La réglementation fera-t-elle de l'UE la puissance la plus fiable en matière d'IA ?
Veiller à ce que l'intelligence artificielle (IA) respecte les lois et les valeurs fondamentales et soit sûre pour les citoyens, tel était l'objectif de l'UE lorsqu'elle a adopté la loi sur l'IA.
00:30The AI Act came into force in 2024 and defines four risk levels for the use of artificial intelligence, from minimal to unacceptable.
00:53It prohibits practices that violate democratic values and human rights.
00:58One example is categorising individuals by their biometric data, which could lead to unfairly profiling minority groups.
01:06The EU now invites AI companies such as the generative chatbots ChatGPT, Mistral, Gemini and Claude to sign a voluntary code of practice on general-purpose AI.
01:18By signing the code and adhering to the rules, they're deemed compliant with the AI Act.
01:23Companies that refuse to sign the code may face more stingent expectations and administrative burdens.
01:30Let's hear what Europeans think about this regulatory move.
01:35I feel like if we leave these programs unregulated, they will also, like, they will make harm as much as good as they do.
01:45It has to be like a global thing.
01:47So then it's the same rules in China and the same rules in here, but also the same rule in America.
01:54Let's break it down with Cynthia Krut, senior tech policy reporter at Aeronews.
02:21When it comes to the code, the European Commission is taking somehow a pedagogical approach to this sector.
02:28But what will happen to AI companies if they will violate the AI Act once it's fully implemented?
02:37Yeah, I think it's important to know that the AI Act is being implemented in a period of time.
02:43So from 2024 to 2027.
02:45And this August, the rules for general purpose AI models, such as JetGPT or Gemini, they will enter into force.
02:54So that means that the products that were already on the market, such as those examples, they have two more years to comply with the rules.
03:01And everything that's been put on the market after August, they have to comply immediately.
03:06And in case there are breaches, the Commission can impose a fine of up to 15 million euros.
03:13Major players like OpenNI and Anthropik support the code, decided that they wanted to sign it.
03:20But others are refusing, like Meta, which is also an important company.
03:25What are the arguments being used and what is the implication of such a refusal in terms of the global market?
03:32Meta has been, I think, since the drafting process of the code started last September, very critical of this whole process.
03:40They say it stifles innovation and they've rolled out a few tools that they cannot fully use in Europe, also because of data protection rules, for example.
03:49But in the end, it doesn't matter much if they sign or not because the AI Act will prevail anyway.
03:55Some analysts claim that this regulation is basically a strategic positioning of the EU that wants to be perceived as the most trustworthy AI provider.
04:09The USA is highly critic of this approach.
04:12So what can come out of these conflicting positions?
04:17It's not only Europe, I think, that's regulating.
04:19China and the US are doing the same.
04:21I think they just take a more sectoral approach instead of more horizontal that the EU is doing.
04:28I think the big difference is really investment in the end.
04:31It's mostly China and the US that have a lot of private investment.
04:36Europe is trying to catch up and they try to mobilize private and public funding, for example, for gigafactories where they train these AI models.
04:45This code establishes rules on three main aspects.
04:49Copyright, solutions to respect the intellectual property of creative works, such as images and academic essays.
04:56Safety, standards for avoiding systematic risks of advanced AI models.
05:01One example would be development of chemical or biological weapons.
05:05And transparency, requiring companies to complete a form on how they comply with the AI law.
05:10Our guest is Laura Lázaro Cabrera, a counsel at the Center for Democracy and Technology.
05:18Denmark, which currently holds the presidency of the EU, wants to simplify the AI Act and another digital rules.
05:29Could this legislation become an empty promise?
05:33The AI Act is already the result of several years of protracted negotiations and hard-fought compromises.
05:39Every sentence in the AI Act, every word is in there for a reason.
05:45So to say that simply a lot of the AI Act can be done away with or can be removed, some sections amended, would be a mistake.
05:54The United States leads the private investment in AI, which is more than 10 times the EU's investment.
06:03Beyond regulation, do you think that the EU is financially committed to providing safer models of AI?
06:12The EU has made great strides towards strengthening the financial support that it provides to AI development in Europe.
06:20Just this year, over 200 billion euros have been announced for AI investment.
06:25We think that finances are an important part of the equation, and indeed it is important for the EU to maintain a leadership role in the development of AI.
06:34But we think that that leadership has to be tied to a strong safety framework that promotes fundamental rights and that promotes people-centered AI systems.
06:44Deepfakes, theft of confidential data, suicides linked to the use of chatbots are some examples of the risks, the dangers of generative AI.
06:57So in addition to regulating companies, should governments also provide awareness to citizens, training about how to use these tools?
07:10Absolutely. One key element of the AI Act is this notion of AI literacy, but the obligations around AI literacy are aimed at companies developing AI and companies deploying AI.
07:24We need to see similar approaches building at EU-wide level, targeted this time at people.
07:31Just as the industrial revolutions driven by steam power and the internet impacted different sectors, so too will AI affect many areas, from defense to life science to energy and manufacturing.
07:46However, as with those revolutions, there are risks to consumers, the environment and the rule of law.
07:52Can the legislation keep up with this lightning-fast innovation?
Écris le tout premier commentaire