Research Context and Challenges

As artificial intelligence (AI), the Internet of Things (IoT), and distributed ledger technologies (DLT) become more prevalent, legal scholars and policymakers are debating how to address the potential for harm. A central fear is the existence of a 'legal vacuum' where victims of AI-related accidents cannot receive compensation because the technology is too complex or unpredictable to fit into traditional fault-based tort law. The author examines whether strict liability—a regime where a party is liable regardless of fault or defect—is the right tool to bridge these perceived gaps.

The Nature of AI Risks

The paper identifies two primary features of AI that challenge traditional liability: 'autonomy,' which makes software reactions to new situations unpredictable, and 'opacity,' or the 'black box' effect, which makes it nearly impossible to trace harmful behavior back to a specific coding error or human negligence. These challenges are compounded by the complexity and connectivity of IoT ecosystems, where multiple components and data feeds interact, making it difficult for victims to identify the source of a failure.

Proposed Liability Framework

The author argues that strict liability is justified for AI applications because it overcomes the evidentiary hurdles victims face when dealing with autonomous and opaque systems. However, the author cautions against applying this broadly. Instead, the paper proposes a risk-based approach:

• Scope: Strict liability should be limited to 'physical risks' in a broad sense, including death, personal injury, property damage, and damage to data or digital environments. It should not extend to immaterial harm or pure economic loss, which are too difficult to quantify and insure.
• Addressees: Liability should be assigned to the party with the most control over the risk. For AI, this is typically the 'backend operator' (who provides updates and defines system features) or the producer, rather than the 'frontend operator' (the user or owner), especially if the user is a consumer.
• Integration: Rather than creating a standalone, potentially disruptive AI liability regime, the author suggests that minor adjustments to existing frameworks, such as the Product Liability Directive, might be more effective and less likely to cause market fragmentation.

Key Terms and Definitions

• Strict Liability — A legal regime where a party is held liable for harm caused by their activity or product regardless of whether they were at fault, negligent, or if the product was defective.
• Autonomy — The capacity of an AI system to react to unseen instances or environments without direct human intervention, often leading to unpredictable behavior.
• Opacity — The 'black box' nature of sophisticated AI algorithms, which makes it difficult to explain why a system behaved in a specific way or to trace that behavior to a specific defect.
• Frontend Operator — The party who uses or deploys an AI device in a specific context, such as the owner of an autonomous vehicle, who controls the immediate risks of operation.
• Backend Operator — The party who continuously defines the features of an AI technology, provides essential support services, and exercises control over the system's algorithms through updates.