Artificial intelligence can recommend medical treatments, operate vehicles, evaluate applicants, generate professional work, and make decisions that affect everyday life. When one of those systems causes harm, however, the technology cannot compensate an injured person, explain its reasoning in court, or assume legal responsibility.

Someone must still answer for what happened. The difficult question is who.

New Technology Meets Established Law

In “Holding AI Accountable: Addressing AI-Related Harms Through Existing Tort Doctrines,” Professor Anat Lior examines whether traditional tort doctrines can address injuries involving artificial intelligence.

Tort law has confronted technological change before. Industrial machinery, automobiles, aircraft, pharmaceuticals, and the internet all introduced risks that earlier courts could not have anticipated. The law responded over time, sometimes by adapting existing principles and sometimes by developing new rules.

AI presents another test of that adaptability. Its complexity may make it difficult to understand how a particular output was produced. Developers may not have specifically programmed the decision that caused the injury, while users and organizations may rely on systems they cannot fully explain.

Does that uncertainty place AI outside established law, or does it make traditional legal reasoning more important?

The Problem of Causation

A person bringing a negligence claim ordinarily must connect the defendant’s conduct to the alleged injury. That connection includes actual cause and proximate cause.

AI can complicate both inquiries. A harmful outcome might involve a developer, model provider, software vendor, organization, operator, user, and data supplier. The system itself may also change through training, interaction, or deployment. Identifying a single responsible act can therefore become difficult.

Foreseeability presents another challenge. A company may argue that it could not have predicted the precise output produced by a complex model. Yet the inability to predict the exact sequence of events does not necessarily mean that the general category of harm was unforeseeable.

If an organization knows that a system can produce inaccurate, discriminatory, dangerous, or otherwise harmful results, how specific must the predicted injury be before the law recognizes it as foreseeable?

That question is not unique to AI. Courts have long considered whether an injury falls within the general risks created by a person’s conduct, even when the exact manner in which it occurred was unexpected.

Responsibility Behind the System

AI is sometimes described as though it were an independent actor. Legally, however, an AI system has no assets, insurance, or ordinary capacity to accept responsibility. It exists because people and organizations created, trained, distributed, selected, and deployed it.

Lior considers whether doctrines such as respondeat superior could help identify the human or organizational principal behind an AI agent. The answer may depend on who exercised meaningful control through design, monitoring, supervision, training, or deployment.

That inquiry may produce different answers in different circumstances. During development, responsibility might rest more heavily with a designer or manufacturer. After deployment, an operator or organization using the system may have greater control over how it is applied. In some cases, several parties may contribute to the same harmful outcome.

Market-share liability presents another possibility when an injured person cannot identify the specific producer responsible for a harm. Courts have historically treated that doctrine cautiously, but increasingly similar AI models, concentrated technology markets, and injuries that may emerge only after time could renew questions about its application.

Law, Technology, and Society

The relationship among law, technology, and society does not move in only one direction.

Society influences which technologies are developed and which risks are accepted. Law affects how those technologies are designed, insured, marketed, and deployed. Technology then changes social behavior and exposes weaknesses in existing legal rules. Those changes lead society to ask the law to respond again.

AI may therefore become part of a continuing feedback loop:

Society shapes law. Law shapes technology. Technology changes society. Society asks law to respond.

Existing tort doctrines may prove flexible enough to address many AI-related harms. Others may require adjustment as courts gain experience with the technology. Liability insurance may also influence conduct by establishing conditions, safeguards, and financial consequences before an injury occurs.

The central issue may not be whether an artificial intelligence system can be treated as a responsible legal actor. The more immediate question is whether the people and institutions surrounding it can remain accountable when its operations become difficult to explain.

AI may produce outcomes that no person specifically directed. It does not follow that those outcomes emerged without human decisions, commercial choices, or organizational responsibility.

As artificial intelligence becomes more deeply embedded in society, the law will have to decide how far existing ideas of causation, foreseeability, agency, and responsibility can reach. At the same time, the way courts answer those questions may influence what kinds of AI society ultimately receives.

Reference

Lior, A. (2024, November 18). Holding AI Accountable: Addressing AI-Related Harms Through Existing Tort Doctrines. The University of Chicago Law Review Online.