Zooming in on AI: Understanding the EU AI Act and Its Obligations for Limited-Risk AI Systems
The EU Artificial Intelligence Act (AI Act) represents a significant step in the regulation of artificial intelligence technologies within the European Union. This comprehensive framework adopts a risk-based approach, classifying AI systems based on their potential hazards and establishing corresponding obligations for developers and deployers.
Overview of the EU AI Act
The AI Act is distinguished by its detailed categorization of AI systems into various risk levels. Each classification is accompanied by specific requirements aimed at mitigating the risks posed by these technologies. This article focuses on the obligations related to limited-risk AI systems, which are subject to lighter regulatory measures compared to their high-risk counterparts.
Defining Limited-Risk AI Systems
Limited-risk AI systems are those that may interact with individuals or generate content, carrying risks of impersonation or deception. According to the AI Act, these systems are unlikely to cause significant harm or infringe upon fundamental rights.
Examples of limited-risk AI systems include:
- AI systems that interact directly with users, such as chatbots and digital assistants.
- Systems that create synthesized content including audio, images, video, or text.
- Technologies that generate or alter media content, leading to deep fakes.
- AI systems designed for emotion recognition or biometric categorization.
Transparency Obligations for Limited-Risk AI Systems
To ensure user awareness and safety, the AI Act mandates distinct transparency obligations for both providers and deployers of limited-risk AI systems.
Provider Obligations
As outlined in Articles 50(1) and 50(2), providers must:
- Chatbots and Digital Assistants: Ensure that users are informed they are interacting with an AI system unless it is evident to a reasonably informed person.
- AI-Generated Synthetic Content: Mark outputs from AI systems in a machine-readable format, ensuring they are identifiable as artificially created or manipulated. Providers must ensure that these solutions are:
- Effective: Solutions must consistently mark and detect synthetic content.
- Interoperable: They must function seamlessly across various platforms.
- Robust: Solutions should maintain effectiveness over time.
These obligations do not apply to AI systems that assist in standard editing functions or do not significantly alter the original data provided.
Deployer Obligations
Articles 50(3) and 50(4) specify obligations for deployers, including:
- Deep Fakes: Disclose when content has been artificially generated or manipulated, especially when the content resembles real individuals or events.
- AI-Generated Text: Notify users when text intended for public dissemination has been artificially produced or altered.
- Emotion Recognition Systems: Inform individuals about the operation of these systems and ensure compliance with the EU General Data Protection Regulation (GDPR).
Timing and Format of Notices
Information regarding limited-risk AI systems must be communicated clearly during the user’s first interaction. Particular attention should be given to vulnerable groups, ensuring that notifications are accessible to individuals with disabilities.
The Role of the European Commission
The European Commission will review the classification of limited-risk AI systems every four years and facilitate the development of guidelines to implement detection and labeling requirements for generated content.
Interplay with Other Regulations
The transparency requirements under the AI Act complement those established by the GDPR and the Digital Services Act (DSA). Providers of large online platforms must also identify and mitigate risks associated with the dissemination of AI-generated content, ensuring compliance with both sets of regulations.
Enforcement and Compliance
Compliance with transparency obligations is overseen by national authorities, with penalties for non-compliance reaching administrative fines of up to EUR 15 million or 3% of the operator’s total annual turnover.
Conclusion
The EU AI Act marks a pivotal moment in the regulatory landscape of artificial intelligence, particularly concerning limited-risk AI systems. By establishing clear obligations for both providers and deployers, the Act aims to foster transparency, enhance user safety, and promote responsible AI practices across Europe.
