EPRS-人工智能的民事责任制度（英文）-2020.9-232页_3mb

Summary of the Document: Civil Liability Regime for Artificial Intelligence

Core Content

This document, prepared by the European Parliamentary Research Service (EPRS), provides an analysis of the civil liability regime for artificial intelligence (AI) systems in the European Union (EU). It outlines the potential economic and social benefits of establishing a unified EU civil liability framework for AI, while also highlighting the current regulatory gaps and the need for coordinated EU action.

Main Socio-Economic Functions of Civil Liability Rules in the AI Context

The civil liability regime for AI is essential due to its significant socio-economic functions:

• Incentive for Rational Use of Resources and Fair Risk Distribution: Liability rules encourage responsible development and use of AI by ensuring that risks are distributed fairly among all actors involved in AI systems.
• Encouragement of Innovation and Reduction of Uncertainty Costs: Clear liability rules reduce the uncertainty surrounding legal processes, which can encourage investment in innovation and research.
• Enhancement of Consumer Trust and Acceleration of AI Diffusion: A coherent liability regime enhances consumer confidence and facilitates the adoption and use of AI across industries.

Current EU Liability Framework and Its Application to AI

The EU does not currently have a specific civil liability regime for AI. Instead, it relies on the EU Product Liability Directive (PLD), which is a partially harmonised system. The PLD establishes a no-fault liability regime for defective products, including AI systems, but its applicability to AI is not clearly defined. The directive covers damage caused by defective products, but the extent to which it applies to AI remains uncertain.

The PLD provides a legal basis for liability in cases where AI systems cause harm, but there are unresolved issues regarding the scope of liability, damages calculation, and exceptions. These ambiguities create legal uncertainty and may hinder the safe and efficient development and deployment of AI.

National Liability Regimes and Their Variability

A comparative legal analysis of 19 EU Member States shows significant divergence in national liability regimes. National laws vary in their approach to contractual and non-contractual liability, fault-based vs. strict liability, and damages calculation. Some countries apply strict liability for things, dangerous activities, or damages caused by animals, while others have more flexible systems.

The lack of a unified approach at the national level leads to regulatory fragmentation, which can create barriers to the internal market and increase costs for producers. This fragmentation also affects the predictability and fairness of liability outcomes for consumers and businesses.

Policy Options and Approaches

The document explores various policy options for regulating AI liability at the EU level:

• Strict Liability and Fault-Based Liability create different incentives for actors. Strict liability encourages producers to take more precautions, while fault-based liability may promote accountability.
• Insurance schemes, whether obligatory or voluntary, could support the liability regime and help manage risks.
• The European Parliament and the European Commission have both proposed legislative initiatives and policy reports that highlight the need for a unified approach to AI liability.

European Added Value Assessment

The quantitative assessment suggests that a unified EU civil liability regime for AI could generate substantial economic added value:

• By 2030, EU action on liability could generate €54.8 billion in added value through increased research and development (R&D) in AI.
• Considering broader impacts, such as reduced accidents and improved health and environmental outcomes, the added value could reach €498.3 billion.
• The assessment is based on a two-step model:
  1. Quantifying the net benefits from increased R&D investment.
  2. Measuring broader economic impacts, including accident reduction, health, and environmental benefits.

Conclusion

A clear and coherent EU civil liability regime for AI is necessary to:

• Reduce regulatory fragmentation and legal uncertainty.
• Enhance consumer trust and business confidence.
• Facilitate safe and efficient AI development.
• Promote innovation and competitiveness in the EU economy.

The study concludes that timely and coordinated EU action on AI liability is crucial to realising the full potential of AI technologies and ensuring fair risk distribution and economic benefits for the EU.

Key Information

• Economic Impact by 2030: €54.8 billion from R&D acceleration; €498.3 billion when broader impacts are considered.
• Current EU Framework: Based on the Product Liability Directive (PLD), which is partially harmonised.
• National Variability: 19 Member States have different liability regimes, leading to regulatory divergence.
• Policy Recommendations: A unified EU liability regime is necessary to ensure fairness, predictability, and innovation in the AI sector.

Authors and Contributors

• Main Author: Dr. Tatjana Evas, European Added Value Unit, EPRS.
• Annexes I and II: Prepared by lawyer-linguists from the Directorate for Legislative Acts, DG Presidency.
• Co-Authors: Philipp Reifenrath, Dr. Brecht Verkempinck, Stephanie Ridley, Raquel Valls, Andreea Puiu, Daniel Roescu, and others from various EU Member States.

Disclaimer and Copyright

• The document is intended for European Parliament members and staff.
• It is not an official position of the Parliament.
• Reproduction and translation are authorized for non-commercial purposes, provided the source is acknowledged.

References

• Market Data: AI market value is expected to grow from $27.23 billion (2019) to $266.92 billion (2027).
• Policy Studies: The analysis builds on the Cost of Non-Europe (2019) and European Added Value Assessment (2018) on autonomous vehicles.
• Legal Theories: The economic and social impacts of liability are discussed in the context of innovation incentives, safety standards, and public trust.