39 hours of negotiations!EU AI regulation reaches a milestone agreement, pointing directly at ChatGPT and Google Bard

The world's first comprehensive draft regulation in the field of artificial intelligence has been agreed.

According to media reports, the European Parliament, EU member states and the European Commission, after nearly 40 hours of lengthy negotiations, reached an agreement on the "Artificial Intelligence Act" on the evening of December 8, local time in Brussels.

It is reported that the bill aims to provide better conditions for the development and use of this technology by comprehensively regulating artificial intelligence, negotiating and agreeing to implement a series of controls on generative AI tools such as OpenAI's ChatGPT and Google's Bard.

European Commissioner Thierry Breton posted on the X platform:

"The Artificial Intelligence Act (AIAct) is more than just a rulebook, it is a springboard for EU start-ups and researchers to lead the global AI race. ”

According to related reports, the draft legislation still needs to be formally approved by EU member states and parliaments, and once formally approved by both parties, the legislation is expected to enter into force early next year and should be implemented in two years, which may set the tone for global regulation of AI technology.

The spearhead is directed at ChatGPT and Google Bard

According to the official website, the European Parliament's top priority in promoting AI regulation is to ensure that the AI systems used in the EU are safe, transparent, traceable, non-discriminatory and environmentally friendly;

As such, it is stated that the draft legislation proposes a balanced, proportionate approach to horizontal regulation of AI, limited to the minimum necessary requirements to address AI-related risks and issues, without unduly restricting or hindering technological development, or otherwise disproportionately increasing the cost of bringing AI solutions to market.

Specifically, the main contents of the draft include the following points:

Adopt a risk-based approach to harmonize rules for the development, market launch and use of AI systems within the EU. Prohibit certain AI practices that run counter to EU values and are particularly harmful. Such practices include the use of AI for "social scoring" and the manipulation of human behavior to circumvent their free will. Specific restrictions and safeguards are proposed for remote biometric systems for certain law enforcement purposes. For example, biometric scanning, which classifies people based on sensitive characteristics such as political or religious beliefs, sexual orientation or ethnicity, will be banned, which the media has described as one of the most difficult and sensitive issues in the negotiations. A credible risk methodology is set out to define "high-risk" AI systems that pose a significant risk to human health and safety or fundamental rights. For certain AI systems, only a minimum transparency obligation is imposed, especially when using chatbots or "deepfakes".

Notably, Breton said lawmakers agreed to a two-tiered approach, namely "transparency requirements for all general AI models, such as ChatGPT," and "stricter requirements for robust models with systemic impact" across the EU.

People familiar with the matter also revealed to the media on December 8 that negotiators have agreed to a series of controls over generative AI tools such as OpenAI's ChatGPT and Google's Bard that can generate content based on commands.

That said, the draft directly requires underlying models such as ChatGPT and Artificial Intelligence General (GPAI) to comply with transparency obligations before they are placed on the market, which includes drafting technical documents, complying with EU copyright law, and disseminating detailed summaries of the training content.

For example, high-impact AI models with systemic risk must undergo model evaluation, assess and mitigate systemic risk, conduct adversarial testing, report serious incidents to the European Commission, ensure cybersecurity, and report on their energy efficiency;

In terms of penalty provisions, the proposed legislation would impose financial penalties of up to €35 million, or 7% of global turnover, on companies that violate the rules, depending on the non-compliance and the size of the company.

Next, the EU and others will discuss the details of this draft, which may change the form of the final legislation. After the law is officially promulgated and implemented, each member state will make certain adaptations based on its own public policy basis.

Taking the lead in legislation, Europe leads the AI race?

If successful, the draft would mean that the EU would become the first major world power to enact AI laws.

Breton Display:

"Europe positions itself as a pioneer and understands the importance of its role as a global standard-setter. Yes, I believe, it was a historic day. ”

According to CCTV News, the EU has been hoping to promote the introduction of the Artificial Intelligence Act as soon as possible.

Since 2019, the EU has been laying out the field of AI regulation, successively issuing AI ethics guidelines and issuing the AI White Paper.

In April 2021, the European Commission presented a draft mandate for the negotiation of proposals for the Artificial Intelligence Act. Since then, the European Parliament and the Council of the European Union have undergone several rounds of revisions and discussions on the draft.

In June 2023, the European Parliament voted 499 in favor, 28 against and 93 abstentions, overwhelmingly passing the draft mandate to negotiate the Artificial Intelligence Act, pushing the bill into the final stage of the legislative process, that is, holding "tripartite negotiations" between the European Parliament, EU member states and the European Commission to determine the final provisions of the bill.

The introduction of the Artificial Intelligence Act will also facilitate the development of AI technology in Europe itself.

The draft mentions:

"Given the pace of technological change and the challenges that may be faced, the EU is committed to working towards a balanced approach. It is in the EU's interest to maintain the EU's technological leadership and ensure that Europeans can benefit from new technologies that are developed and operated in accordance with EU values, fundamental rights and principles. ”

Opposing voices

In June this year, dozens of large European companies, including Airbus and Siemens, took collective action to publicly oppose the EU's passage of the Artificial Intelligence Act, citing "potential damage to European competitiveness" and "failure to address potential challenges".

Open AI's Altman has also previously criticized the European Union's plans to regulate AI technology.

The proposed law, which was drafted earlier, implied that both ChatGPT and GPT-4, a large language model, could be designated as high-risk. Altman has responded: "If we can comply, we will, and if we can't, we will stop operating." We need to try, but there may be technical limitations. ”

After the results of the draft agreement were announced, Ella Jakubowska, senior policy adviser to the European Union, said:

"It's hard to be excited about the EU's first steps to legalise real-time public facial recognition across the EU. ”

"While parliament has worked hard to limit the damage, the overall approach to biometric surveillance and analytics has been lukewarm at best. ”

Cecilia Bonefeld-Dahl, Director General of the business organization DigitalEurope, said:

"We made an agreement, but at what cost? We fully support the risk-based approach based on the use of AI, not the technology itself. ”

However, perhaps the more important significance behind the agreement on the draft bill is that it provides a policy blueprint for regulating AI technology.

Ursula von der Leyen, President of the European Commission, the EU's executive body, said: "We will support companies and developers in anticipating the new rules until the law is fully applicable. Around 100 companies have already expressed interest in joining our Artificial Intelligence Act.