数字化转型网人工智能专题
翻译：
The EU’s Artificial Intelligence Law was officially released, ushering in a new chapter in AI regulation
Digital Transformation Network Artificial Intelligence topics
Learn with the world’s top AI professionals! Digital Transformation Network has established a research and learning community dedicated to discussing artificial intelligence technology, industry, and academia, and grow together with you! Welcome to scan code to join!
On July 12, 2024, the Official Journal of the European Union officially published the Artificial Intelligence Act (AI Act). This landmark legislation contains 180 preambles, 113 articles and 13 annexes, establishing a comprehensive framework for the development, deployment and use of AI systems in the EU. The law will officially take effect on August 1, local time, and August 2, Beijing time, 20 days after its announcement, and a phased and risk-based approach has been adopted in legislation.
1. It will be implemented in phases
● Before November 2, 2024
Each member State designates an AI regulator
● From 2 February 2025
The bill’s general provisions and prohibitions will come into effect, and a list of prohibitions against artificial intelligence that poses an “unacceptable risk,” such as social credit ranking systems, will be applied within six months of the law’s enactment
● Before May 2, 2025
The EU Office for Artificial Intelligence will issue a code of conduct
● From August 2, 2025
The General Artificial Intelligence Model (GPAI) regulations come into effect, and providers will be required to meet obligations such as transparency
● Before February 2, 2026
The Commission needs to propose a template for a post-market monitoring plan and a list of elements to be included in the plan
● Before August 2, 2026
The act will come into full effect, meaning everything else in the act will apply. Includes all obligations for high-risk AI systems as defined in Annex III (List of High-risk AI Systems). The infrastructure related to governance and compliance assessment systems should be operational, and Member States should establish at least one AI regulatory sandbox at the national level, which should be operational by 2 August 2026 at the latest
● Before August 2, 2027
Some obligations related to high-risk AI systems will begin to apply. Examples include high-risk AI systems for medical devices, machinery, and radio equipment. Providers of general-purpose AI models placed on the market before 2 August 2025 shall take the necessary steps to comply with their obligations under this Regulation by 2 August 2027
● By November 2028
The European Commission should draw up a report on the devolution of regulatory powers
● Before December 31, 2030
Artificial intelligence systems that are part of large-scale information technology systems that have been placed on the market or put into use before August 2, 2027 shall comply with the provisions of this Law.
2. Scope of application of the Act
The EU Artificial Intelligence Act establishes a comprehensive regulatory framework with jurisdiction extending to public and private entities operating within the EU. The scope of the bill is remarkably broad, reflecting the proliferation of AI technology and its potential impact on EU citizens. The main entities subject to this regulation include:
Ai System Provider:
Entities developing AI systems designed to be placed on the EU market or deployed within the EU fall within the scope of the Act. Article 3(2) defines a provider as “a natural or legal person, public body, institution or other body that develops or commissions the development of an AI system in order to place it on the market or put it into use under its own name or trademark, whether or not for a fee”.
Artificial Intelligence system users:
The Act applies to organisations and individuals using AI systems within the EU. According to Article 3(4), a user is defined as “any natural or legal person, public authority, institution or other body that uses an AI system under its authority, unless the AI system is used for personal non-professional activities”.
Importers and Distributors:
Entities involved in the import or distribution of AI systems on the EU market are subject to specific obligations under the Act. Articles 26 and 27 detail, respectively, the responsibility of importers and distributors to ensure compliance with the regulations.
Third country entities:
As stated in Article 2(1), the extraterritorial effect of the Act extends its application to providers and users of AI systems located outside the EU, as long as the output produced by these systems is used within the EU. This regulation ensures that AI systems affecting EU citizens are regulated.
The scope of the Act is further refined through a series of exemptions and specific application provisions:
Military and Defense:
Article 2(3) specifically excludes artificial intelligence systems developed or used exclusively for military purposes.
Research and Development:
Section 2(5a) provides certain exemptions for AI systems used solely for research and development purposes.
Legacy systems:
Article 83(2) Outlines transitional provisions for AI systems placed on the market or put into service before the date of application of the regulation.
3. Outline of the provisions of the Bill
The EU Artificial Intelligence Act, which will come into force on August 1, 2025, has 13 chapters, 113 articles and 13 appendices.
4. Risk-based approach
The EU AI Act adopts a risk-based approach, dividing AI systems into four tiers based on their potential impact on individual rights and security.
(1) Unacceptable Risk.
Artificial intelligence systems, defined as posing a clear threat to safety, livelihood and fundamental rights, are prohibited under Section 5 of the Act.
Examples include social scoring systems, systems that exploit the vulnerability of specific individuals or groups of people, systems that use the subconscious beyond human consciousness, real-time remote biometrics systems in public Spaces, predictive policing systems for individuals, and emotional recognition systems in the workplace and educational institutions.
(2) High Risk.
Defined as an AI system that is likely to endanger safety, fundamental rights or cause significant adverse effects, subject to strict compliance and transparency obligations under Articles 6-51 of the Act.
Examples include artificial intelligence for critical infrastructure, education, employment, essential services, law enforcement, immigration management, and justice administration. Determining high risk needs to be based on the function, intended purpose, and behavior of the AI system. Suppliers of such high-risk AI systems must run conformity assessment procedures before they can sell and use their products in the EU.
(3) Limited Risk.
Defined as an artificial intelligence system that requires specific transparency measures.
For example, chatbots, emotion recognition systems, etc. The transparency obligations outlined in sections 52-54 of the Act. Such systems are subject to information and transparency requirements, and the deployer of an AI system that generates or manipulates image, audio or video content (deepfakes) must disclose that the content has been artificially generated or manipulated, except in certain circumstances (such as when it is used to prevent criminal offences). Providers of AI systems that generate large amounts of synthetic content must implement techniques and methods (such as watermarking) that are reliable, interoperable, efficient, and robust enough to be able to flag and detect that the output is generated or manipulated by AI systems rather than humans.
(4) Minimal Risk.
It is defined as an application of AI that poses minimal or no risk to the rights or safety of citizens. For example, AI-enabled video games, spam filters, etc. There are no additional legal requirements in addition to existing laws. According to the European Commission, the vast majority of AI systems currently in use or likely to be used in the EU are low-risk systems.
5. Main provisions of the Act
The bill introduces several key provisions that stakeholders must comply with, including:
Risk-based classification:
Ai systems are classified according to their level of risk, with specific requirements for high-risk AI systems, including transparency, data governance, documentation, and human oversight.
Prohibited acts:
Certain AI practices are deemed unacceptable and banned, such as those that manipulate human behavior to circumvent users’ free will, or that allow governments to conduct “social scoring.”
Transparency obligations:
Ai systems that interact with humans or are used to detect emotions or determine social categories based on biometric data must be designed to be transparent.
Data Governance:
High-risk AI systems must be trained, validated, and tested on high-quality data sets that are relevant, representative, unbiased, and respectful of privacy.
Market supervision:
Member States must designate national authorities to conduct market surveillance in order to ensure the enforcement of the Act.
6. Important definitions of the Bill
Artificial intelligence system:
Is a machine-based system designed to operate with varying degrees of autonomy, may exhibit adaptability after deployment, and inferences from the inputs it receives how to generate outputs that can affect a physical or virtual environment, such as predictions, content, recommendations, or decisions, for explicit or implicit goals.
Widespread infringement:
Any act or omission that violates the laws of the Union protecting the interests of individuals.
Risks:
Refers to the combination of the likelihood of a hazard occurring and the severity of the hazard.
Put on the market:
It refers to the first AI system or general AI model available on the EU market.
7. Prohibited artificial intelligence practices
The use of subconscious techniques or purposeful manipulation or deception techniques beyond the individual’s conscious awareness.
The purpose or effect is to materially distort the conduct of a person or group of persons by manifesting impairment of the ability of that person or group of persons to make informed decisions, thereby causing or potentially causing significant harm to that person, another person or group of persons by causing that person or group of persons to make decisions he or she would not otherwise make;
Exploit any vulnerability of a particular individual or group due to their age, disability or particular social or economic situation
To have the purpose or effect of materially distorting the conduct of that person or a person belonging to that group to cause or be reasonably likely to cause significant harm to that person or others.
8. Fines for general AI model providers
The Commission may impose a fine on providers of general AI models not exceeding 3% of their total global turnover for the previous financial year or EUR 15 million, whichever is higher.
9. Countries have joined hands to introduce policies
The agreement follows years of discussions among EU member states and MEPs on how AI should be regulated. In 2021, the European Commission proposed the adoption of the Artificial Intelligence Act. After the release of ChatGPT last year, the regulation of AI became urgent, and countries including China, the United States, and the United Kingdom are rapidly advancing the construction of rules on AI governance.
On October 30, the United States issued an executive order “on the Safe, Reliable and Trusted Development and Use of Artificial Intelligence” on artificial intelligence, proposing new standards for AI security. The executive order lays out eight guiding principles and priorities to advance the development and use of AI, including: Setting new safety standards for AI; Protecting Americans’ privacy; Promoting equity and civil rights; Safeguarding the rights and interests of consumers, patients and students; Support workers; Promote innovation and competition; Advancing American leadership abroad; Ensure that governments use AI responsibly and effectively.
Some legal experts believe that on the one hand, “administrative order” is to strengthen the right to control the industrial chain and supply chain; On the other hand, through information disclosure, intelligence sharing, security and risk traceability control.
In July this year, the National Cyberspace Administration of China and seven other departments jointly issued the Interim Measures for the Management of Generative Artificial Intelligence Services, which took effect on August 15. On October 18, the CAC released the Global Artificial Intelligence Governance Initiative. The initiative has put forward 11 initiatives from the three dimensions of development, security and governance. It mentioned that support for artificial intelligence technology to prevent artificial intelligence risks, artificial intelligence needs to be viewed dialectically, it may have a “far-reaching impact”, but also bring “unpredictable risks and complex challenges.”
It is worth noting that among the more than 20 consensus reached by the two sides after the meeting between the heads of state of China and the United States in San Francisco in November, one is specifically about artificial intelligence – China and the United States agreed to establish an intergovernmental dialogue mechanism on artificial intelligence.
In the same month, at the first Global Artificial Intelligence (AI) Security Summit, China and 27 other countries and the European Union also signed a major agreement on AI, the Bletchley Declaration, which promotes “a shared understanding of the opportunities and risks presented by cutting-edge AI, and the need for governments to work together to address the most significant challenges.”
European Union Artificial Intelligence Law, English original, Chinese translation download
Digital Transformation Network Artificial Intelligence topics
Learn with the world’s top AI professionals! Digital Transformation Network has established a research and learning community dedicated to discussing artificial intelligence technology, industry, and academia, and grow together with you! Welcome to scan code to join!
