The law and economics of AI liability
Introduction:
The use of AI systems poses challenges to liability rules. This paper identifies these challenges and assesses how liability rules should be adapted to address them.
This paper discusses the responsibility gap that arises when AI systems act unpredictably or (semi-)autonomously. It takes into account when errors occur in AI systems.
Problems of proving failure and causality when monitoring responsibilities are foreseeable and difficult to define for producers and users. From an economic perspective, this paper considers which liability rules can minimize the cost of damage associated with AI.
Based on an analysis of risk and best liability rules, this paper evaluates the recently published EU proposals for the Product Liability Directive and the Artificial Intelligence Liability Directive.
One: safety regulations
Within the EU, product safety rules are largely ex ante and are divided into two levels of legislation. On the one hand, as described below, some specific laws regulate certain sectors or products.
On the other hand, in the absence of such specific requirements, General Rule 3 set forth in the General Product Safety Directive (hereinafter referred to as "GPSD") applies. The directive aims to ensure that manufacturers only place safe consumer goods on the market.
In particular, the GPSD requires manufacturers and distributors to provide safe products to consumers, to take all possible steps to identify any hazards of their products, to inform consumers of the existence of such hazards, and, if necessary, to remove dangerous products from the market.
Following a review of the GPSD, the committee recommended that the GPSD be converted into regulation and issue a proposal for the General Product Safety Regulation ("GPSR").
The proposed GPSR aims to regulate the security risks posed by new technologies. In addition to the GPSD or soon-to-be-become the GPSR, the Market Surveillance Regulations also set out requirements for certification and market surveillance.
Two: Liability Rules
In most Member States, fault-based liability is the general standard. When claiming damages under the fault liability regime, the claimant needs to prove that the infringer was at fault.
They suffered harm, and there is a causal link between the harmful activity and the damage. However, proving these conditions can be challenging for the injured party.
That is why other, more claimant-friendly liability regimes have been introduced for specific situations where legislators aim to allow easier access to damages.
One way to facilitate the filing of damage claims is to insist on a fault-based liability regime while reversing the burden of proof. A rebuttable presumption of fault or causation can help the claimant obtain compensation.
and reduce the information asymmetry between the injured party and the infringer. The system of presumption may be relevant to a variety of factual circumstances giving rise to different types of risk and damage.
Examples include the liability of parents for damage caused to their children, the liability of employers to employees acting on their behalf, owners of buildings, or people engaged in hazardous activities.
Three: Product Liability Directive and its proposed amendments
Over the past few years, the European Commission has identified several issues with the application of PLD provisions in the context of digital, connected and autonomous systems. One challenge is the complex product or service value chain, where suppliers, manufacturers and third parties are interdependent.
Another issue revolves around the uncertainty of the legal nature of digital goods, namely that they are products or services. Autonomous technology poses specific issues for product liability. Both the expert group report and the AI white paper concluded.
Some key concepts in PLDs need to be clarified in response to emerging digital technologies. The European Parliament also called on the European Commission to "review the directive and consider adjusting the concepts of 'product', 'damage' and 'defect', as well as adjusting the rules governing the burden of proof".
Four: Risks associated with artificial intelligence
AI can be complex due to the involvement of multiple stakeholders and the interdependence of AI components. Parts of digital goods, such as hardware and digital content, can be sold separately and produced by multiple parties.
This can make it difficult to trace the source of the failure or attribute responsibility for the failure to a single manufacturer. The injured party may face a hardware manufacturer, software designer, software developer, facility owner, or others.
It can be difficult for consumers to justify why their product isn't working. Artificial intelligence characterized by autonomy, unpredictability, complexity, or opacity.
It also brings new risks and challenges to the attribution of responsibility. In the light of these challenges, three possible gaps in existing liability regimes could be identified.
Conclusion:
As AI technologies make their way into everyday products and services, they are bound to play a role in liability litigation as well. This raises the question of whether liability rules are appropriate to deal with AI.
This paper identifies several possible gaps in liability rules, analyzes effective liability regimes for AI, and evaluates recent EU proposals on producer and operator liability.
EU non-contractual liability rules should not be considered in isolation, but should be seen as part of a broader rule, as together they shape the incentives of the parties.
In particular, liability rules need to be aligned with EU general regulations and sector-specific safety regulations, mandatory or self-regulated certification schemes, and national non-contractual and contractual liability rules and insurance rules.
Bibliography:
1. Commission, "Building a European Data Economy" (Communications) COM (2017) 09.
2.COM (2020) 65 finals; Commission, "Report on the Safety and Accountability Impact of Artificial Intelligence, Internet of Things and Robotics" COM (2020) 64 final; Responsibility and Responsibility of the Group of Experts on New Technologies for the Formation of New Technologies, Artificial Intelligence and Other Emerging Digital Technologies (2019) (hereinafter referred to as the "Panel Report"), 27-28.
4. European Parliament, Resolution on Automated Decision-Making Processes: Ensuring Consumer Protection and Free Movement of Goods and Services (2019/2915 (RSP)) (EP Resolution)
5. Commission, "On Artificial Intelligence – A. Europe's Approach to Excellence and Trust" (White Paper) COM (2020) 65 final, 17. This is also known as the "black box effect" of AI (see COM (2020) 65 final1)
