On March 13th, local time, at the plenary session of the European Parliament held in Strasbourg, France, the European Parliament formally voted to pass and approve the EU Artificial Intelligence Act. This is the world’s first artificial intelligence (AI) bill, marking the beginning of a new era of AI supervision.
National business daily reporter noted that the European Union’s "Artificial Intelligence Act" was first born in 2021, which will regulate AI technology based on the degree of risk.
After the bill is approved, how to supervise the generative tools such as ChatGPT will become the focus of attention. Some analysts say that the generative model is included in the general AI model, and there will be different regulatory requirements according to whether there is "systemic risk", which may cause difficulties for the actual supervision of large language models.
In an interview with national business daily, Steven Farmer, a partner of Pillsbury Winthrop Shaw Pittman LLP, one of the world’s top law firms, said, "I think the EU’s AI bill is likely to become a template for the world, which is a critical moment for the global supervision of AI competition." Taking the EU’s General Data Protection Regulation as an example, he explained that in terms of AI regulation, the EU has determined the global direction.
It is reported that the core goal of the Artificial Intelligence Act is to establish a regulatory framework, use the potential of AI responsibly, and ensure that technology benefits the public interest while protecting individual rights.
It is worth noting that the bill sets different requirements and implementation mechanisms according to the degree of risk that the AI system may cause to users. The reporter found that the AI system was divided into four different risk levels: unacceptable risk, high risk, finite risk and low risk. According to the function and influence of AI system, the Act prohibits eight types of AI applications.
It mentioned that "all AI systems that are considered to pose a clear threat to people’s safety, livelihood and rights will be banned, from the government’s social rating to the use of voice-assisted toys to encourage dangerous behavior."
National business daily reporter further reviewed the bill and found that according to the function and actual or possible impact of AI tools,The bill prohibits eight types of AI applications.
After the bill was passed by the European Parliament, how the generated AI tools such as ChatGPT will be regulated immediately became the focus of attention.
According to foreign media, the bill only explicitly mentions the "generative model" twice and incorporates these systems into the so-called "General AI (GPAI) model". Depending on whether there is "systemic risk" in these models, they may be subject to different requirements.
Charles-Albert Helleputte, head of EU data privacy, network security and digital assets business of Squire Patton Boggs, a global law firm, said, "For large language models, it is not entirely high risk, and there is an intermediate category (specific transparency risk), which may lead to many thorny problems in actual implementation."
According to foreign media reports of science and technology law, since the first legal dispute of generative AI tools, the attribution of responsibility between AI service providers and users has been unclear. And this bill gives a clear answer: both sides have responsibilities. Although AI system providers or developers undertake most compliance requirements, how users or deployers use these tools will determine the types of responsibilities they face.
In this regard, Ashley Casovan, managing director of the AI Governance Center of the International Privacy Professionals Association, explained: "The bill shows that if the deployers make significant enough changes to the development of the platform, they are both developers and deployers." But he mentioned that the meaning of "significant" determines how to implement this bill.
However, Helleputte believes that it may be difficult to distinguish the responsibilities between developers and users. "When we look at the differences between developers and users in the bill, you will find that there is a transfer of responsibilities between one party and the other … The boundary may not be so clear."
It is reported that the bill will take effect 20 days after its publication in the official gazette, and it will be fully applied two years later, and some rules will be implemented in stages.The ban will take effect in six months; The governance rules and obligations of the general AI model will apply after 12 months; AI system rules embedded in regulated products will be applied after 36 months. The bill will apply to any company doing business in the EU,Companies that violate the law may be fined up to 35 million euros, or 7% of the company’s global revenue.
Some legal professionals said that the bill is an important milestone in the supervision of AI by international law, which may pave the way for other countries to follow suit.
Steven Farmer, an AI expert and partner of Pillsbury Winthrop Shaw Pittman LLP, one of the world’s top law firms, said in an interview with national business daily, "I think the EU’s AI bill is likely to become a template for the world. This is a critical moment for the global supervision of the AI competition. The EU has determined the global direction."
Some foreign media mentioned that one of the biggest challenges for this bill is whether the EU can achieve the "Brussels effect" in the global AI technology field, that is, the ability to radiate the EU’s legislative framework to other countries and regions.
Steven Farmer mentioned to every reporter an important precedent in data supervision in the European Union: the General Data Protection Regulation (GDPR) promulgated in 2016. Within two years after the promulgation of this regulation, global technology giants such as Meta and Microsoft have updated their related services. Today, privacy standards and awareness are commonplace in most jurisdictions.
"In terms of regulatory data, the EU took the lead in providing GDPR for us, and (later) we also saw the global convergence in this direction. The Artificial Intelligence Act seems to be repeating history. " Farmer told reporters like this.
In fact, the passage of the AI bill faces many controversies. During the negotiation, some EU countries, such as France and Germany, thought that the AI rules were too strict, which might hurt European start-ups.
Cecilia Bonefeld-Dahl, Director-General of Digital Europe, a European technology industry organization, once said: "We will pay the price for this bill, which means that we need to invest more in studying ways to combat the risk of AI use, rather than investing in the technology itself."
"Some people worry that the bill will stifle innovation, but in fact I think the bill sets some safety parameters for innovation. Look at the list of AI systems that are prohibited (in the bill). I really don’t see what these are worth opposing. " Farmer told every reporter. "The EU’s actions (regulated by AI) have always been controversial, but I think it balances safety and innovation, which is a very comprehensive framework. When other countries seek to formulate similar measures, the EU may enjoy a first-Mover advantage. "
Although this is the world’s first AI bill, the EU is not the only one in terms of AI technology supervision. Last year, the United States signed a comprehensive executive order on AI, and legislators in at least seven states in the country are making their own AI legislation. In addition, China, Japan, Brazil and other countries and regions, as well as organizations such as the United Nations and G7, are all working on formulating AI regulatory rules.
However, Emma Wright, a partner of Harbottle & Lewis Law Firm, is worried that the EU bill may soon become obsolete with the rapid development of technology. "Considering the speed of technological change-as shown by the introduction of generative AI last year-the EU AI bill will soon become obsolete, especially considering the time frame of implementation," she said.
national business daily