TL;DR:
- The European Parliament’s leading committees have approved the AI Act, paving the way for plenary adoption in mid-June.
- The AI Act aims to regulate AI and mitigate potential harm, becoming a landmark legislation for Europe and the world.
- The definition of AI in the legislation aligns with that of the OECD, anticipating potential revisions.
- Prohibited practices include biometric categorization, predictive policing, and emotion recognition in specific contexts.
- General-purpose AI systems have a tiered approach, with obligations falling on economic operators and more stringent requirements for foundation models.
- Generative AI models like ChatGPT must disclose AI-generated texts and provide a summary of copyright-protected training data.
- High-risk AI providers face detailed obligations in risk management, data governance, and documentation.
- Users of high-risk AI solutions must conduct fundamental rights impact assessments.
- A centralized enforcement architecture is favored, but the proposed AI Office’s tasks were reduced, with a focus on guidance and coordination.
- The Commission is responsible for resolving disputes among national authorities regarding dangerous AI systems.
Main AI News:
The AI Act, a groundbreaking legislation aimed at regulating Artificial Intelligence (AI) and mitigating potential harm, has received approval from the leading parliamentary committees of the European Parliament. This crucial development sets the stage for plenary adoption in mid-June, marking a significant milestone in the quest to establish comprehensive AI regulations. The joint adoption of the AI Act by the Parliament’s Civil Liberties and Internal Market committees, with a substantial majority, underscores the importance and urgency of this legislation.
Once the plenary adoption takes place, tentatively scheduled for 14 June, the proposal will enter the final phase of the legislative process. This phase involves engaging in negotiations, known as trilogues, with the EU Council and Commission. These trilogues aim to refine and finalize the legislation, ensuring that it aligns with the diverse perspectives and interests of all stakeholders.
Brando Benifei, one of the co-rapporteurs responsible for the AI Act, expressed enthusiasm about the imminent implementation of this legislation, stating, “We are on the verge of building a real landmark legislation for the digital landscape, not only for Europe but also for the entire world.” Such optimism reflects the profound impact that the AI Act is poised to have on shaping the global AI landscape.
At the heart of the legislation lies the critical task of defining AI itself, which establishes the boundaries and scope of its regulation. Conservative MEPs successfully aligned the AI Act’s definition of AI with that of the Organisation for Economic Cooperation and Development (OECD), a group comprising 38 affluent nations.
According to the legislation, an “Artificial intelligence system” (AI system) refers to a machine-based system designed to operate autonomously, generating outputs such as predictions, recommendations, or decisions that influence physical or virtual environments.
It is worth noting that the OECD is already contemplating potential revisions to its AI definition. Consequently, EU lawmakers have taken proactive measures to anticipate future changes in the OECD’s wording, ensuring the AI Act remains relevant and adaptable.
The AI Act also explicitly prohibits certain practices deemed to pose an unacceptable risk. Through the advocacy of left-to-center MEPs, the list of prohibited practices has been significantly expanded. Notably, the legislation bans the use of AI models for biometric categorization, predictive policing, facial image scraping for database construction, and the employment of emotion recognition software in law enforcement, border management, workplace, and education.
The inclusion of biometric identification systems, initially allowed in specific circumstances such as kidnapping or terrorist attacks, was subject to intense debate. Despite resistance from the conservative European People’s Party, the Parliament ultimately secured a majority in favor of a complete ban.
Recognizing the rapid development of AI systems without specific purposes, exemplified by ChatGPT and similar advanced language models, EU lawmakers faced the challenge of effectively regulating these general-purpose AI (GPAI) systems. The resulting approach entails a tiered framework within the AI Act. General-purpose AI systems will not be automatically covered by the legislation. Instead, the primary responsibility for compliance with obligations will fall upon the economic operators integrating these systems into high-risk applications.
However, GPAI providers must support downstream operators’ compliance by providing comprehensive information and documentation on the AI model. More stringent requirements have been proposed for foundation models and powerful GPAI systems like Stable Diffusion that can empower other AI applications. Independent experts will review aspects such as risk management, data governance, and the robustness of the foundation model.
Finally, generative AI models such as ChatGPT occupy the highest tier of the regulatory framework. These models will be obligated to disclose when a text is AI-generated and provide a detailed summary of the training data protected by copyright law. This transparency aims to foster trust and accountability in the utilization of AI technologies.
High-risk AI providers will face more stringent and detailed obligations under the European Parliament’s revised text. These obligations primarily focus on risk management, data governance, technical documentation, and record-keeping. By setting clear guidelines and requirements in these areas, the legislation aims to ensure responsible and accountable practices in the deployment of high-risk AI systems.
Furthermore, a novel requirement has been introduced for users of high-risk AI solutions. They are now obligated to conduct a fundamental rights impact assessment, taking into consideration factors such as the potential negative effects on marginalized groups and the environment. This assessment emphasizes the importance of safeguarding fundamental rights and preventing any discriminatory or harmful outcomes arising from the use of AI technologies.
In terms of governance and enforcement, EU lawmakers have reached a consensus on the need for a centralized enforcement architecture, particularly for cross-border cases. Co-rapporteur Dragoș Tudorache proposed the establishment of an AI Office, which was envisioned as a new body with a level of authority akin to that of an EU agency.
However, due to budgetary constraints and limited flexibility, the tasks assigned to the AI Office have been significantly reduced during the negotiations. Consequently, the AI Office will primarily assume a supporting role, offering guidance and facilitating coordinated investigations.
In contrast, the Commission has been entrusted with the responsibility of resolving disputes among national authorities concerning dangerous AI systems. This role ensures a unified and effective approach to handling conflicts and maintaining consistency in the enforcement of AI regulations across the European Union.
Conlcusion:
The approval of the AI Act by the European Parliament’s leading committees represents a significant step toward comprehensive regulation of Artificial Intelligence (AI). This landmark legislation will have profound implications for the AI market. With clear guidelines and obligations for high-risk AI providers, the Act establishes a framework that promotes responsible and accountable practices. The ban on certain practices and the emphasis on fundamental rights impact assessments reflect a commitment to safeguarding individuals and marginalized groups.
Additionally, the alignment of the AI Act’s definition with the OECD ensures harmonization with international standards. As the AI market adapts to these regulations, businesses will need to navigate stricter compliance requirements, potentially leading to more robust risk management and data governance practices. Overall, the AI Act sets a new standard for the ethical and responsible deployment of AI technologies, instilling trust and confidence in the market while protecting individuals and society at large.
