. Tammy Haddad: I call it dragging. You call it inspiring. Anne Bouverot: You are a great diplomat. But we see appetite for that. In having this summit in February in France. I was in Germany last week, I'm meeting with other countries. They're very, very willing to come to have their startups, their ecosystem. They're asking, are you going to do something like Choose France for others? I mean, yeah, there is appetite. Okay. Tammy Haddad: All right. like I said, I call it dragging. Okay. So let's go back. So the AI Soeul summit was the last gathering. What, on And so we, we have in November, we had London, then we had Seoul, and now you have a few more months. [00:07:00] What's your top goal? Anne Bouverot: So the top goal, I guess, is we really want to focus on deliverables. There's already been two summits. They've been doing a great job at convening people at agreeing voluntary commitments. Anne Bouverot: I think at some point you need to stop just doing incremental voluntary commitments and move to concrete actions. And we really want to make sure this is not only a conversation for the happy few, but a conversation for all of us. all the people who want to participate in AI. So we have a few tracks and a very important one is public interest AI. I think we have to be able to tell all the people in all the countries in Africa, in Latin America, in Southeast Asia, in some of the countries in Europe that don't have that level of innovation ecosystems that they can participate and that we will help to make it possible by helping access to compute, access to data, training, [00:08:00] and not theoretical training, but training into how does that work and what do you want to do in your country. I think those topics, whenever I go around the world or meet with representatives of civil society, of companies, of countries, that's a very loud and clear message. Tammy Haddad: Right, because the issue is it's the haves and have nots, right? And it's not going to really be as effective as it can in society if everyone doesn't believe it's for all of them. Anne Bouverot: Completely. Tammy Haddad: Is that why you changed the name to Action Summit from Safety Summit? Anne Bouverot: Well, safety, we also think, has a tendency to focus on the risks and therefore sort of prevent people from being on the front foot. So yes, we wanted something else, but the action part is because after two summits, you do want to tell people that you're not just meeting to meet. Anne Bouverot: You want to see what concretely it means. So we're also focusing on AI in work, AI in the future of work, which is a major, major topic [00:09:00] for. workers for companies for administrations. What does AI mean? And once you've started from, Oh my gosh, all the jobs will disappear in two years and you realize that actually, well, no, they're not disappearing that fast, but, but they're changing. And so some of them are disappearing and some of them being created and the new AI experts, et cetera. But in the middle, you have like 80 percent of the jobs that are changing, and you need to train, to re skill, to reorganize work, to discuss with the unions, to, to, you need to do a lot of work, and everyone has questions around that. Tammy Haddad: Well, and it's different in every country, so how do you envision any sort of consolidation or agreement when you do get together? Anne Bouverot: So this in this area, we believe it's more in terms of best practice sharing and examples,for example, some countries and Germany is one of them as a longstanding tradition of having high level discussions with unions about any change and including about [00:10:00] AI. Anne Bouverot: We believe that that's a good practice to have because you cannot do this transformation against the unions. You want to bring them with you so that everyone works in the same direction and hopefully, and we really believe this is possible, works towards better jobs and not just productivity. So if you can have productivity and better jobs, then that's a win win and we really think this win win is possible. Tammy Haddad: There has been numerous deals made recently, strikes and resolutions and unions here in the U. S. The entertainment industry, the car industry. Are you looking at any of those ideas coming out of there and applying them? Anne Bouverot: So we're definitely looking at these ideas. We're, we're now in the phase, as you said, we have a bit of time ahead of the summit. So we're really in the consultative phase. This is not France going to tell the world what to do on AI. This is, we're reaching out to people to see what solutions they have. They've started to experiment and to see whether they can be brought [00:11:00] as examples or best practices to the summit. So yes, we're meeting with some of the union representatives to understand what they're doing and to see whether this is something that we could recommend. Tammy Haddad: How is the industry participating in the action summit? Anne Bouverot: Yeah, great question. So, of course, on something like AI, it cannot just be government. So it is a meeting of heads of states on 11th of February. But the day before is really the ecosystem day for startups, large companies, VCs, academia, civil society artists. So we really think about the community on a wide basis. And everywhere I go, I meet with representatives of all these categories and invite the CEOs and invite the VCs, invite the leaders of civil society to come. And then the weekend before, It's Saturday and Sunday. We want to have a really broad set of events in Paris around [00:12:00] AI and creation. Anne Bouverot: So any ideas you have of Tammy Haddad: Oh, we have lots of ideas. You've told the wrong people. Oh, great. We have lots of ideas. Well, I mean, people are really interested and there isn't really a place yet where you can show your way or lack of a better way to say it. Yes. But even the Kennedy Center here is putting together an AI summit in September. Anne Bouverot: Excellent. Tammy Haddad: Because the industry wants to be ahead. I mean, individuals, whether you're an actor, I work with a lot of actors at HBO, they're trying to figure out who gets their voice. How do they do it? Companies like YouTube, et cetera, are trying to figure out, well, how do we handle it? So that's, I like that idea. Tammy Haddad: Okay. So I'll see you set. What time do you want me to come on Saturday? Anne Bouverot: But please come and also we're very happy that there are events, we call that on the road to the summit, and we're very happy to partner to say that this is an event that we like and that we actually hope will generate more [00:13:00] interest for the summits like the Kennedy Center event or whatever. Anne Bouverot: We're super happy to partner. Tammy Haddad: That sounds great. Okay. Let's turn to technical testing. Yes. Okay. So we've heard red teaming, technical testing. What have you, and of course that is part of what's already taken place at these other summits. How are you going to look at that and address that issue? Refine or leave it the way it is? Can you give us some insight? Anne Bouverot: Yes. So of course, this is a track where we want to continue. We don't think we should have any further voluntary commitments, but we should take stock of what has been agreed. And we are looking at the various initiatives that are progressing with the network of safety and evaluation institutes. Tammy Haddad: Well, you've done testing, right? Anne Bouverot: Yes. Tammy Haddad: Can you talk about that process? Because part of my complaint, not to you, part of my complaint is if you can explain it more to people and they can visualize it, if they can see it, then they can [00:14:00] understand it and they don't feel like it's in a room in a black box somewhere, some elite people “who have nothing to do with me” are making this call. Anne Bouverot: I think that's a super question because one of the things we want to do is to actually make that more understandable to people because otherwise there is that feeling within the country or actually even more for countries who do not have a testing institute. They're like, what is going on that I don't even understand, but maybe to try it and explain it a bit. Anne Bouverot: An AI model is something that produces when you ask a question, produces an answer, and it's a statistically generated question and it typically makes sense, but it's not always the same one. So the way you test it is you take a set of data and you apply that data, that set of questions and look at the answers and want to check that it makes sense. In very simple terms, it's not. You need to have a set of data that you put at the beginning, you look at the output and you think. It makes sense. Just like in the way you [00:15:00] test medicine, you don't always understand how a particular medicine works, but you want to make sure that if you test it in theory and then in real life, that it does what you want it to be and that you feel it's it's okay.So that's the type of things. Yeah, Tammy Haddad: Well, that's really helpful. But will that information ever be published? Anne Bouverot: I think it has to. So it has to. There's national security testing, of course, which each country will want. to keep, and we understand that, but there's much broader testing on, for example, risks on privacy or just speed of the answer or to take some very simple things or capacity in a given language, for example, Spanish or French or whatever language. So those tests absolutely should be open in public. And there's a number already of leader boards that exist on things like Hugging Face platform, which is an open source platform or, or that companies like Scale AI publish or, and the [00:16:00] more we have public leader boards and public information, the more people will be able to go and see and say, yeah, okay, I trust it. Tammy Haddad: So they'll be able to look it up. They'll be able to search and get an answer. I think until you have that. It's still not going to be co Anne Bouverot: I completely agree with you. Tammy Haddad: Okay, good. I like it when you agree with me. Excellent. Okay, so let's, let's finish up. I know you have lots of meetings. You're here meeting with the U.S.government and tell us what you're, tell us what you're going to tell the U. S. government to do or what do you think they're going to tell you to do? Anne Bouverot: Well, of course we want to work with the U. S. government. We want to work with all countries. Yesterday I was at the U. N. and met with representatives of countries from Nigeria to South Korea to Netherlands to India, etc. So we're really in the face of gathering inputs and insights and good ideas. So, of course, we don't know what the result of the U. S. elections will be. [00:17:00] But this is an ecosystem summit, so, uh, meeting with companies, with academics, with civil society. We're inviting them, they're agreeing to come. We really hope that this will be as broad as possible. And hoping to align on the main deliverables as much as we can. And the main one, as I say, is around public interest AI. Tammy Haddad: I like to hear that. So, today, we're taping today a day that there was this public outage with Microsoft CrowdStrike. It makes me worry about how people view AI because if this just can happen, do you feel like things like that put more pressure on you? Anne Bouverot: Well, that's a good question. I think it puts more pressure on transparency, but also we have to remember that. Today, when we talk about AI, it's really when we talk about Chat GPT or Mistral Le Chat is the name of Mistral's model, which I love, or we talk about Aleph Alpha or Claude [00:18:00] or others. They're not critical systems. I'm in there. If you don't get access to your model and you don't have the answer to your question, well, that's annoying, but it's not going to be a delay on your flight. So what we really need to look at is use cases. So when these models are being used in the airline industry, in the Hilton's Industry in specific markets that's where we need to look at specific procedures to ensure it's well done and handled in the right way, as we have been doing in these industries for other types of innovations. Tammy Haddad: You've been talking about the voluntary commitments. And it's so odd to say voluntary commitments in the world we live today, because that basically means no one's checking, right? So how do you convert voluntary commitments to actual rules globally? Anne Bouverot: That's where we think standard setting is really the way to go, both standard setting and voluntary setting. and transparent publication of where you stand against [00:19:00] those. Because you're right, otherwise it's, it's something where you just maybe trust people or you don't know. So we want to, and, and it's normal to start with that, but we want to move as much as possible the route of standards and build on science because this is really science based and also focus on solutions that this can bring to people because if we do a summit where we talk about. As you say, red teaming or frontier models. What does that mean for citizens around the world? What we want to talk about is what can AI do in helping to detect cancers? What can AI do in terms of changing education? What can AI do in climate? Of course, there's more energy consumption but there's also help in terms of mitigating some of the impacts of climate change. Those are some of the things we'd like to bring to the AI Action Summit. Tammy Haddad How much time do you spend worrying about the bias issues? It seems to be an overwhelming issue all around the world. Anne Bouverot: It is, but it is also an issue in the real [00:20:00] world. So if, if I take an example, the number of women in AI, there's not that many. Anne Bouverot: And we do want to make sure that the photos from the Paris event are gender balanced or at least a bit more diverse than what we can see elsewhere. So if you have any suggestions about great women in AI that I don't yet know, uh, I would love to invite them. Tammy Haddad: I would be thrilled to introduce you. One of the key things of the Washington A. I. Network. We've met so many women. They're not public facing who've done so much work. Anne Bouverot:Excellent. Tammy Haddad: So I volunteer all of their names and all of the work they've done. And I cannot thank you enough for stopping by here. I hope you come back before February. We have lots of recommendations. We'll spare this podcast. We'll tell you after, but we really appreciate your time. And honestly, the focus that you have on making AI available to the whole world. We really appreciate your time. Thank you.

Anne Bouverot: Oh, thank you for having me. Thank you so [00:21:00] much. [a]nell ina wants us to highlight the news