- il y a 2 jours
Digital Sovereignty
Catégorie
🤖
TechnologieTranscription
00:00Sous-titrage MFP.
00:30Sous-titrage MFP.
00:35Hi, everyone.
00:36So as we all know here at VivaTech, AI and Generative AI
00:42is really the talk of the conference.
00:45And with this topic, we are going
00:47to try to know with this session if France and Europe
00:50can achieve digital sovereignty in this field, that is AI,
00:56regarding what happened in the past in cloud computing, search,
01:01advertising, et cetera.
01:03With us today, we have Jean-Norel Barraud,
01:05the Digital Minister in France.
01:07Nathalie Collin, Executive Vice President of La Post,
01:11which is the French postal company
01:13and that has many digital services.
01:15And Olivier Sichel, the Deputy Director of like SD Depot,
01:20a big investor in tech.
01:21You're also the founder of the Digital New Deal Foundation.
01:26I want to start with AI and obviously the news of Emmanuel
01:31Macron coming yesterday here at VivaTech.
01:33He spoke about AI and regulation in Europe.
01:37It was a day of the vote of the AI act yesterday.
01:43From what we understood, he was a bit critical of that regulation.
01:47What is going to be France's stance on this regulation
01:51and that tension between innovation and regulation?
01:56Our top priority, and President Macron repeated it yesterday,
02:01is to help the emergence in Europe in the next few months
02:07of generative AI models at the state of the art.
02:12To do that, we need to invest massively.
02:14And this is what the President has done yesterday
02:18on this very stage by saying what we were going to do for talents,
02:23for data, for computing power,
02:26and, you know, for the support to our AI ecosystem in France.
02:33And then we need to regulate smartly.
02:37There is no contradiction between regulation and innovation.
02:42To the very contrary, regulation can help bring competition
02:46in markets where you have sort of monopolistic situations.
02:52And there, regulation is very helpful
02:55because it helps unlock economic activity and innovation.
02:59Moreover, you need regulation in order to build
03:02a single digital market in Europe.
03:05One of the key obstacles that startups in France face,
03:09startups in France but in other EU member states,
03:13is that they have now access to financing,
03:16but they need a broader market to accelerate their growth.
03:19But this is not possible when there is a fragmented market
03:22across the European Union.
03:24So when we regulate, we also set common grounds,
03:28common rules across the EU,
03:30which is going to simplify and accelerate the growth of European startups and scale-ups.
03:37So the way we want to approach the regulation of AI is this.
03:41The EU was the first democracy in the world to decide to invent regulation or a regulatory framework for AI.
03:52It's called the AI Act and it does something very, very important and useful,
03:57which is to set or to highlight what the responsibilities and liabilities are for each of the players in the
04:07value chain.
04:08This is very important because when there is uncertainty about liabilities and responsibilities,
04:13this can curb adoption of new technologies.
04:20Let me try and explain why very briefly.
04:23When we are going to introduce AI in mobility systems, in health solutions,
04:29there is going to be issues when human lives are at stake.
04:35And when there are accidents, liabilities are going to need to be established.
04:39If there is no clarity as to how liability is going to be established,
04:44then there will be reluctance in terms of adoption of such technologies.
04:50The AI Act that the EU has decided to adopt clarifies liabilities
04:57and says that for sensitive use cases of AI,
05:00those that develop AI systems need to provide sort of, you know,
05:05disclosure documentation, they need to provide documentation,
05:11they need to audit their systems before they are used for these sensitive use cases.
05:17But that for non-sensitive use cases, think video games, for instance,
05:23well, AI systems are not subject to specific constraints
05:27because there is no life at risk at the end of the day.
05:31So this is great because it clarifies things very nicely and we help adoption in Europe.
05:37What's a concern for us is that the, you know, the adoption of such a regulation at European level
05:49is a negotiation between the Council, the European Parliament and the European Commission.
05:54The position of the European Council is the one that I just mentioned.
05:58The position of the European Parliament was formed yesterday and the European Parliament tightened strings,
06:06perhaps a bit too much, by saying that generative AI models that are associated with a number of new risks
06:14should be assigned to the same constraints, whichever is the final use cases of these generative AI models,
06:21to the AI systems used for very sensitive and critical use cases.
06:27To us, it is going too far because it's going to generate too much of a sort of a drag
06:35on the development in Europe
06:37in the next few months of these models.
06:39So we are going to push back in the negotiation with the EU Parliament against this sort of thing they
06:45added
06:45to the version that we adopted last December.
06:49So you see what we're trying to do and what the President spoke about yesterday on this very stage,
06:54is investing massively, regulating smartly, in order to ensure that France and Europe can develop these models in the next
07:03few months.
07:04Olivier, do you agree on this regulation? This is the way to go?
07:08And is there any risk that it would actually, by protecting the European framework,
07:16that it kind of kills innovation in this new field?
07:19Yes, absolutely.
07:21I'm going to start by saying something that may be a little bit surprising.
07:26I'm one of the very few Europeans who signed the open letter calling for a pause in AI.
07:33And, you know, having set up a think tank called G10U, fighting against Google, Apple, Facebook and big tech monopolies,
07:42I couldn't imagine to be aligned with this letter.
07:46But in fact, I was not calling for the pause, but the time used for the pause was very interesting.
07:54What did it say, this open letter?
07:56He said the future of AI should not be that only to tech leaders, unelected tech leaders.
08:03That was exactly what was written.
08:04And that the future of AI should be also determined by policy makers.
08:09Policy makers such as Jean-Noel.
08:12And why so?
08:13Because, you know, we're living in a democracy.
08:15And if we go back to democracy, democracy is the power to the people.
08:20And Jean-Noel has been elected.
08:21He used to be a congressman.
08:23And he's been elected.
08:24So he's got a voice to do what?
08:26To protect the people for AI.
08:29Of course, AI is going to bring a lot of things.
08:32And I'm sure Nathalie, who is in business, is going to develop what AI can bring because it's great technology.
08:37Like nuclear is great technology.
08:39But we also know that there are risks.
08:41And to defend us from our risk, I think that in this letter, we're saying we should work with policy
08:45makers to do what?
08:47To invent the smart regulation Jean-Noel is calling for.
08:51We need to invest massively, yes, but also to regulate smartly.
08:55And I was very surprised to see, for instance, that in this letter, they were calling for increasing public budget
09:03for funding safety AI research.
09:07Public budget.
09:07Because we know that, you know, tech leaders and the private sector is going to develop AI, is going to
09:13do it for what?
09:14For profit.
09:15I mean, they are VC-backed.
09:17They need to do some profit.
09:18And even, you know, OpenAI started as a non-profit.
09:21And at the end of the day, it's worth one billion.
09:23Something I cannot figure out, how you start non-profit and you're worth one billion.
09:27There's something missing in my scheme.
09:29But maybe, you know, I'm too much European and I don't see Silicon Valley minded.
09:34So that's why, you know, I'm falling towards.
09:37Natalie, LaPost has millions of customers.
09:41It would actually, I'm sure, help the LaPost employees to use Generative AI to have interactions with them and save
09:48time, be more efficient.
09:50Is LaPost experimenting with this?
09:52Are you using any models?
09:54Yes, of course.
09:56First of all, I'd like to say that ChatGPT now is clearly the emerged part of AI.
10:03But we have been working on AI for many years.
10:06Acquiring companies such as ProBias, for example, which is a spin-off of Enveria.
10:13Enveria is a famous researcher institute.
10:15So we have been in AI for ages at LaPost.
10:19And now we are working on Generative AI and for businesses.
10:22There is not only ChatGPT as a possibility for Generative AI.
10:27For example, we are working here right now with LightOn and Laial.
10:31And we are working with them on setting up Generative AI on our own data centers and on NIMSBOT as
10:38well.
10:38To have a really sovereign facility on Generative AI.
10:44And of course, with 900,000 people a day in our post offices, 250,000 people, employees, and billions, obviously,
10:56of objects that we are dealing with.
10:58Generative AI for us is a big opportunity.
11:01It's a big opportunity in SEO.
11:03It's a big opportunity.
11:05We want to launch, effectively, LaPost GPT.
11:09LaPost GPT will be the whole content we have, the whole data of LaPost, on-premises with Generative AI to
11:17support and really enhance the work of all our postal workers.
11:22And it doesn't mean we are going to replace them.
11:24And then, at the contrary, we are going to augment them with AI.
11:28Minister, does France, through his civil servants in France, does France want to experiment with Generative AI?
11:38And which models would you use?
11:40First, let me congratulate Nathalie Collin and LaPost for launching such a great project, LaPost GPT.
11:48Yes, I think you can give her a round of applause.
11:51Go, go, go, go, go, go, go, go, go.
11:55Why?
11:59It is very important when we talk about Generative AI that we think about these public investments that we need
12:08to make in talents, in data, in computing power.
12:12It's important that we talk about regulation.
12:14mais comme Olivier l'a dit,
12:18les gens qui n'ont pas accepté l'arrivée dans leur vie quotidienne
12:24sont nos co-citizens.
12:26Nos co-citizens ne importe pas les détails de régulation,
12:33ils ne importe pas les détails de nombre de GPUs que vous avez
12:36pour qu'on a train la générique de la modèle de France et de Europe.
12:40Ce qu'ils carent est ce qu'ils vont changer dans leur vie quotidienne.
12:45Et il y a à l'heure, il y a deux choses que nous avons déjà vu
12:49et qui peuvent donner des choses qui sont mieux et qui vont changer
12:55les jours de la vie quotidienne.
12:57Une est la production.
13:00Il y a peu de recherche que nous avons donc,
13:03mais il y a un article qui a été publié en April
13:07by researchers at the NBR in the US
13:11that shows what the impact of the adoption of GPT-like models
13:18for customer service did in terms of productivity.
13:23What do we see?
13:24Think of the customer service of a large software company,
13:295,000 people.
13:30et compare the productivity of teams of the customer service
13:34that are equipped or powered with GPT-like tools
13:39and those that are not.
13:41So those that are powered with these tools,
13:44when they receive a customer call or a customer email,
13:47they're going to put it into the GPT tool
13:50and they're going to receive suggestions
13:53in terms of how to solve the customer request
13:55or the customer problem.
13:57What are the results of this study?
13:59is that productivity goes up by 14%.
14:02And most importantly, what we're seeing
14:05and that is kind of counterintuitive
14:07is that it is the least productive workers
14:11that see their productivity goes up by the most.
14:15So you see a 14, let's say 15% increase in productivity
14:18that is entirely driven by a 30% increase in productivity
14:24by sort of the least productive half of the workers
14:29at this company.
14:32Such amount of productivity increases
14:35is a major driver of prosperity
14:40and let me tell you that this our co-citizens will understand.
14:44And the second important dimension
14:46of what these systems will change in their daily lives
14:49is access to public services.
14:52And that's why I wanted to congratulate Nathalie
14:54for what Lapos is currently doing.
14:56We, old democracies,
15:00have sort of missed the digitalization of our public services.
15:05Why?
15:06Because we had public services delivered through layers of bureaucracy.
15:10And what we've done when digital technology has come
15:15is that we've digitalized bureaucracy.
15:18And what it has done to our co-citizens
15:21is that it's made their lives somewhat even more difficult
15:25when they need to access public services.
15:28While in younger democracies,
15:30I'm thinking of India, I'm thinking of Ukraine,
15:33I'm thinking of some African countries
15:35that skipped the bureaucracy sort of stage of their development.
15:41digitalizing public services
15:44has not led to exclusion of part of the population,
15:48but rather to inclusion of hundreds of millions of people
15:51into democracy, into access to public services.
15:55And I think we can catch up on what we've perhaps missed
15:58thanks to Generate TVI.
16:00I think we can make access to public services much more easier
16:05because you can access those public services
16:07without a keyboard, without a mouse,
16:10without, you know, a very complicated website,
16:13but just with common language.
16:15So you see productivity and access to public services.
16:19This is what our co-citizens are going to see.
16:22And this is the reason why they will support
16:25the adoption of Generate TVI.
16:27If we talk about digital sovereignty, we have to talk about cloud.
16:31Obviously, big American companies are dominating this world,
16:34Amazon, Google, Microsoft.
16:37But France is developing trusted cloud offers at La Poste.
16:43And like SD Depot, you're doing new spot
16:45with actually French partners, Dassault System, Brigg.
16:49Other groups like Orange are working with Microsoft.
16:53Thales is working with Google.
16:55Olivier, do you believe it's possible to do
16:58a sovereign cloud with these American partners?
17:00Thank you.
17:01Yes.
17:01As a chairman of Newspot,
17:03first of all, maybe to tell you what is Newspot.
17:06Newspot is a French cloud, a trusted cloud.
17:09So it's been set up by the combination of four actors.
17:14That's the system.
17:15It's providing the factory of the service, which is outscaled.
17:20So we're talking about the most advanced software development company in France.
17:26WIC Telecom, because you need to know telecommunication in cloud.
17:30La Poste, which is providing software of the service.
17:34And it was really a reference to developers in administration.
17:38and we, CDC, were providing the fund.
17:41And our will is not to go against Google, Amazon, and Microsoft
17:48because it's too late in certain things.
17:50But we want to focus where trustworthy cloud is necessary.
17:55Talking about health, Jean-Marie,
17:57well, nobody wants to have his data health,
18:01personal data health, everywhere in the cloud.
18:03You need to have trust where you are going to give you data health.
18:06administration, tax systems.
18:09You don't want to have that everywhere in the cloud.
18:11So we've been launching that.
18:14The community welcomed it increasingly positive.
18:17It's been amazing the number of companies calling us.
18:20And what we discovered is that we just arrived at the right moment
18:23because some of the people, I'm talking about the banking system, for instance,
18:27they just said, well, I'm just realizing that I'm locked in the system,
18:31and Amazon, and Google, and also that the prices were going up like mad.
18:37And we're saying, that is now a CEO concern, saying, okay, I've saved the bills.
18:43The bills are raising.
18:44I was told that cloud was scalable, flexible, at that time, but it cost a lot.
18:49So it helped me to change that, helped me to make competition.
18:52And we go back to regulation because there is no competition.
18:55At that stage, you really just realize that some of the customers are locked legally,
19:00technically, and they cannot give back and get back their data, their systems.
19:05So NuneSpot was very welcome by the community, not only because it was sovereign,
19:09yes, it's an asset, but also because it increased competition.
19:13Natalie?
19:14Yes, of course, we are part of that story, and it's very important to us as well.
19:19Just to put generative AI, you have to put it on the cloud, and it will be definitely on NuneSpot.
19:24And we believe that sovereignty is very, very, very important.
19:29For example, if you use your data, and you train, and you improve GPT,
19:33and you get to GPT-4, GPT-5, GPT-6, at what time?
19:37Is a GPT-7 version coming back to Europe when it will stay in the U.S.?
19:42We know that, for example, they trained the GPT in the U.S.
19:45for six, nine months with American companies.
19:48It comes back in Europe later, so we lose in competition.
19:51So we don't want to lose in competition.
19:53We want to be in the competition.
19:55When we speak digital sovereignty, there is a question of talent.
19:59We know we have a lot of engineers in France, but still, this is key,
20:04and especially in tech, there is a lack of female talent being hired in startups
20:09and founding startups.
20:11Also, at La Post, I think you care about this a lot.
20:13Yes, look at the room.
20:16I could ask the women to stand up.
20:19Maybe they won't, but you're not that many, maybe 25% of the audience,
20:25and it's not enough.
20:26You're just not enough, and we cannot hear your voice enough.
20:31So what we did first is that we set up an AI and data school,
20:38and we decided, I have to say that I decided, it would be really a gender parity.
20:46And we achieved that because it was just a decision,
20:49and we would expand the school if we have enough women to attend that school.
20:53It's a school you can attend if you are a postal worker or if you are an external person,
20:58so you are welcome.
20:59We have PhDs, we have Masters in Science, and so on.
21:02And we have the first promotion this year, 51%, 26 women.
21:07And you are going to change the world.
21:10And if I may make a call to action, maybe.
21:13It's you in this room, the men and the women, you are designing the world of tomorrow.
21:21And you have a tremendous responsibility in making this world inclusive,
21:28in making this world open to gender equality.
21:32So you have to look for talents.
21:34You have to push them.
21:36You have to support them, to train them.
21:38Can you imagine that at La Poste, for example, we already trained 50,000 postal workers to AI?
21:46Just because if you don't do that, you're not doing your job.
21:49Your job is to make the tech the cement of our society tomorrow.
21:56Minister, on this, well, people appreciate that.
22:02On this question of talents and retaining talents in France and in Europe,
22:07we know it's particularly key in terms of engineers, top engineers, and the diversity of them.
22:13Sure.
22:14So let me once again congratulate Nathalie and Olivier for NUMSPOT,
22:22for this newcomer in the cloud market that is both trusted in the sense that
22:27it's going to allow some of our sensitive companies and administration
22:32to protect their sensitive data.
22:35And it is a newcomer, so it brings competition.
22:38And competition is good, because competition gives more freedom to users,
22:42in particular to compete.
22:43So we support the development of NUMSPOT,
22:46and we are very grateful to the four partners of this project.
22:51In terms of talent, talent is clearly our key asset in the race for generative air.
23:01The best way to illustrate this is to go back to yesterday,
23:07about this time on this very stage with the French president,
23:11was Arthur Mensch.
23:13Who's Arthur Mensch?
23:14Arthur Mensch graduated from a PhD at INRIA.
23:19He then joined Meta, or Google DeepMind, one of the two.
23:26DeepMind.
23:27Where he basically learned the ropes about generative AI models.
23:31And then he recently decided to join forces with some of his,
23:39perhaps classmates or some of his friends that are also involved in generative AI
23:44in other major players, major incumbents in this field,
23:48to set up its own company.
23:50This company is called Mistral.
23:52It raised the largest ever seed round of financing in Europe,
23:58that was announced this week.
23:59And we have great hopes in this ability of sort of French-grown talent
24:05that are coming back to France
24:08to help us achieve leadership in this field.
24:12But this also tells us that we need to continue growing talent.
24:16And here, what it means is keep investing in our research and higher education system.
24:24But most importantly, it's doing what Nathalie said,
24:28which is to draw half of the population that is today remote
24:33from the digital and the entrepreneurial careers to come to it.
24:40And that's bringing women to entrepreneurship,
24:43to tech and science more generally.
24:47The Prime Minister will be here tomorrow
24:49to sort of basically report on our advancements
24:54on some of the commitments she made on March 8th,
24:58which is the International Day for the Rights of Women.
25:01We want to achieve parity in scientific sort of path in 27, 2027.
25:08And we want to launch 10,000 grants for young women and women
25:17that want to undertake a scientific career path.
25:22Thank you.
25:23Our time is over, but I've got one last question.
25:25Tomorrow here also will be Elon Musk.
25:28You've been quite vocal about Twitter
25:30and the content moderation that it was lacking in Europe
25:34and actually said that France could block Twitter.
25:38if he didn't respect it.
25:40Are you going to discuss this with Elon Musk tomorrow?
25:43So, you know, the entire world gathers together in VivaTech,
25:48so it's no surprise that Elon Musk would show up.
25:51Of course, Elon Musk is a great entrepreneur.
25:55He's a visionary leader.
25:58And we have many things to discuss with him.
26:01But in particular, of course, we're going to chat about Twitter
26:05that as any social network will need to conform to the new European rules
26:13that will come into force on August 25th of this year
26:16because last year we adopted the Digital Market Act
26:22at the European level
26:24that imposes to these social networks to act responsibly.
26:31And so it's very important that Twitter conforms to its obligation.
26:34Otherwise, Twitter will face fines up to 6% of global sales.
26:38and in case Twitter does not make the adjustments,
26:45then indeed Europe can ban Twitter if Twitter does not comply.
26:51We don't want this to happen.
26:52And so we are having discussions on a daily, weekly basis
26:57with Twitter teams to try and ensure that they can conform in time
27:01to these new European rules.
27:03and we're looking for this tomorrow, certainly.
27:06Thank you to the three of you.
27:07Thank you very much.
27:08Thank you.
27:08Thank you.
27:09Thank you.
Commentaires