'EU AI Act is first step in regulation of AI & we need advances in technology to detect deep fakes'

FRANCE 24 English

Visit our website: http://www.france24.com  Like us on Facebook: https://www.facebook.com/FRANCE24.English  Follow us on Twitter: https://twitter.com/France24_en

Transcript

00:00A Global Summit on Artificial Intelligence is underway in Paris.

00:14The event, which is taking place at the Grand Palais, is a chance for world leaders to rub

00:18shoulders with tech executives who are leading the change.

00:22In the run-up to the gathering, OpenAI's Chief Executive penned an op-ed in French

00:26daily Le Monde, where he wrote,

00:28If we want growth, jobs and progress, we must allow innovators to innovate, builders to

00:33build and developers to develop.

00:36For more on the summit, here's our correspondent reporting.

00:43If there was ever a sign that drumming up investment had taken priority at this summit,

00:49which is called the AI Action Summit, rather than the AI Safety Summit, as previous editions

00:54were called, it's the fact that today is really about doing business.

00:59We're seeing many more tech leaders here than we are world leaders so far.

01:03The plenary with the world leaders is tomorrow, and that's when there's supposed to be a political

01:08declaration to be signed by lots of them.

01:10But today, the headline is that 109 billion euro investment in French AI infrastructure

01:18that Emmanuel Macron announced in his speech on Sunday night.

01:24Now, one of the contributors to this pot of money is Hugging Face.

01:31That's an American company led by a French man called Clément Delong.

01:36He talked to us a little bit earlier.

01:39So we're going to double our investment with more team members, more infrastructure here

01:44in France.

01:45We will focus on open source AI, of course, that we've been doing for four years now.

01:50And also in robotics, we're really excited about in France, and a focus on AI apps.

01:58We're lucky to be the most used platform for AI builders at Hugging Face today, with more

02:04than 10 million AI builders using the platform.

02:08So hopefully, more investment will mean more impact.

02:12So you heard there him talking about open source AI.

02:14They are the leader in open sourcing AI projects.

02:18He also told me that he'd be doubling Hugging Face's investment in France as part of that

02:25109 billion euros announced by Emmanuel Macron.

02:30For more on this story, we can bring in Frederic Pascal, director of the AI Institute of Paris-Saclay

02:35University.

02:36Good afternoon.

02:37Thank you so much for joining us here on the program today.

02:39You were at this summit, this AI summit at the Grand Palais.

02:42Can you tell us what it's like, what the atmosphere is like?

02:46Yeah, for sure.

02:47Thank you for the invitation.

02:48And yeah, the atmosphere is very interesting because there's a big spotlight on Paris and

02:55also France, many announcements.

02:58And there's this connection between companies, academic, policy makers.

03:04So yeah, it's great.

03:06Can you tell us where exactly France and Europe is positioned in this AI race?

03:10Because it seems that the United States and China, out of nowhere, are leading the charge.

03:16I mean, the U.S. and China, they announced very big LLMs, so these AI models.

03:23We have also AI models in France with Mistral AI.

03:28So I think only one point is missing in France and Europe is investment.

03:34And so that's why the announcement of President Macron was very good.

03:39OK, you're talking about the 109 billion dollars.

03:42I'm going to ask you about that in just a little bit.

03:44Emmanuel Macron's team had a little fun using artificial intelligence ahead of the summit.

03:49Take a look at what they came up with.

04:14OK, so there you had the French president essentially talking about how AI can be used,

04:23but it also shows the dangers of this technology, doesn't it?

04:27Yes.

04:28So I think this is something interesting because we can see that anybody can use AI, can use

04:35AI models, AI apps.

04:38But of course, there is danger.

04:40So that's why it's a way to open discussion on regulation, what we can do with AI, how

04:47we have to specify that new video, new images is from AI or not.

04:54We had Sam Altman who wrote an op-ed in Le Monde in the run up to this conference taking

05:00place where he essentially called for, if we want growth, jobs and progress, we must

05:05allow the innovators to innovate, builders to build and developers to develop.

05:10But he's not talking about regulation, is he?

05:13Where are we on regulation?

05:14So this is one of the objectives of the summit, to find a good tradeoff between innovation

05:19on one aspect.

05:21And I mentioned Mistral, but we saw Hugging Face as a French company.

05:26So innovation on one part and regulation on the other part.

05:30Regulation on the data.

05:31So there is already the GDPR law, but also regulation on the usage of AI, just to specify

05:40the law around the usages of AI.

05:44Because for example, a couple of months ago, one of my colleagues on the French channel,

05:49he had his, there was his face, his voice, but they were using a different language and

05:57saying something totally different.

05:59This is just one example of how AI can be used negatively and essentially spread disinformation.

06:05Yeah, that is the main problem.

06:08And this is a risk to use AI in a wrong way.

06:11So that's why I think something essential is to train people, to train people to AI,

06:18how to use AI, what is a good use or a bad use, what are the risks of AI.

06:24This is something essential.

06:25I think France starts in that way.

06:29So you are director of an AI institute at a university here in Paris.

06:35What are some of the positive ways students can be using artificial intelligence?

06:39OK, so in our institute, one main axis is on AI for medicine.

06:45We saw the medical imaging yesterday during the President Macron interview.

06:51And those images came from research from my institute.

06:57So this is one aspect that is very positive, how to help doctors to find where a cancer

07:04is exactly, to have a personalized medicine.

07:07This is very important implication of AI.

07:12Another aspect we work on is on the mathematical foundation of AI.

07:15So how to explain those black box model?

07:19Because we know it's easy to generate images, video, but we have to explain how it works.

07:25So this is another aspect of the research activities in the institute.

07:31And what are some of the downfalls of it?

07:33What are some of the downfalls of artificial intelligence on college campuses?

07:41It's kind of difficult.

07:44There's many, many uses of that.

07:47I talk about research, but also for training.

07:51For us, AI brings something completely different.

07:54It's not only for society.

07:56We know Chagipiti or Le Chat, but also in teaching at the university, how it would transform

08:02all the services.

08:04So that's also something essential for us to train with or at the hour of AI.

08:13Stay with us, Frédéric, because another thing we've been talking about is can we safely

08:17embrace artificial intelligence?

08:20For decades, sci-fi films and novels have imagined scenarios where intelligent machines

08:24shed the shackles of human control to take over the world.

08:28Now once a concept of our collective imaginations, AI is now part of our everyday lives.

08:32And it's not uncommon to hear concerns about such systems getting too much power.

08:38Eliza Herbert reports.

08:42Joe Biden and Donald Trump closer than ever.

08:45Or Emmanuel Macron embracing Marine Le Pen.

08:49As deepfakes grow more and more prolific, so do the abuses of artificial intelligence.

08:55In just a few clicks, an image can be falsified and virally spread across the internet.

09:01But bots, too, are adding to disinformation.

09:04According to experts, programs that impersonate real people can be a dangerous digital tool,

09:10especially when generating myths or rumours around election campaigns.

09:15Now important questions that we should tackle, the future of the workforce, of course, the

09:22impact on our society, on our democracies.

09:25Automation threatens to replace jobs like customer service, retail and many other professions.

09:32Generative AI can clone voices used in impersonation scams.

09:36And services such as ChatGPT, capable of producing text, image and sounds, directly endanger

09:43artists' livelihoods, copyright and the collective imagination.

09:49How can we fail to see that this is the Trojan horse for giving up the use of our most fundamental

09:55faculties?

09:56Another issue is on the battlefield.

09:59Militaries already use AI in warfare for mass surveillance and with autonomous drones.

10:06And recently, Google dropped its ban on using AI to develop weapons.

10:11It could also be used in much more dangerous and harmful ways.

10:18For example, kill anyone who fits the following description.

10:25Bias and discrimination, privacy and data leaks and the environmental cost of the heavily

10:32energy-dependent services are a few more of the risks that come with AI.

10:39We can now bring in Frédéric Pascal once again.

10:42He's director of the AI Institute for Paris-Saclay University.

10:47Frédéric, talk us through the dangers of the disinformation and deepfakes.

10:52Yeah, so the video was very interesting because I think AI does not work alone.

10:58There's always human working with AI, deciding AI can advise, help, assist, but the decision

11:05should remain human, especially we saw war images.

11:10So that's something important.

11:12For generative images, it's very important to be able to recognize, to make the difference

11:18between generative images and true images.

11:22Same thing for the voice.

11:23Same thing for video.

11:24So generative images are sort of Macron, his face on it.

11:30Okay, so that's what a generative image is.

11:32Yes, it's something that does not exist, that is generated by an AI system.

11:37And generally, it's very difficult to recognize that it's not a true image.

11:43That's why we mentioned deepfakes, because it's very difficult and it asks a new question.

11:48I heard about democracy, that is something very important, where the truth is.

11:53And what about deepfakes that appear on social platforms like Facebook, for instance?

11:59What should be done about policing that?

12:04It would be difficult to answer because, you know, the CEO of Facebook or Meta didn't share

12:10the same policy as Europe, for instance, who wants to regulate, to control.

12:17This is something that has been decided with an international governance of AI.

12:24This is essential because we cannot produce anything without explaining what is this production,

12:32especially for images, voice or videos.

12:36Do you think it's possible for Europe to keep a regulation in check here on the continent

12:41or is it going to be difficult for Europe if there is no regulation in the United States,

12:46for instance?

12:47Yeah, I agree it would be difficult, but I think AI Act is the first piece of this regulation.

12:55And also now we need advances in technology, I mean, to be able to detect deepfake.

13:02That is not easy, by the way.

13:05Now, the French president, as you said, announced this 109 billion dollars, billion euros rather

13:10investment in the years to come.

13:13Where will this money be used and where will it come from exactly?

13:16So mainly the money will be for computing resources because, as you know, AI system,

13:23to build AI system, we need computing resources.

13:27And also for using AI, we need computers.

13:30So this is the main part of the usage of this money.

13:35There's several hedge funds that have announced investing in this envelope of 109 billions.

13:44OK, so we'll see how this pans out and how this summit pans out.

13:49Frédéric, thank you so much for taking the time and coming to speak to us here on François

13:52et Cat.

13:53Thank you for the invitation.

13:55Now, while using AI tools like ChatGPT may seem revolutionary.

13:59The truth is new data centers are run at a huge cost to the environment.

14:04Eliza Herbert has been finding out.

14:08Artificial intelligence is rapidly advancing.

14:11Computer programs are able to replicate the world in detail with precision.

14:16But will it come at a cost to the real thing?

14:20Estimates say that a single ChatGPT inquiry requires around 10 times the electricity of

14:26a Google search.

14:28According to the International Energy Agency, the local data centers that power AI are already

14:34straining local grids, often burning fossil fuels which jeopardize climate targets.

14:40It's a topic on the mind of some world leaders at the AI summit in Paris this week.

14:46How we make sure that we have the energy to power AI in a clean enough way and an efficient

14:53enough way that's not going to set us back in our fight to reduce carbon emissions.

14:59Ahead of the summit, over 120 organizations signed a joint statement calling for AI systems

15:06to be made within the demands of planetary boundaries.

15:10Fossil fuels must be phased out to reduce the greenhouse gas emissions heating the planet.

15:15Yet there is increasing evidence of AI systems driving up emissions and locking in reliance

15:20on fossil fuels while exhausting critical resources like water, land and raw materials.

15:27Data centers are large water consumers relying on water to cool servers and prevent them

15:33from overheating.

15:34According to the UN, global AI-related infrastructure may soon consume six times more water than

15:40Denmark, a country of six million people.

15:44Another problem is mining.

15:47One two kilogram computer requires 800 kilograms of raw materials and their microchips require

15:54rare earth minerals.

15:55AI tools such as mapping greenhouse gas emissions also have the potential to combat environmental

16:02crises.

16:03The question is, will they do more harm than good?

Category

Transcript

Recommended