2026 The Geopolitics of AI
Transcript
Jean-Marie Guéhenno:
Okay, we are about to start and I want to thank our panelists. I think we should be joined by Emmanuel Bacry in a few minutes, but we'll start anyway now. This morning, we heard about the substance of AI, we heard about what's going on backstage, all the infrastructure of the data economy, discussing data centers, submarine cables, satellites, and all that is necessary for this connected world to work.
In this panel, we really want to draw some conclusions or at least sharpen our questions on the geopolitical dimension of what's going on. For that, we have four panelists who each bring a lot to the discussion. I will start with Kori Schake, who I have known for many years.
Delighted to see you again, Kori. She works at the American Enterprise Institute and has been working on security for many years. When we first met, when the security was not seen as it is seen now, it's really profoundly different.
I want to ask you, Kori, how disruptive is AI for the international security architecture? Is it going to consolidate the dominance of a few AI superpowers, or is it going to disseminate power in an unpredictable way for the good and for the bad?
Kori Schake:
It's too soon to tell, because we're at the early stage of what the people who understand the technology best are telling us is going to be a bigger economic and social upheaval than the industrial revolution. The industrial revolution took about 100 years before the social and political and economic effects were really fully evident. It's too soon to tell.
It's also to be on with Audrey Tang, who I admire enormously and would defer to her answers on these things. But my sense is that there's a two-stage battle going on. One line of effort is the countries that are pushing the frontier of the development of the technology.
And there, it does really matter who's first, because it looks like a first-mover advantage will drive the frontier forward in potentially different directions and will have enormous national security consequences. I mean, even if you just look at the way artificial intelligence is changing military operations and intelligence practices by allowing the assessment of enormous pools of data, so making things more transparent, easier to target. But you can also see that that doesn't add up to winning wars.
So it's early on in the development. But the second track, I think, which goes to the latter part of your question, Jean-Marie, is that it's not clear the first-mover advantage of people at the frontier are going to be how artificial intelligence is adopted across economies, across states, across political communities. And there, it looks to me that the first-mover advantage might actually be detrimental because lesser models can use the data pools that the frontier organizations have developed and create solutions to problems that are plenty good for what is needed in that space.
So I think it's too soon to tell, but there look to be enormous advantages on the frontier for being a technological leader. And it's not clear that those advantages will pertain across broad-scale adoption.
Jean-Marie Guéhenno:
So we are going to continue in a way with the same question, but focusing on China and the United States. Melanie Hart, you are focusing on that tech competition actually goes beyond AI between China and the United States. You're regularly going to China, checking on the state of play.
How do you see that competition? How do you see China leverage its technical and technological prowess into geopolitical advantage or not?
Melanie Hart:
That's a really great question. And there are three ways that a lot of people in Washington and elsewhere are looking at how to measure who's ahead in AI capability. One, Kori mentioned frontier models.
Who's doing the things that no one has done before with AI? So, ChatGPT has had moments where they've done that, for example. DeepSeek has also had moments where they've done that out of China.
But overall, when it comes to frontier models, who's first? The US tends to be first out of the gate. And I think we're looking at that too much.
I agree with Kori, if I understood you correctly. There's too much focus on that in the United States sometimes. Thinking about whoever spends the billions and billions of dollars to do the first model before anyone else will maintain leadership is not necessarily the case.
The second metric to look at and where people are is compute. How much data can you crunch to power these models? And some of that is to do with chips.
The United States and its allies have the lead on the most advanced AI chips. China's about five to seven years behind the US and its allies on the advanced AI chips. And we're hearing from Chinese companies and the Chinese military that that is constraining what they can do with some of the AI deployment.
But another aspect of compute is electricity and data center infrastructure. And that's where we're really starting to run into barriers in the United States. I live in Virginia where our electricity bills have shot up because there are, I think, something like 1,500 new data centers being built in Virginia.
And now American citizens are starting to protest them because there's water pollution. They're raising our electricity bills. People don't want to live next to a data center.
So I think we're going to see some shift in that balance of power and compute because China can roll out nuclear projects much faster than we can. They don't really care if you want to live next to a data center or not. In China, the data center will go wherever Beijing wants to go.
So compute is the second. But the third, I think, is going to be the most important, and that's deployment and distribution. Who uses the AI that you come up with in a lab?
And that's where China is really focusing its efforts. And this is where China has the opportunity to really shape the way that AI is deployed globally. Here in the United States, you know, Chad GPT and Claude and these different models are at the frontier in what they can do, but they're also relatively expensive.
What China's doing is deploying deep-seek and finding ways to do an almost as good capability, but a lot cheaper. You don't need as good chips. You don't need as much electric power.
They're doing open-weight, locally deployed models. That means you can just take it and install it on your own computer and tinker with it and change the way that the model works. You can own it a little bit more, whereas with Chad GPT, it's kind of like doing a Google search.
Everything lives within OpenAI and Chad GPT. China's experimenting with smaller, cheaper, easier-to-deploy solutions that are probably going to be more like the kind of solution that Global South nations in particular are going to want, because they're going to want to be able to have their data in-country or in-company and not give it to OpenAI. They're going to want to know, how can I deploy AI at the lowest cost with the easiest, lowest-end chips, without consuming too much electric power?
Here in the United States, a lot of leading American universities can't get access to the most advanced chips to experiment with in the lab, because OpenAI and some of these other companies are buying them all up. Deployment is probably going to be most critical for who shapes the global landscape. That's an area where right now I think Washington is paying too much attention to, let's be ahead on frontier.
Let's be first, as Kori put it. Let's maintain our chip edge. Both of those are important, but if nobody buys what you're selling, or nobody can afford to buy what you're selling, then that's where we really lose the game to China.
Then just four areas that I'll be watching really closely in terms of US-China competition. The first one is China likes to turn a cost advantage into dependencies. We've seen with 5G telecom that Huawei was always the cheapest.
Once you bought Huawei 2G, you could only go for Huawei for 3G. You could only go for Huawei for 4G. They aren't compatible with other systems, so they tend to lock in their customers to the Chinese ecosystem.
There's a lot of risk there with AI that maybe you go for DeepSeek or another China alternative because it's cheap today. Then you find, well, the only chips you can use are Huawei chips. The only data center you can use is a Chinese data center.
Then that opens you up for risk. A second factor that I'm really watching is information shaping. My team here at the Atlantic Council did the same search in DeepSeek, China's AI model, and in Western AI models, and the information is just different.
The Chinese government shapes what you hear and know about the world. I'm sure, Audrey is more aware than I, you're definitely going to get different information about Taiwan, for example, but also about human rights and how well the Chinese government performs. There is a big risk if the entire planet is getting their AI search information from DeepSeek because when it came to general basic internet search, China's alternative to Google was Baidu.
It's horrible. Nobody ever wanted to use Baidu because it is such a mess. It is so difficult to use.
We were in China a couple months ago and I tried to use Baidu just to get a restaurant address and it was giving me search results dated in 2003 for no clear reason whatsoever. It's just very messy. China never took off on internet search, but it could on AI-powered internet search.
That would really shape the information that the global public has access to. Then just one other point, China, we really are watching closely. How is China working to export AI surveillance models?
We know that in previous years, there have been issues with China exporting surveillance networks that would be, for example, a network of cameras that you can deploy around a city. Now they're looking at new models that can predict in advance what kind of person, who is going to be a possible dissident in the future, who's going to be a political agitator in the future, using AI models to look at who are their friends, to listen to their telefilm conversations, to review their emails, to use geolocation and see where do they go around the city, what internet websites did they look at.
China is experimenting with AI for a forward looking surveillance program, not just tracking what people do now or in the past, but what kind of person are you and should we be worried about you in the future. I have huge concerns about that being exported to other countries as well. That's it for me.
Jean-Marie Guéhenno:
I'm going to soon turn to Audrey, but before I want to ask Emmanuel a couple of questions because it happens that this morning we already discussed briefly, as he was presenting, when the various types of AI, we discussed briefly DeepSeek. The question is really, do you, I mean Emmanuel is a AI scientist, and do you really believe that there are shortcuts in a way to using much less data and being much cheaper? I mean, how do you see the competition from your standpoint as a...
Emmanuel Bacry:
Yeah, before answering strictly speaking the question, I would like to comment on some of the things that were said. First of all, I think we are focusing a little too much on LLMs and chatbots. AI is not just that.
So we don't have to forget. I mean, the whole buzz in the whole world is about that, but we should be careful not just, you know, being part of this buzz. It is clearly a big thing.
I'm not saying, I'm not trying to minimize it. And it's a big, I mean, with plenty of applications and it will transform quite a lot of things. But AI is also a lot of other stuff.
So when we are saying about the fact that we need a huge amount of data, this is really about LLMs, a huge amount of compute, it's also about LLM. So we have to be careful a little bit about that and talk also about other AIs that could have huge impacts. I mean, if we talk about health, I know health very well, there are huge impacts on health without LLMs. That's the first thing I wanted to say. Second thing, maybe it's not politically correct what I'm going to say, but what you are saying, Melanie, about China, the fact that, I think it's you who said that, I'm sorry, maybe it's not you, but when the information shaping, the search information, there is one thing that I like in the, I'm being a little bit provocative, OK, but there is one thing that I like in the Chinese legislation and regulation, which is that it says clearly that the alignment of the foundational models should be done according to the Chinese Communist Party. But it is clear, at least, open AI, we don't know anything about the way they align the model.
And OK, you are saying that when we are doing search information on deep-seek, it is clear that there are plenty of problems and so on. But because we know China very well, but it is totally biased what we get with ChatGPT, with Claude, with Gemini and so on, that it is more subtle and nobody knows exactly how it was trained. I don't know which one is the most dangerous.
It is not clear for me. It is an open question. I don't have the answer, but I think both are kind of dangerous and not only the one which is openly biased and we know very well the bias.
So I think we should be also careful about that. And so now I come back to the, and yeah, sorry, one more point. I think one of the very big challenges is also, and you mentioned, I don't remember who, deployment.
In deployment, there is one challenge which is extremely important and very hard, is the evaluation. The evaluation, there are plenty of challenges about evaluation, not even talking about LLMs. On LLMs, it is a nightmare to evaluate. But even if we talk about regular AI, evaluation in real life is a very big problem.
And again, please, I will take an example in health because a lot of things are very clear in health because it is the center of so many problems and so many touches everybody. So when we deploy a medical device that has AI in a hospital, of course it is certified. It is FDA cleared or in Europe it is certified and every country has this certification.
But certification is kind of a still ideal world where you have a closed space and you test specific things. When you deploy in real life, when you deploy in a chain of process where human beings interact with the AI, everything can go wrong, even if the AI is very good. But you have also to evaluate the whole chain of process that will change because you change the one brick, one step by an AI model.
And there are examples in hospitals where some AI were deployed, very good level AI, and where the number of deaths increased dramatically. So there are many challenges about that. Now, sorry, I come to your question now, Jean-Marie.
So your question, sorry, it was about the fact that do we need a lot of data or can we have... Yes, yes. So, yeah, as I said this morning, but nobody is sure.
So maybe some people in this roundtable would say, but it is likely that DeepSeek and Quen used what we call distillation of the big American models. So, of course, they didn't need as much data and it costs less money because they basically, they used the big models to take all the knowledge of these models. Now, would it have been possible really to do a DeepSeek or Quen if ChatGPT, Claude and Gemini didn't exist?
It is not that clear. I don't have a clear-cut answer about that. But I want to...
And about the quantity of data, this is necessary for LLMs. This is very clear. It is necessary for LLM. But not at all, not in the same size for many other AIs.
Melanie Hart:
Can I ask a question? I think this is an important point. So if you go to the DeepSeek website, for example, and look into their pricing, it is fascinating.
They describe their pricing in direct relation to OpenAI. And they say, like other models who require a monthly subscription minimum, we only charge by the search or by the token. So OpenAI...
Am I getting to the heart of this follow-up question correctly? OpenAI is cheaper for users to run. So, for example, a company based in South Africa or Colombia or anywhere in the world, if they're comparing some enterprise options between DeepSeek and Claude and OpenAI, everything that I'm seeing is that the DeepSeek version is going to be currently much cheaper for them.
And I think that's something that we need to fix.
Emmanuel Bacry:
Yeah, but that is different from the question of Jean-Marie, which was the training with the quantity of data. But I totally agree that it seems like they have found a way of running it in a cheaper way. And when you look at the researchers in AI today, everybody is looking for models that are simpler and that do as efficient LLMs as what we have today, but in a simpler manner so that the energy needed to run would be much less.
And there was one announcement, I mean, almost every month there is such an announcement of a big improvement about that.
Jean-Marie Guéhenno:
I want to turn to Audrey Tang. When you've been the first digital minister in Taiwan, you are the Taiwan cyber ambassador, and you've been working a lot on the positive views of AI. And as Emmanuel reminded us, I mean, AI, large language model are just one compartment of AI.
There are many uses of AI. And so if one looks beyond the US-China competition and looking at AI and not focusing just on the security side of things, what do you see as the upside opportunities created by AI in connecting citizens, in increasing capacity to solve societal problems? What is the potential?
And I think it's important for this group of diplomats. I mean, they come from rich countries, from developing countries, from really all over the world. And what is the potential of AI to redistribute power for good, possibly for bad?
Audrey Tang:
Certainly. So I want to continue the thoughts that we heard about evaluation, actually, because I'm now also academically in Oxford, working on AI alignment, but not aligning our outcome utility, but rather aligning with a participatory process. So I want to tell a story about how we aligned AI systems with the participatory process.
And it happens in my home, actually. So my dad is a political scientist, journalist, who was in Tiananmen, actually, until 1st of June in 1989, so covered the whole thing. And a few months ago, he has a health issue.
And so he found that no matter which chatbot he talks to, the chatbot keeps suggesting fantastic ideas, yours, brainstorming, and so on. And he gets lured into this kind of spiral now situation. But because he's a journalist, he immediately diagnosed that it is just trying to earn the subscription and not loyal to the family relationship.
So my point here is that we very quickly found a solution to his dilemma. So with my parents' explicit consent, we built a local model that's entirely running on a Mac. And so it doesn't need a data center.
And it is a language model plus a few other narrow models, but everything fits into one Mac computer. And it only joins our family signal group. And of course, he can directly message the bot.
But the evaluation metric was done with my mom, who said that every time the bot speaks, it should be reducing my dad's dependency on the screen and should restore his peace of mind to the reality around them. So it's a little bit like how in Taiwan, we make sure that the kids in classrooms use only large screens, one laptop or a tablet per pair of children. So it's always pro-social, not anti-social.
And the bot, every time it gives a slightly isolating answer, we can just use what's called directional steering, which is a very small amount of data, just 10 sentences. And within one minute, we can train the network to behave in the direction we're tuning. And because it's running entirely locally, it's not sending the data anywhere to the cloud, not to Beijing, not to Silicon Valley.
So my point here is really a governance point, which is, are we thinking about AI on top or AI on tap? If it's AI on tap, then the people touched by AI can't steer it in real time. But if it's AI on top, then the extractive relationship, whether in Beijing or Silicon Valley, we are like planktons, our human data being extracted into this huge oil rig.
But if it's within the local community, then people can hand it together. This is why I call data soil, not the data oil. It's purely regenerative.
And by the way, it's faster than the cloud models and higher quality. So I really don't see the reason why not to use this kind of local models.
Jean-Marie Guéhenno:
On what Audrey just said, I saw Emmanuel, you were... Sorry?
Emmanuel Bacry:
I thought you had a reaction, No, no, positive reaction. I agree. And I do believe in small models more than the big GPT models, where nobody sees a business model in these big models.
Everybody's losing money with these models. And they are too generic. They are supposed to know everything about everything, which is stupid.
It will never reach this goal. This is very clear. So they keep, you know, they keep adding layers and layers on top of it so that they repair something that they see and they repair something else.
But we should totally change by design these things. And so I do believe absolutely in much smaller models that you train on specific data, local data, specific to a small population and with a purpose, a specific purpose, trying to answer specific questions. This is clearly has a lot of applications and a lot of value.
So I totally agree with Audrey.
Jean-Marie Guéhenno:
Thank you. I want to come back to Kori. You said at the beginning that it's much too early to answer the kind of existential question that I asked.
But as you see it today, not projecting in the future, do you see AI being in the existing conflict, in the existing battles that we see? Do you see AI playing, I mean, rebalancing situation between actors? We have the Ukraine war.
We have the Iran, Israel, US war. So these are war that are existing, that are happening today with extensive use of artificial intelligence and data in both cases. How do you see the impact in those existing conflicts?
Kori Schake:
So I think artificial intelligence is changing some of the trade craft of national security because it's harder for spies, for example, to have unsupervised engagements, right? It's increasing the transparency both on the battlefield and its increasing search capability. So you can, you know, if an intelligence agency records every foreign telephone call, the constraint isn't on the recording, it's on the processing of information, people to listen to it and identify where there are patterns.
And artificial intelligence is genuinely brilliant at that work. Just to take one small example, I was on the historian advisory committee to the State Department and we did a test run on whether using artificial intelligence to evaluate freedom of information document releases could assist. Would it increase the need for human supervision?
Would it reduce? And what we found is that it takes away the majority of cases which are easy to solve by a decision rule. And that freed up the actual talented people to focus only on the hard cases.
And so it didn't so much reduce the need for human labor as focus the need for human labor on the most intellectually challenging elements of the problem. So that's one, I think, really valuable test case. And as Emmanuel said, I can see a thousand applications in the healthcare sphere for that, identifying new vaccines or new treatments or solving difficult problems that a decision rule can solve for you, but it can't substitute for judgment.
And so in warfare, identifying where things are on a battlefield, both for situational awareness and for targeting is an enormous advantage. And if those targets are targets that don't require human judgment, right? Every tank moving on a battlefield, every enemy tank moving on a battlefield should be targeted.
That doesn't require human judgment unless you are worried about your munitions stockpiles, in which case you can also put in an algorithm to prioritize certain types of targets or locations. But what it cannot do for you is say, should I do this? That's a fundamentally human decision-making.
And in warfare, that's really important decision-making. Back in the 1990s, the Commandant of the Marine Corps, General Charles Krulak talked a lot about what he called the strategic corporal, which is the lowest ranking soldier on the battlefield can make a choice that will have strategic implications. And there's no amount of automation that can prevent bad human choices in conflict situations and can replicate good human choices.
So what is changing, I think we are seeing is the persistent surveillance at low cost that drones either relying on a communications network or independently targeted. That has reduced the ability of mobility on the battlefield. It's reinforced the advantages of defenders over the prospects of attackers.
That's a legitimate change. The second legitimate change of what's happening as a result of artificial intelligence is it's so easy to print low cost drones. And if you don't care whether 947 of a thousand get destroyed, then you can have enormous effects.
So the cost effectiveness, it used to be, you had to be a great power to afford persistent surveillance on the battlefield. That's not true anymore. And so that's an advantage to insurgents.
It's an advantage to terrorist organizations. It's an advantage to countries of the global South in defending themselves against higher capability militaries. So I think it's changed those things, but what it hasn't changed is the grit of societies refusing to accept bad outcomes and other fundamentally human judgments on which winning and losing wars continues to rely.
Jean-Marie Guéhenno:
Thank you very much. That's actually very important, I think, for many countries around the world, what you just said. But let's now move beyond war and think about the positive views of AI.
Audrey gave us a wonderful example of a small model that produces very good results at relatively low cost. And so I'm tempted to ask the question to all of you. As we look forward to the future development of AI, I mean, Melanie, I think you said there was too much obsession, people were too obsessed as who is going to be number one.
As you look at, as you compare what's going on in China, what's been going on in the United States, and there is not enormous transparency because each company has its own strategy and we don't always understand what's being done. How do you see the potential positive development of AI? And I think that's a question for Melanie, for Emmanuel, for Audrey, for Kori.
How do you see the positive developments that can happen? And in the case of Emmanuel, I would add, how do you see Europe playing its part there? Because Europe has, in a way, missed part of the first data revolution, all the big tech giants are either American or Chinese.
But Europe is sitting on a trove of quality data, which for a number of users of AI have great value. So there's a question there also on how countries that are developing AI can link up together for the positive. Who wants to start first?
Audrey Tang:
May I come in with this point about a social license? I think what we have just heard is that AI gives people situational awareness, common knowledge. But what does not take away is the ability to make judgment together.
So in Taiwan, we have been using AI systems for more than a decade now for civic decision making. So one quick example. A couple of years ago, in social media in Taiwan, if you scroll, you always see celebrity like Jensen Huang trying to give you cryptocurrency or investment advice.
And it happens in Beijing-controlled and US-controlled social media platforms, all of them. And so instead of doing censorship, because we have the freest internet in Asia, we cannot do that. As minister, I sent 200,000 text messages to random people around Taiwan, asking for a lotacracy to decide the social license to operate for social media ad companies.
And thousands volunteered, and we chose 447 people randomly to mirror the entire population. So in a table of 10, they're assisted by civic AI. And I should say, because we know what we're doing, it's just for making transcriptions, summarizations, social translation, chess clock with manners, basically encouraging quiet people to speak up.
It's better than frontier model at doing such things. So the small model that we built for my dad, if we know what we're doing, it's performed better than frontier models. And so with the help of such civic AI systems, each table of 10 virtually came up with their own idea that leaves everybody slightly happier and nobody very unhappy.
One table say, let's just label all ads on social media as probably skin, like a cigarette label. If Jensen Huang or somebody signed it to own it, of course you take it down, but otherwise it's labeled. Another table say an unlabeled ad, if you push to people who didn't subscribe and they lost 7 million, well, the social media company should be liable for the full 7 million damage, joint liability, another good idea.
Another say foreign companies ignore our liability rules. Every day they ignore us, slow down connection to their short video by 1%, another very good idea. And so by the end of the day, we put it to vote.
And had it been 44 human facilitators, they will need a few days to come to this uncommon ground. But again, using small language models, we will fit immediately and people voted and more than 85% agree the other 15 can live with it. So that was two years ago and it became law in just a couple of months.
So throughout last year, there's just almost no defect ads anymore. No impersonation, according to Reuters, is down by more than 94%. So it's all but solved in Taiwan.
So my point being, instead of requiring each person to kind of earn a driver license to use social media, or to, I don't know, abandon or whatever, we instead have the entire nation come together and draw the social license to operate. So that no matter which AI recommendation system come from, we apply this communal rule to them. So again, it's tamed, not as a extractive oil rig, but rather regenerated as soil.
So I think that's a very clear use of AI for good. And again, it also helps us to train our so-called trustworthy AI dialogue engine, which is a sovereign model of Taiwan. Again, it's just a layer that you can apply on existing open models like Gemma or Nemetron, which by the way, it's now both faster than DeepSeek for flash and also better.
Jean-Marie Guéhenno:
Any words on that? Any comment on Audrey, on all your views? I mean, as you monitor technology, when comparing US and China, do you see, apart from strengthening the power of the state, do you see developments comparable to what Audrey described for Taiwan?
Melanie Hart:
Yeah, so first of all, I'm thrilled to hear that example. I think the US needs to learn from Taiwan on this. You know, the US is really behind Europe on data protections, for example, and I think really behind Taiwan on helping people manage propaganda flooding from all kinds of actors.
Just two comments. First, I think I agree with Emmanuel, his earlier comment on transparency behind these models is going to be really important, because you cannot trust a model's decision-making if you aren't sure what are the factors it's using to make the decision. And if folks, I highly recommend the book Empire of AI, which really gets into some of the details about the kind of human decisions that go into some of the model design and what some of the challenges are there.
The US needs to do a better job still on transparency and having open, clear models so that people can trust them more. And just an example, again, I live in Virginia, and Virginia is one of the states here in the United States that are having huge political battles on how to draw our districts. So if you have one state, where do you draw the lines to decide what parts of the state vote together to elect a senator or to elect a representative?
And everyone's fighting over doing crazy lines and crazy shapes to try to get, how do we get the most Republicans or how do we get the most Democrats? As a Virginian, I would love a neutral AI version that would just say, okay, here's a clear kind of neutral option that everyone can look at, but that's really hard to do if you don't know the assumptions that went into it. And then just an example on AI for good, I'm doing a lot of work in the biotech space, and a lot of our new medicines, medicines that are being discovered to treat diseases that we previously didn't have treatments for are coming from mapping of the genome of not only humans, but other species as well.
So here in the United States, a lot of people are talking about GLP-1 medication, that emerged from mapping the genome of the gila monster, which is a kind of lizard. And there are just so many species of fish and mammals and amphibians and creatures on the planet, and we haven't fully mapped all of those genomes, and AI-empowered science is helping us to do that much faster. So we now increasingly have the capabilities to map genomes and then test how that might can translate into medicines in human bodies at a speed and scale that we never had before.
And China has companies like Xtalpi that are building huge factories that are nothing but robots doing biological testing to get data points to feed into an AI model that's doing this mapping to try to come up with novel disease. And as American, if I have a rare disease or a hard to treat disease, and a Chinese company uses these models to come up for medicine, to create a medicine that works for me, that I can't get anywhere else, to me that's great, that's a net positive. I will definitely want access to that.
So I think the more we are empowering science around the planet to come up with new information faster, that's a great thing.
Jean-Marie Guéhenno:
Emmanuel and Kori, the last word before we open to questions.
Emmanuel Bacry:
Let me rebounce on the rare diseases. Rare diseases are interesting because we are, I mean, we were several to say that we need local systems, because local systems are potentially more neutral than, I don't know if I believe in a neutral AI, I don't think that will ever exist as a neutral human being, then they don't exist, okay, I'm not saying. But okay, we need local models to limit the bias and so on.
Now with the rare disease, we have a problem because if you just use the local data, you don't have enough data. So you have to manage in some way. So data becomes very important, and interoperability, quality of data becomes a major issue.
I do think, as I said this morning, that when we talk about AI revolution, AI revolution is before all a data revolution first. And it's not easy to operate the data revolution, meaning setting up quality data with agile governance, but with privacy, if it's personal data, and interoperability, and so on. It is extremely hard to operate.
So there is a unique initiative in Europe, I don't know if it will work to tell you the truth, but clearly it's a unique initiative on health data. I don't know if you heard about the European health data space. So it's a new regulation that has been voted at the same time as the AI Act, and that tries to homogenize all the rules about governance, access, quality of data.
And so today, I mean in a few years, every data producers, health data producers will be required to share the data for public interest research. Otherwise there will be fines to those who don't want to do that. So we are talking about private data producers and public data producers.
It's a major change, really a major change. I'm at one of these, I'm working at one of these national platforms called Health Data Hub, which is the national French health data platform. We play a major role in this thing.
It's a unique, I think it's really a unique initiative. I hope it will succeed because this data revolution was not solved by any country. You could have companies that are very good at data and so on, but at the national level, nobody solved this problem of data, from what I know.
Okay, maybe I don't know some of the countries. And what is interesting is that since the publication of this regulation, now a lot of countries are looking at Europe, saying, ha ha, it is interesting, there is a new model for health data. Let's look at it because maybe there is something to take.
And now before this regulation, I was going all over the world and trying to set a partnership and now people come to me. So Audrey, actually the Ministry of Health, they came, 28 people came to France to visit us and we are going to sign a partnership. And we do that with Japan, we do that with South Korea, with Israel, with the US also, with Quebec, with so many countries, more and more countries.
And just to mention one pretty original maybe partnership, and it's always about data. The bottleneck for health, for operating AI on health, like partnerships, international partnerships, the main bottleneck, but like way beyond everything else, is the problem of data and finding data that are matching and that you can make them work together. Once you have that, the rest is easy.
It's easy apart from the fact that if you are working transnationally, you have governance problems and you have access problems. So what we are doing, I think it's a nice project, I like very much, with India, and it was announced by Modi and Macron in the last two summits in New Delhi. India has developed an architecture for a securitized space that is so well securized for privacy purposes that you can put it anywhere and they are very confident about the fact that privacy is preserved.
And as a proof of concept, it's open source, it's called DIPA, Data Empowerment Protection Architecture. And as a proof of concept, what they are telling to us, to France, they said, OK, if you take it, if you build such an infrastructure, such a secure space, well, we would be ready to send to you any health personal data from Indian people. So we are building this thing and we are going to run five use cases.
So we are going to take personal health data from India, we are going to bring them in France, put together some French data and doing AI on that. So that's one. I think that's a very important project because around data, governance is a major issue, major issue.
Kori Schake:
I got nothing to add, Jean-Marie.
Jean-Marie Guéhenno:
OK, maybe we can open to questions because we can't hear anything.
Graziene de Souza
I don't know if we're Thank you very much for the excellent presentation. It's very interesting. I would like to know your opinion on the whole process since the supply chain to really sustain AI and all the technologies involved in the process.
So how that has impacted job politics across the globe. For example, we see now different alliances being made on the rare minerals and how the power base now is defined based on that. When you look at China as the main supply chain that regard Brazil, DRC, Ukraine.
So how do you see that influencing the structure that supports AI per se in the database at all?
Jean-Marie Guéhenno:
Yes, go ahead.
Datcha Byangoy:
Thank you very much. Great panel. I'm much more confused than I was this morning because I'm so new to this.
So apologies if my questions don't make sense. And I apologize, I have two questions rather than one. The first one would be beyond access, I'd be interested to understand how AI is deepening inequalities amongst countries.
It'd be really interesting to hear from you on that front. And the second question was great to hear from Audrey on local language models, which seems to be a great example of how countries should be approaching this question. But I was wondering whether there is such a thing as hierarchy of AI models, whether they come from a small country versus a big power.
And then how then those local languages inform global frameworks around particularly governance and ethical considerations. Thank you.
Raphaël Ollivier-Mrejen
Thank you very much to the panelists. You made clear that governance and regulation are key. My question is about the prospects of such a regulation to ensure transparency of the of the underlying assumptions.
Do you think it is feasible? And should it be done at the national, regional or international level? Thank you.
Kawtar Zerouali:
Yeah, my question is a little different. Just to give you some context, most of us here around the room, either work for the UN, or they are diplomats that help in the UN to stay relevant. The UN is trying to be updated when it comes to the AI.
It customizes an AI tool. It's providing training to its employee to kind of catch up with the world. It started recently recruiting new positions as chief digital officers.
So in where you sit and in your work, if you are to be recruited as an advisor to the UN, how will you advise the UN to stay relevant and helpful in this sphere? So I know that a lot they see the UN as an enabler to access data, given how the UN works and with whom it works. But I'm interested to hear it from your own perspective.
Thank you.
Jean-Marie Guéhenno:
The panel. So you pick the questions that you feel most comfortable.
Kori Schake:
Sorry. Okay. I think I actually got a pass.
I don't think any of those questions were directed at me. So I will give my time to my co-panelists.
Jean-Marie Guéhenno:
Okay, who wants to start?
Emmanuel Bacry:
I can comment on a few questions. And maybe I'm not going to answer all the questions, because some I don't know how to answer. But I would like to comment on the first one.
I'm not going to give an answer. I'm sure Melanie or Audrey will have a better, will have an answer. I don't have an answer.
But I just want to warn, to do one warning, because there is something that I see too often, a mistake that is done when we're talking about rare minerals and the whole cost of the supply chain of the AI. We have to be careful today. I mean, that's at least when we're talking about what's happening today, we have to be careful to differentiate the impact of AI and the impact of digital transformation.
Because the fact that we have a phone, this is digital transformation. It is not an impact on AI. So we have the fact that we have cars with electric batteries.
This is not AI. This is digital. And this has a huge impact on ecology and energy and so on and climate.
It has the digital transition. Today, the proportion of this digital transition, the proportion of AI in this digital transition impact is very small. It's a small part.
It is increasing exponentially. So I'm not saying that it's OK and we should not care about that. But I think it is important to say so because we mix everything.
Today, when we talk about artificial intelligence, it's such a buzzword that you include everything in it. And when you do some streaming of a video, this is not AI. The AI is just for accommodation system.
But after that, and you need clouds and stuff like that for that. So this is just a remark on the first question. And for the second question, I just want about the inequalities between the countries.
Yeah, the data is very, very important for that. I mean, as I said this morning, the fear is that we are building basically a Western AI and the most terrible thing that would be to just deploy them everywhere in the world. That would be really, really bad.
And it would generate a lot of inequalities and so on. So we have to be very careful. And of course, there are inequalities between the countries, because even in terms of data, because some countries, a lot of countries haven't started collecting data.
And I think this is the main asset. And I help a French agency, public agency in designing deployment of AI in sub-Saharan African countries, and helping them making their AI. So the main bottleneck, again, is data.
I always say that in Europe, the day where we would have high quality data sets, we would have a huge impact on public health without AI. There are so many things to be done without AI. Then with AI that are on the shelves, ready to be used.
Then with the latest models of AI. But this is very, I think, important. And very, very quickly for the last question, the UN question.
You understood I'm a data person, so I would advise UN to focus on the data, access, government and stuff like that. I think that's a major. And I know I would have some colleagues at the UN doing the regular lobbying for other matters.
So I would focus on that. Thank you.
Audrey Tang:
Maybe I can come in a little bit. Thank you for the great questions. I would like to tackle also the inequality.
As Emmanuel said, it's mostly about the direction of value. If data flow out of communities, then value accrue to whoever owns the model. And what comes back is just rented so-called intelligence that you cannot correct.
So again, the data as oil direction, just make people closest to the harm, make it having less standing. People would not have any way to fix it when it goes wrong. And so I think this circular economy model, this data as soil model, is not just good to have.
It is actually a must if you are in such a community suffering from epistemic inequality. Like my grandma speaks Taiji, Taiwanese Holo, and not Mandarin as her first language. And if she uses a cutting-edge frontier LLM, that's not specifically tuned for her Taiji.
She is, of course, at a disadvantage. So in Taiwan, when we train so-called sovereign AI models, because we have 20 national languages, 16 indigenous nations, 42 language variations, the sovereignty is actually communal. So each community may be working with Mozilla, Common Voice, Data Collective, and so on, do their own curation.
And it is actually about co-authorship because each language carry a way to dissolve agreements, also to understand outside world. And they own that data. And again, using local knowledge, artifact management, intelligence, which I call calming, they can negotiate and even do social translation across different epistemic standings.
And we're already seeing other countries adopting this social translation because polarization is a big problem everywhere. And social translation to convey, for example, climate justice views and biblical creation care views can de-escalate their conflict and then agree on the uncommon ground. So I'm very happy to also report that this kind of waging peace work, which is a core part of UN, is now taking place in many places.
You can look up Democracy R&D for such a network. So that is what I will recommend UN to look into. And it can also resolve the issue about statewide or national regulations about the social license to operate of AI, including transparency rules.
In fact, if you are a California citizen, you have three more days to go to engaged.ca.gov, Engage California, where we're using this Taiwan-inspired model to ask every Californian about their work, how it's impacted by AI. So it could be apprenticeship, belonging, community, dignity, ethics, privacy, anything of those ABCDE issues. And again, using such social translation, the hope is that we can find this uncommon ground that is cross-partisan, cross-cultural, and then the state of California can simply say, as we did to the deepfakes social media, to the Frontier Labs, this is not the governor's idea or the MP's idea.
This is not partisan. Everyone in California thinks you should do at least this. Lee?
Melanie Hart:
Yeah. So just a few comments. On the supply chain question, the United States and China are currently the AI leaders, depending on how you look at the metrics.
I'll say that they're the two AI leaders. And it will be interesting in that there are areas where we can collaborate and areas where we just can't. So on critical minerals, on rare earths, China has made very clear that they will cut us off from those supplies as a way to control U.S. decision making. So you will see a growing decoupling between the U.S. and China in critical minerals. We just have to have alternative supply. We cannot depend on China for that.
And when it comes to the higher end chips, there's very clear evidence that the Chinese military is trying to get ahold of those to run AI in ways that in our view undermines our national security. And we have to have export controls there. What I find really interesting is that there's also areas where we can collaborate, but doing so safely is going to require new mechanisms.
And I agree totally with Emmanuel that on data, cross-border movement of data in a way that is safe and secure and benefits both sides equally is really going to be the big key to unlock what we can actually do with AI. And there's tremendous potential there with China, but also tremendous risk. So I was really interested to hear about the French India framework.
I think maybe that could be a model for the U.S. and Europe and China and others, right? Because we will have to have a kind of independent standards and mechanism for how we can exchange data. It can't sit in any one country.
I'm doubtful that the UN could be the solution, but hey, welcome some ideas on that front. On the question about AI increasing inequality between nations, I totally agree with Kori's comment. This is going to be the biggest upheaval since the industrial revolution, and we still don't fully understand it.
But I think the economic losses where countries are not deploying AI are going to be real and serious. And just to give an example, last fall we went to a car and automobile manufacturing factory in China, the Xiaomi factory that uses AI plus robotics to make one electric vehicle in 76 seconds. And next week, we'll go to California to visit a Ford factory that's using similar AI and robotic innovation to make a Ford truck rapidly and at a price comparable with what the Chinese companies are reaching.
So you see the U.S. and China in this race. How can we use the in-factory AI models with robotics to manufacture higher quality faster with fewer people? If you're not deploying the same models, you're not going to be able to compete.
You won't be able to compete on scale, on cost, or on quality. What does that mean for Mexico or for another country that's an automobile manufacturer? You may wind up with just two different worlds, and you can't be in that higher level of competition without the most advanced AI and robotics, which are expensive.
So I think a lot of countries, including the United States, including China, are going to have to figure out what's our new economic plan, because some of what we were doing before may not work going forward. And then just finally to the question about what can governments do to push models to be more transparent, looking at the U.S. political system and sitting here in Washington, I don't think Washington can do this effectively, because our political leaders are always so far behind the companies in understanding what the AI is doing, much less trying to regulate the models. I think a lot of this, at least in the U.S. system and probably globally as well, will be based on market choice. So companies and citizens are going to want the more transparent, the cheaper, the simpler AI models that you get, kind of how both Audrey and Emmanuel talked about sometimes what you get back is garbage, right? I get a lot of garbage back when I experiment with U.S. language models, for example. So I think the best thing that government can do is to make sure citizens have wide access and choice, models from the U.S., models from China, models from Europe, local ones, big ones that house their data in a data center. I think if consumers and companies have a lot of choice, then people will gravitate to the ones that work best for them. And I just want to point out that not that many months ago, OpenClaw took the world by storm, and OpenClaw was an innovation from Austria, I think, right? So big, big global models can come from anywhere as long as consumers have choice.
And if they have choice, then transparency is ultimately going to win out at the end of the day.
Jean-Marie Guéhenno:
On this optimistic and positive note, I think that's a good ending. And I want to thank all our panelists, Kori, who had to leave, Melanie, Audrey, and Emmanuel. I think it was a very good discussion.
For me, the key takeaway is that, in a way, data are the oil of the future, and that every country sits on a trove of data which is valuable if it makes those data reliable, if it makes them accessible, interoperable, if it protects privacy. So all sorts of ifs that require a lot of thinking and efforts. But in a way, we all produce data around the world, rich and poor countries.
If we manage to produce them well, then there's an enormous field for cooperation and enormous potential for progress. And I think that's a pretty encouraging thought at the moment when we feel that there are only a few big companies that are going to control us through the data. I mean, the image of Audrey, I mean, is it extractive or is it not extractive?
And if we move from an extractive economy to a non-extractive economy, there's real hope. So thank you all. Thank you very much.
Melanie Hart:
Thank you.
Okay, we are about to start and I want to thank our panelists. I think we should be joined by Emmanuel Bacry in a few minutes, but we'll start anyway now. This morning, we heard about the substance of AI, we heard about what's going on backstage, all the infrastructure of the data economy, discussing data centers, submarine cables, satellites, and all that is necessary for this connected world to work.
In this panel, we really want to draw some conclusions or at least sharpen our questions on the geopolitical dimension of what's going on. For that, we have four panelists who each bring a lot to the discussion. I will start with Kori Schake, who I have known for many years.
Delighted to see you again, Kori. She works at the American Enterprise Institute and has been working on security for many years. When we first met, when the security was not seen as it is seen now, it's really profoundly different.
I want to ask you, Kori, how disruptive is AI for the international security architecture? Is it going to consolidate the dominance of a few AI superpowers, or is it going to disseminate power in an unpredictable way for the good and for the bad?
Kori Schake:
It's too soon to tell, because we're at the early stage of what the people who understand the technology best are telling us is going to be a bigger economic and social upheaval than the industrial revolution. The industrial revolution took about 100 years before the social and political and economic effects were really fully evident. It's too soon to tell.
It's also to be on with Audrey Tang, who I admire enormously and would defer to her answers on these things. But my sense is that there's a two-stage battle going on. One line of effort is the countries that are pushing the frontier of the development of the technology.
And there, it does really matter who's first, because it looks like a first-mover advantage will drive the frontier forward in potentially different directions and will have enormous national security consequences. I mean, even if you just look at the way artificial intelligence is changing military operations and intelligence practices by allowing the assessment of enormous pools of data, so making things more transparent, easier to target. But you can also see that that doesn't add up to winning wars.
So it's early on in the development. But the second track, I think, which goes to the latter part of your question, Jean-Marie, is that it's not clear the first-mover advantage of people at the frontier are going to be how artificial intelligence is adopted across economies, across states, across political communities. And there, it looks to me that the first-mover advantage might actually be detrimental because lesser models can use the data pools that the frontier organizations have developed and create solutions to problems that are plenty good for what is needed in that space.
So I think it's too soon to tell, but there look to be enormous advantages on the frontier for being a technological leader. And it's not clear that those advantages will pertain across broad-scale adoption.
Jean-Marie Guéhenno:
So we are going to continue in a way with the same question, but focusing on China and the United States. Melanie Hart, you are focusing on that tech competition actually goes beyond AI between China and the United States. You're regularly going to China, checking on the state of play.
How do you see that competition? How do you see China leverage its technical and technological prowess into geopolitical advantage or not?
Melanie Hart:
That's a really great question. And there are three ways that a lot of people in Washington and elsewhere are looking at how to measure who's ahead in AI capability. One, Kori mentioned frontier models.
Who's doing the things that no one has done before with AI? So, ChatGPT has had moments where they've done that, for example. DeepSeek has also had moments where they've done that out of China.
But overall, when it comes to frontier models, who's first? The US tends to be first out of the gate. And I think we're looking at that too much.
I agree with Kori, if I understood you correctly. There's too much focus on that in the United States sometimes. Thinking about whoever spends the billions and billions of dollars to do the first model before anyone else will maintain leadership is not necessarily the case.
The second metric to look at and where people are is compute. How much data can you crunch to power these models? And some of that is to do with chips.
The United States and its allies have the lead on the most advanced AI chips. China's about five to seven years behind the US and its allies on the advanced AI chips. And we're hearing from Chinese companies and the Chinese military that that is constraining what they can do with some of the AI deployment.
But another aspect of compute is electricity and data center infrastructure. And that's where we're really starting to run into barriers in the United States. I live in Virginia where our electricity bills have shot up because there are, I think, something like 1,500 new data centers being built in Virginia.
And now American citizens are starting to protest them because there's water pollution. They're raising our electricity bills. People don't want to live next to a data center.
So I think we're going to see some shift in that balance of power and compute because China can roll out nuclear projects much faster than we can. They don't really care if you want to live next to a data center or not. In China, the data center will go wherever Beijing wants to go.
So compute is the second. But the third, I think, is going to be the most important, and that's deployment and distribution. Who uses the AI that you come up with in a lab?
And that's where China is really focusing its efforts. And this is where China has the opportunity to really shape the way that AI is deployed globally. Here in the United States, you know, Chad GPT and Claude and these different models are at the frontier in what they can do, but they're also relatively expensive.
What China's doing is deploying deep-seek and finding ways to do an almost as good capability, but a lot cheaper. You don't need as good chips. You don't need as much electric power.
They're doing open-weight, locally deployed models. That means you can just take it and install it on your own computer and tinker with it and change the way that the model works. You can own it a little bit more, whereas with Chad GPT, it's kind of like doing a Google search.
Everything lives within OpenAI and Chad GPT. China's experimenting with smaller, cheaper, easier-to-deploy solutions that are probably going to be more like the kind of solution that Global South nations in particular are going to want, because they're going to want to be able to have their data in-country or in-company and not give it to OpenAI. They're going to want to know, how can I deploy AI at the lowest cost with the easiest, lowest-end chips, without consuming too much electric power?
Here in the United States, a lot of leading American universities can't get access to the most advanced chips to experiment with in the lab, because OpenAI and some of these other companies are buying them all up. Deployment is probably going to be most critical for who shapes the global landscape. That's an area where right now I think Washington is paying too much attention to, let's be ahead on frontier.
Let's be first, as Kori put it. Let's maintain our chip edge. Both of those are important, but if nobody buys what you're selling, or nobody can afford to buy what you're selling, then that's where we really lose the game to China.
Then just four areas that I'll be watching really closely in terms of US-China competition. The first one is China likes to turn a cost advantage into dependencies. We've seen with 5G telecom that Huawei was always the cheapest.
Once you bought Huawei 2G, you could only go for Huawei for 3G. You could only go for Huawei for 4G. They aren't compatible with other systems, so they tend to lock in their customers to the Chinese ecosystem.
There's a lot of risk there with AI that maybe you go for DeepSeek or another China alternative because it's cheap today. Then you find, well, the only chips you can use are Huawei chips. The only data center you can use is a Chinese data center.
Then that opens you up for risk. A second factor that I'm really watching is information shaping. My team here at the Atlantic Council did the same search in DeepSeek, China's AI model, and in Western AI models, and the information is just different.
The Chinese government shapes what you hear and know about the world. I'm sure, Audrey is more aware than I, you're definitely going to get different information about Taiwan, for example, but also about human rights and how well the Chinese government performs. There is a big risk if the entire planet is getting their AI search information from DeepSeek because when it came to general basic internet search, China's alternative to Google was Baidu.
It's horrible. Nobody ever wanted to use Baidu because it is such a mess. It is so difficult to use.
We were in China a couple months ago and I tried to use Baidu just to get a restaurant address and it was giving me search results dated in 2003 for no clear reason whatsoever. It's just very messy. China never took off on internet search, but it could on AI-powered internet search.
That would really shape the information that the global public has access to. Then just one other point, China, we really are watching closely. How is China working to export AI surveillance models?
We know that in previous years, there have been issues with China exporting surveillance networks that would be, for example, a network of cameras that you can deploy around a city. Now they're looking at new models that can predict in advance what kind of person, who is going to be a possible dissident in the future, who's going to be a political agitator in the future, using AI models to look at who are their friends, to listen to their telefilm conversations, to review their emails, to use geolocation and see where do they go around the city, what internet websites did they look at.
China is experimenting with AI for a forward looking surveillance program, not just tracking what people do now or in the past, but what kind of person are you and should we be worried about you in the future. I have huge concerns about that being exported to other countries as well. That's it for me.
Jean-Marie Guéhenno:
I'm going to soon turn to Audrey, but before I want to ask Emmanuel a couple of questions because it happens that this morning we already discussed briefly, as he was presenting, when the various types of AI, we discussed briefly DeepSeek. The question is really, do you, I mean Emmanuel is a AI scientist, and do you really believe that there are shortcuts in a way to using much less data and being much cheaper? I mean, how do you see the competition from your standpoint as a...
Emmanuel Bacry:
Yeah, before answering strictly speaking the question, I would like to comment on some of the things that were said. First of all, I think we are focusing a little too much on LLMs and chatbots. AI is not just that.
So we don't have to forget. I mean, the whole buzz in the whole world is about that, but we should be careful not just, you know, being part of this buzz. It is clearly a big thing.
I'm not saying, I'm not trying to minimize it. And it's a big, I mean, with plenty of applications and it will transform quite a lot of things. But AI is also a lot of other stuff.
So when we are saying about the fact that we need a huge amount of data, this is really about LLMs, a huge amount of compute, it's also about LLM. So we have to be careful a little bit about that and talk also about other AIs that could have huge impacts. I mean, if we talk about health, I know health very well, there are huge impacts on health without LLMs. That's the first thing I wanted to say. Second thing, maybe it's not politically correct what I'm going to say, but what you are saying, Melanie, about China, the fact that, I think it's you who said that, I'm sorry, maybe it's not you, but when the information shaping, the search information, there is one thing that I like in the, I'm being a little bit provocative, OK, but there is one thing that I like in the Chinese legislation and regulation, which is that it says clearly that the alignment of the foundational models should be done according to the Chinese Communist Party. But it is clear, at least, open AI, we don't know anything about the way they align the model.
And OK, you are saying that when we are doing search information on deep-seek, it is clear that there are plenty of problems and so on. But because we know China very well, but it is totally biased what we get with ChatGPT, with Claude, with Gemini and so on, that it is more subtle and nobody knows exactly how it was trained. I don't know which one is the most dangerous.
It is not clear for me. It is an open question. I don't have the answer, but I think both are kind of dangerous and not only the one which is openly biased and we know very well the bias.
So I think we should be also careful about that. And so now I come back to the, and yeah, sorry, one more point. I think one of the very big challenges is also, and you mentioned, I don't remember who, deployment.
In deployment, there is one challenge which is extremely important and very hard, is the evaluation. The evaluation, there are plenty of challenges about evaluation, not even talking about LLMs. On LLMs, it is a nightmare to evaluate. But even if we talk about regular AI, evaluation in real life is a very big problem.
And again, please, I will take an example in health because a lot of things are very clear in health because it is the center of so many problems and so many touches everybody. So when we deploy a medical device that has AI in a hospital, of course it is certified. It is FDA cleared or in Europe it is certified and every country has this certification.
But certification is kind of a still ideal world where you have a closed space and you test specific things. When you deploy in real life, when you deploy in a chain of process where human beings interact with the AI, everything can go wrong, even if the AI is very good. But you have also to evaluate the whole chain of process that will change because you change the one brick, one step by an AI model.
And there are examples in hospitals where some AI were deployed, very good level AI, and where the number of deaths increased dramatically. So there are many challenges about that. Now, sorry, I come to your question now, Jean-Marie.
So your question, sorry, it was about the fact that do we need a lot of data or can we have... Yes, yes. So, yeah, as I said this morning, but nobody is sure.
So maybe some people in this roundtable would say, but it is likely that DeepSeek and Quen used what we call distillation of the big American models. So, of course, they didn't need as much data and it costs less money because they basically, they used the big models to take all the knowledge of these models. Now, would it have been possible really to do a DeepSeek or Quen if ChatGPT, Claude and Gemini didn't exist?
It is not that clear. I don't have a clear-cut answer about that. But I want to...
And about the quantity of data, this is necessary for LLMs. This is very clear. It is necessary for LLM. But not at all, not in the same size for many other AIs.
Melanie Hart:
Can I ask a question? I think this is an important point. So if you go to the DeepSeek website, for example, and look into their pricing, it is fascinating.
They describe their pricing in direct relation to OpenAI. And they say, like other models who require a monthly subscription minimum, we only charge by the search or by the token. So OpenAI...
Am I getting to the heart of this follow-up question correctly? OpenAI is cheaper for users to run. So, for example, a company based in South Africa or Colombia or anywhere in the world, if they're comparing some enterprise options between DeepSeek and Claude and OpenAI, everything that I'm seeing is that the DeepSeek version is going to be currently much cheaper for them.
And I think that's something that we need to fix.
Emmanuel Bacry:
Yeah, but that is different from the question of Jean-Marie, which was the training with the quantity of data. But I totally agree that it seems like they have found a way of running it in a cheaper way. And when you look at the researchers in AI today, everybody is looking for models that are simpler and that do as efficient LLMs as what we have today, but in a simpler manner so that the energy needed to run would be much less.
And there was one announcement, I mean, almost every month there is such an announcement of a big improvement about that.
Jean-Marie Guéhenno:
I want to turn to Audrey Tang. When you've been the first digital minister in Taiwan, you are the Taiwan cyber ambassador, and you've been working a lot on the positive views of AI. And as Emmanuel reminded us, I mean, AI, large language model are just one compartment of AI.
There are many uses of AI. And so if one looks beyond the US-China competition and looking at AI and not focusing just on the security side of things, what do you see as the upside opportunities created by AI in connecting citizens, in increasing capacity to solve societal problems? What is the potential?
And I think it's important for this group of diplomats. I mean, they come from rich countries, from developing countries, from really all over the world. And what is the potential of AI to redistribute power for good, possibly for bad?
Audrey Tang:
Certainly. So I want to continue the thoughts that we heard about evaluation, actually, because I'm now also academically in Oxford, working on AI alignment, but not aligning our outcome utility, but rather aligning with a participatory process. So I want to tell a story about how we aligned AI systems with the participatory process.
And it happens in my home, actually. So my dad is a political scientist, journalist, who was in Tiananmen, actually, until 1st of June in 1989, so covered the whole thing. And a few months ago, he has a health issue.
And so he found that no matter which chatbot he talks to, the chatbot keeps suggesting fantastic ideas, yours, brainstorming, and so on. And he gets lured into this kind of spiral now situation. But because he's a journalist, he immediately diagnosed that it is just trying to earn the subscription and not loyal to the family relationship.
So my point here is that we very quickly found a solution to his dilemma. So with my parents' explicit consent, we built a local model that's entirely running on a Mac. And so it doesn't need a data center.
And it is a language model plus a few other narrow models, but everything fits into one Mac computer. And it only joins our family signal group. And of course, he can directly message the bot.
But the evaluation metric was done with my mom, who said that every time the bot speaks, it should be reducing my dad's dependency on the screen and should restore his peace of mind to the reality around them. So it's a little bit like how in Taiwan, we make sure that the kids in classrooms use only large screens, one laptop or a tablet per pair of children. So it's always pro-social, not anti-social.
And the bot, every time it gives a slightly isolating answer, we can just use what's called directional steering, which is a very small amount of data, just 10 sentences. And within one minute, we can train the network to behave in the direction we're tuning. And because it's running entirely locally, it's not sending the data anywhere to the cloud, not to Beijing, not to Silicon Valley.
So my point here is really a governance point, which is, are we thinking about AI on top or AI on tap? If it's AI on tap, then the people touched by AI can't steer it in real time. But if it's AI on top, then the extractive relationship, whether in Beijing or Silicon Valley, we are like planktons, our human data being extracted into this huge oil rig.
But if it's within the local community, then people can hand it together. This is why I call data soil, not the data oil. It's purely regenerative.
And by the way, it's faster than the cloud models and higher quality. So I really don't see the reason why not to use this kind of local models.
Jean-Marie Guéhenno:
On what Audrey just said, I saw Emmanuel, you were... Sorry?
Emmanuel Bacry:
I thought you had a reaction, No, no, positive reaction. I agree. And I do believe in small models more than the big GPT models, where nobody sees a business model in these big models.
Everybody's losing money with these models. And they are too generic. They are supposed to know everything about everything, which is stupid.
It will never reach this goal. This is very clear. So they keep, you know, they keep adding layers and layers on top of it so that they repair something that they see and they repair something else.
But we should totally change by design these things. And so I do believe absolutely in much smaller models that you train on specific data, local data, specific to a small population and with a purpose, a specific purpose, trying to answer specific questions. This is clearly has a lot of applications and a lot of value.
So I totally agree with Audrey.
Jean-Marie Guéhenno:
Thank you. I want to come back to Kori. You said at the beginning that it's much too early to answer the kind of existential question that I asked.
But as you see it today, not projecting in the future, do you see AI being in the existing conflict, in the existing battles that we see? Do you see AI playing, I mean, rebalancing situation between actors? We have the Ukraine war.
We have the Iran, Israel, US war. So these are war that are existing, that are happening today with extensive use of artificial intelligence and data in both cases. How do you see the impact in those existing conflicts?
Kori Schake:
So I think artificial intelligence is changing some of the trade craft of national security because it's harder for spies, for example, to have unsupervised engagements, right? It's increasing the transparency both on the battlefield and its increasing search capability. So you can, you know, if an intelligence agency records every foreign telephone call, the constraint isn't on the recording, it's on the processing of information, people to listen to it and identify where there are patterns.
And artificial intelligence is genuinely brilliant at that work. Just to take one small example, I was on the historian advisory committee to the State Department and we did a test run on whether using artificial intelligence to evaluate freedom of information document releases could assist. Would it increase the need for human supervision?
Would it reduce? And what we found is that it takes away the majority of cases which are easy to solve by a decision rule. And that freed up the actual talented people to focus only on the hard cases.
And so it didn't so much reduce the need for human labor as focus the need for human labor on the most intellectually challenging elements of the problem. So that's one, I think, really valuable test case. And as Emmanuel said, I can see a thousand applications in the healthcare sphere for that, identifying new vaccines or new treatments or solving difficult problems that a decision rule can solve for you, but it can't substitute for judgment.
And so in warfare, identifying where things are on a battlefield, both for situational awareness and for targeting is an enormous advantage. And if those targets are targets that don't require human judgment, right? Every tank moving on a battlefield, every enemy tank moving on a battlefield should be targeted.
That doesn't require human judgment unless you are worried about your munitions stockpiles, in which case you can also put in an algorithm to prioritize certain types of targets or locations. But what it cannot do for you is say, should I do this? That's a fundamentally human decision-making.
And in warfare, that's really important decision-making. Back in the 1990s, the Commandant of the Marine Corps, General Charles Krulak talked a lot about what he called the strategic corporal, which is the lowest ranking soldier on the battlefield can make a choice that will have strategic implications. And there's no amount of automation that can prevent bad human choices in conflict situations and can replicate good human choices.
So what is changing, I think we are seeing is the persistent surveillance at low cost that drones either relying on a communications network or independently targeted. That has reduced the ability of mobility on the battlefield. It's reinforced the advantages of defenders over the prospects of attackers.
That's a legitimate change. The second legitimate change of what's happening as a result of artificial intelligence is it's so easy to print low cost drones. And if you don't care whether 947 of a thousand get destroyed, then you can have enormous effects.
So the cost effectiveness, it used to be, you had to be a great power to afford persistent surveillance on the battlefield. That's not true anymore. And so that's an advantage to insurgents.
It's an advantage to terrorist organizations. It's an advantage to countries of the global South in defending themselves against higher capability militaries. So I think it's changed those things, but what it hasn't changed is the grit of societies refusing to accept bad outcomes and other fundamentally human judgments on which winning and losing wars continues to rely.
Jean-Marie Guéhenno:
Thank you very much. That's actually very important, I think, for many countries around the world, what you just said. But let's now move beyond war and think about the positive views of AI.
Audrey gave us a wonderful example of a small model that produces very good results at relatively low cost. And so I'm tempted to ask the question to all of you. As we look forward to the future development of AI, I mean, Melanie, I think you said there was too much obsession, people were too obsessed as who is going to be number one.
As you look at, as you compare what's going on in China, what's been going on in the United States, and there is not enormous transparency because each company has its own strategy and we don't always understand what's being done. How do you see the potential positive development of AI? And I think that's a question for Melanie, for Emmanuel, for Audrey, for Kori.
How do you see the positive developments that can happen? And in the case of Emmanuel, I would add, how do you see Europe playing its part there? Because Europe has, in a way, missed part of the first data revolution, all the big tech giants are either American or Chinese.
But Europe is sitting on a trove of quality data, which for a number of users of AI have great value. So there's a question there also on how countries that are developing AI can link up together for the positive. Who wants to start first?
Audrey Tang:
May I come in with this point about a social license? I think what we have just heard is that AI gives people situational awareness, common knowledge. But what does not take away is the ability to make judgment together.
So in Taiwan, we have been using AI systems for more than a decade now for civic decision making. So one quick example. A couple of years ago, in social media in Taiwan, if you scroll, you always see celebrity like Jensen Huang trying to give you cryptocurrency or investment advice.
And it happens in Beijing-controlled and US-controlled social media platforms, all of them. And so instead of doing censorship, because we have the freest internet in Asia, we cannot do that. As minister, I sent 200,000 text messages to random people around Taiwan, asking for a lotacracy to decide the social license to operate for social media ad companies.
And thousands volunteered, and we chose 447 people randomly to mirror the entire population. So in a table of 10, they're assisted by civic AI. And I should say, because we know what we're doing, it's just for making transcriptions, summarizations, social translation, chess clock with manners, basically encouraging quiet people to speak up.
It's better than frontier model at doing such things. So the small model that we built for my dad, if we know what we're doing, it's performed better than frontier models. And so with the help of such civic AI systems, each table of 10 virtually came up with their own idea that leaves everybody slightly happier and nobody very unhappy.
One table say, let's just label all ads on social media as probably skin, like a cigarette label. If Jensen Huang or somebody signed it to own it, of course you take it down, but otherwise it's labeled. Another table say an unlabeled ad, if you push to people who didn't subscribe and they lost 7 million, well, the social media company should be liable for the full 7 million damage, joint liability, another good idea.
Another say foreign companies ignore our liability rules. Every day they ignore us, slow down connection to their short video by 1%, another very good idea. And so by the end of the day, we put it to vote.
And had it been 44 human facilitators, they will need a few days to come to this uncommon ground. But again, using small language models, we will fit immediately and people voted and more than 85% agree the other 15 can live with it. So that was two years ago and it became law in just a couple of months.
So throughout last year, there's just almost no defect ads anymore. No impersonation, according to Reuters, is down by more than 94%. So it's all but solved in Taiwan.
So my point being, instead of requiring each person to kind of earn a driver license to use social media, or to, I don't know, abandon or whatever, we instead have the entire nation come together and draw the social license to operate. So that no matter which AI recommendation system come from, we apply this communal rule to them. So again, it's tamed, not as a extractive oil rig, but rather regenerated as soil.
So I think that's a very clear use of AI for good. And again, it also helps us to train our so-called trustworthy AI dialogue engine, which is a sovereign model of Taiwan. Again, it's just a layer that you can apply on existing open models like Gemma or Nemetron, which by the way, it's now both faster than DeepSeek for flash and also better.
Jean-Marie Guéhenno:
Any words on that? Any comment on Audrey, on all your views? I mean, as you monitor technology, when comparing US and China, do you see, apart from strengthening the power of the state, do you see developments comparable to what Audrey described for Taiwan?
Melanie Hart:
Yeah, so first of all, I'm thrilled to hear that example. I think the US needs to learn from Taiwan on this. You know, the US is really behind Europe on data protections, for example, and I think really behind Taiwan on helping people manage propaganda flooding from all kinds of actors.
Just two comments. First, I think I agree with Emmanuel, his earlier comment on transparency behind these models is going to be really important, because you cannot trust a model's decision-making if you aren't sure what are the factors it's using to make the decision. And if folks, I highly recommend the book Empire of AI, which really gets into some of the details about the kind of human decisions that go into some of the model design and what some of the challenges are there.
The US needs to do a better job still on transparency and having open, clear models so that people can trust them more. And just an example, again, I live in Virginia, and Virginia is one of the states here in the United States that are having huge political battles on how to draw our districts. So if you have one state, where do you draw the lines to decide what parts of the state vote together to elect a senator or to elect a representative?
And everyone's fighting over doing crazy lines and crazy shapes to try to get, how do we get the most Republicans or how do we get the most Democrats? As a Virginian, I would love a neutral AI version that would just say, okay, here's a clear kind of neutral option that everyone can look at, but that's really hard to do if you don't know the assumptions that went into it. And then just an example on AI for good, I'm doing a lot of work in the biotech space, and a lot of our new medicines, medicines that are being discovered to treat diseases that we previously didn't have treatments for are coming from mapping of the genome of not only humans, but other species as well.
So here in the United States, a lot of people are talking about GLP-1 medication, that emerged from mapping the genome of the gila monster, which is a kind of lizard. And there are just so many species of fish and mammals and amphibians and creatures on the planet, and we haven't fully mapped all of those genomes, and AI-empowered science is helping us to do that much faster. So we now increasingly have the capabilities to map genomes and then test how that might can translate into medicines in human bodies at a speed and scale that we never had before.
And China has companies like Xtalpi that are building huge factories that are nothing but robots doing biological testing to get data points to feed into an AI model that's doing this mapping to try to come up with novel disease. And as American, if I have a rare disease or a hard to treat disease, and a Chinese company uses these models to come up for medicine, to create a medicine that works for me, that I can't get anywhere else, to me that's great, that's a net positive. I will definitely want access to that.
So I think the more we are empowering science around the planet to come up with new information faster, that's a great thing.
Jean-Marie Guéhenno:
Emmanuel and Kori, the last word before we open to questions.
Emmanuel Bacry:
Let me rebounce on the rare diseases. Rare diseases are interesting because we are, I mean, we were several to say that we need local systems, because local systems are potentially more neutral than, I don't know if I believe in a neutral AI, I don't think that will ever exist as a neutral human being, then they don't exist, okay, I'm not saying. But okay, we need local models to limit the bias and so on.
Now with the rare disease, we have a problem because if you just use the local data, you don't have enough data. So you have to manage in some way. So data becomes very important, and interoperability, quality of data becomes a major issue.
I do think, as I said this morning, that when we talk about AI revolution, AI revolution is before all a data revolution first. And it's not easy to operate the data revolution, meaning setting up quality data with agile governance, but with privacy, if it's personal data, and interoperability, and so on. It is extremely hard to operate.
So there is a unique initiative in Europe, I don't know if it will work to tell you the truth, but clearly it's a unique initiative on health data. I don't know if you heard about the European health data space. So it's a new regulation that has been voted at the same time as the AI Act, and that tries to homogenize all the rules about governance, access, quality of data.
And so today, I mean in a few years, every data producers, health data producers will be required to share the data for public interest research. Otherwise there will be fines to those who don't want to do that. So we are talking about private data producers and public data producers.
It's a major change, really a major change. I'm at one of these, I'm working at one of these national platforms called Health Data Hub, which is the national French health data platform. We play a major role in this thing.
It's a unique, I think it's really a unique initiative. I hope it will succeed because this data revolution was not solved by any country. You could have companies that are very good at data and so on, but at the national level, nobody solved this problem of data, from what I know.
Okay, maybe I don't know some of the countries. And what is interesting is that since the publication of this regulation, now a lot of countries are looking at Europe, saying, ha ha, it is interesting, there is a new model for health data. Let's look at it because maybe there is something to take.
And now before this regulation, I was going all over the world and trying to set a partnership and now people come to me. So Audrey, actually the Ministry of Health, they came, 28 people came to France to visit us and we are going to sign a partnership. And we do that with Japan, we do that with South Korea, with Israel, with the US also, with Quebec, with so many countries, more and more countries.
And just to mention one pretty original maybe partnership, and it's always about data. The bottleneck for health, for operating AI on health, like partnerships, international partnerships, the main bottleneck, but like way beyond everything else, is the problem of data and finding data that are matching and that you can make them work together. Once you have that, the rest is easy.
It's easy apart from the fact that if you are working transnationally, you have governance problems and you have access problems. So what we are doing, I think it's a nice project, I like very much, with India, and it was announced by Modi and Macron in the last two summits in New Delhi. India has developed an architecture for a securitized space that is so well securized for privacy purposes that you can put it anywhere and they are very confident about the fact that privacy is preserved.
And as a proof of concept, it's open source, it's called DIPA, Data Empowerment Protection Architecture. And as a proof of concept, what they are telling to us, to France, they said, OK, if you take it, if you build such an infrastructure, such a secure space, well, we would be ready to send to you any health personal data from Indian people. So we are building this thing and we are going to run five use cases.
So we are going to take personal health data from India, we are going to bring them in France, put together some French data and doing AI on that. So that's one. I think that's a very important project because around data, governance is a major issue, major issue.
Kori Schake:
I got nothing to add, Jean-Marie.
Jean-Marie Guéhenno:
OK, maybe we can open to questions because we can't hear anything.
Graziene de Souza
I don't know if we're Thank you very much for the excellent presentation. It's very interesting. I would like to know your opinion on the whole process since the supply chain to really sustain AI and all the technologies involved in the process.
So how that has impacted job politics across the globe. For example, we see now different alliances being made on the rare minerals and how the power base now is defined based on that. When you look at China as the main supply chain that regard Brazil, DRC, Ukraine.
So how do you see that influencing the structure that supports AI per se in the database at all?
Jean-Marie Guéhenno:
Yes, go ahead.
Datcha Byangoy:
Thank you very much. Great panel. I'm much more confused than I was this morning because I'm so new to this.
So apologies if my questions don't make sense. And I apologize, I have two questions rather than one. The first one would be beyond access, I'd be interested to understand how AI is deepening inequalities amongst countries.
It'd be really interesting to hear from you on that front. And the second question was great to hear from Audrey on local language models, which seems to be a great example of how countries should be approaching this question. But I was wondering whether there is such a thing as hierarchy of AI models, whether they come from a small country versus a big power.
And then how then those local languages inform global frameworks around particularly governance and ethical considerations. Thank you.
Raphaël Ollivier-Mrejen
Thank you very much to the panelists. You made clear that governance and regulation are key. My question is about the prospects of such a regulation to ensure transparency of the of the underlying assumptions.
Do you think it is feasible? And should it be done at the national, regional or international level? Thank you.
Kawtar Zerouali:
Yeah, my question is a little different. Just to give you some context, most of us here around the room, either work for the UN, or they are diplomats that help in the UN to stay relevant. The UN is trying to be updated when it comes to the AI.
It customizes an AI tool. It's providing training to its employee to kind of catch up with the world. It started recently recruiting new positions as chief digital officers.
So in where you sit and in your work, if you are to be recruited as an advisor to the UN, how will you advise the UN to stay relevant and helpful in this sphere? So I know that a lot they see the UN as an enabler to access data, given how the UN works and with whom it works. But I'm interested to hear it from your own perspective.
Thank you.
Jean-Marie Guéhenno:
The panel. So you pick the questions that you feel most comfortable.
Kori Schake:
Sorry. Okay. I think I actually got a pass.
I don't think any of those questions were directed at me. So I will give my time to my co-panelists.
Jean-Marie Guéhenno:
Okay, who wants to start?
Emmanuel Bacry:
I can comment on a few questions. And maybe I'm not going to answer all the questions, because some I don't know how to answer. But I would like to comment on the first one.
I'm not going to give an answer. I'm sure Melanie or Audrey will have a better, will have an answer. I don't have an answer.
But I just want to warn, to do one warning, because there is something that I see too often, a mistake that is done when we're talking about rare minerals and the whole cost of the supply chain of the AI. We have to be careful today. I mean, that's at least when we're talking about what's happening today, we have to be careful to differentiate the impact of AI and the impact of digital transformation.
Because the fact that we have a phone, this is digital transformation. It is not an impact on AI. So we have the fact that we have cars with electric batteries.
This is not AI. This is digital. And this has a huge impact on ecology and energy and so on and climate.
It has the digital transition. Today, the proportion of this digital transition, the proportion of AI in this digital transition impact is very small. It's a small part.
It is increasing exponentially. So I'm not saying that it's OK and we should not care about that. But I think it is important to say so because we mix everything.
Today, when we talk about artificial intelligence, it's such a buzzword that you include everything in it. And when you do some streaming of a video, this is not AI. The AI is just for accommodation system.
But after that, and you need clouds and stuff like that for that. So this is just a remark on the first question. And for the second question, I just want about the inequalities between the countries.
Yeah, the data is very, very important for that. I mean, as I said this morning, the fear is that we are building basically a Western AI and the most terrible thing that would be to just deploy them everywhere in the world. That would be really, really bad.
And it would generate a lot of inequalities and so on. So we have to be very careful. And of course, there are inequalities between the countries, because even in terms of data, because some countries, a lot of countries haven't started collecting data.
And I think this is the main asset. And I help a French agency, public agency in designing deployment of AI in sub-Saharan African countries, and helping them making their AI. So the main bottleneck, again, is data.
I always say that in Europe, the day where we would have high quality data sets, we would have a huge impact on public health without AI. There are so many things to be done without AI. Then with AI that are on the shelves, ready to be used.
Then with the latest models of AI. But this is very, I think, important. And very, very quickly for the last question, the UN question.
You understood I'm a data person, so I would advise UN to focus on the data, access, government and stuff like that. I think that's a major. And I know I would have some colleagues at the UN doing the regular lobbying for other matters.
So I would focus on that. Thank you.
Audrey Tang:
Maybe I can come in a little bit. Thank you for the great questions. I would like to tackle also the inequality.
As Emmanuel said, it's mostly about the direction of value. If data flow out of communities, then value accrue to whoever owns the model. And what comes back is just rented so-called intelligence that you cannot correct.
So again, the data as oil direction, just make people closest to the harm, make it having less standing. People would not have any way to fix it when it goes wrong. And so I think this circular economy model, this data as soil model, is not just good to have.
It is actually a must if you are in such a community suffering from epistemic inequality. Like my grandma speaks Taiji, Taiwanese Holo, and not Mandarin as her first language. And if she uses a cutting-edge frontier LLM, that's not specifically tuned for her Taiji.
She is, of course, at a disadvantage. So in Taiwan, when we train so-called sovereign AI models, because we have 20 national languages, 16 indigenous nations, 42 language variations, the sovereignty is actually communal. So each community may be working with Mozilla, Common Voice, Data Collective, and so on, do their own curation.
And it is actually about co-authorship because each language carry a way to dissolve agreements, also to understand outside world. And they own that data. And again, using local knowledge, artifact management, intelligence, which I call calming, they can negotiate and even do social translation across different epistemic standings.
And we're already seeing other countries adopting this social translation because polarization is a big problem everywhere. And social translation to convey, for example, climate justice views and biblical creation care views can de-escalate their conflict and then agree on the uncommon ground. So I'm very happy to also report that this kind of waging peace work, which is a core part of UN, is now taking place in many places.
You can look up Democracy R&D for such a network. So that is what I will recommend UN to look into. And it can also resolve the issue about statewide or national regulations about the social license to operate of AI, including transparency rules.
In fact, if you are a California citizen, you have three more days to go to engaged.ca.gov, Engage California, where we're using this Taiwan-inspired model to ask every Californian about their work, how it's impacted by AI. So it could be apprenticeship, belonging, community, dignity, ethics, privacy, anything of those ABCDE issues. And again, using such social translation, the hope is that we can find this uncommon ground that is cross-partisan, cross-cultural, and then the state of California can simply say, as we did to the deepfakes social media, to the Frontier Labs, this is not the governor's idea or the MP's idea.
This is not partisan. Everyone in California thinks you should do at least this. Lee?
Melanie Hart:
Yeah. So just a few comments. On the supply chain question, the United States and China are currently the AI leaders, depending on how you look at the metrics.
I'll say that they're the two AI leaders. And it will be interesting in that there are areas where we can collaborate and areas where we just can't. So on critical minerals, on rare earths, China has made very clear that they will cut us off from those supplies as a way to control U.S. decision making. So you will see a growing decoupling between the U.S. and China in critical minerals. We just have to have alternative supply. We cannot depend on China for that.
And when it comes to the higher end chips, there's very clear evidence that the Chinese military is trying to get ahold of those to run AI in ways that in our view undermines our national security. And we have to have export controls there. What I find really interesting is that there's also areas where we can collaborate, but doing so safely is going to require new mechanisms.
And I agree totally with Emmanuel that on data, cross-border movement of data in a way that is safe and secure and benefits both sides equally is really going to be the big key to unlock what we can actually do with AI. And there's tremendous potential there with China, but also tremendous risk. So I was really interested to hear about the French India framework.
I think maybe that could be a model for the U.S. and Europe and China and others, right? Because we will have to have a kind of independent standards and mechanism for how we can exchange data. It can't sit in any one country.
I'm doubtful that the UN could be the solution, but hey, welcome some ideas on that front. On the question about AI increasing inequality between nations, I totally agree with Kori's comment. This is going to be the biggest upheaval since the industrial revolution, and we still don't fully understand it.
But I think the economic losses where countries are not deploying AI are going to be real and serious. And just to give an example, last fall we went to a car and automobile manufacturing factory in China, the Xiaomi factory that uses AI plus robotics to make one electric vehicle in 76 seconds. And next week, we'll go to California to visit a Ford factory that's using similar AI and robotic innovation to make a Ford truck rapidly and at a price comparable with what the Chinese companies are reaching.
So you see the U.S. and China in this race. How can we use the in-factory AI models with robotics to manufacture higher quality faster with fewer people? If you're not deploying the same models, you're not going to be able to compete.
You won't be able to compete on scale, on cost, or on quality. What does that mean for Mexico or for another country that's an automobile manufacturer? You may wind up with just two different worlds, and you can't be in that higher level of competition without the most advanced AI and robotics, which are expensive.
So I think a lot of countries, including the United States, including China, are going to have to figure out what's our new economic plan, because some of what we were doing before may not work going forward. And then just finally to the question about what can governments do to push models to be more transparent, looking at the U.S. political system and sitting here in Washington, I don't think Washington can do this effectively, because our political leaders are always so far behind the companies in understanding what the AI is doing, much less trying to regulate the models. I think a lot of this, at least in the U.S. system and probably globally as well, will be based on market choice. So companies and citizens are going to want the more transparent, the cheaper, the simpler AI models that you get, kind of how both Audrey and Emmanuel talked about sometimes what you get back is garbage, right? I get a lot of garbage back when I experiment with U.S. language models, for example. So I think the best thing that government can do is to make sure citizens have wide access and choice, models from the U.S., models from China, models from Europe, local ones, big ones that house their data in a data center. I think if consumers and companies have a lot of choice, then people will gravitate to the ones that work best for them. And I just want to point out that not that many months ago, OpenClaw took the world by storm, and OpenClaw was an innovation from Austria, I think, right? So big, big global models can come from anywhere as long as consumers have choice.
And if they have choice, then transparency is ultimately going to win out at the end of the day.
Jean-Marie Guéhenno:
On this optimistic and positive note, I think that's a good ending. And I want to thank all our panelists, Kori, who had to leave, Melanie, Audrey, and Emmanuel. I think it was a very good discussion.
For me, the key takeaway is that, in a way, data are the oil of the future, and that every country sits on a trove of data which is valuable if it makes those data reliable, if it makes them accessible, interoperable, if it protects privacy. So all sorts of ifs that require a lot of thinking and efforts. But in a way, we all produce data around the world, rich and poor countries.
If we manage to produce them well, then there's an enormous field for cooperation and enormous potential for progress. And I think that's a pretty encouraging thought at the moment when we feel that there are only a few big companies that are going to control us through the data. I mean, the image of Audrey, I mean, is it extractive or is it not extractive?
And if we move from an extractive economy to a non-extractive economy, there's real hope. So thank you all. Thank you very much.
Melanie Hart:
Thank you.