Tort Law and AI: Principles & Practice

What is Tort Law?

A "tort" is distinct from a crime or a breach of contract.

• A crime is a wrong against the state, prosecuted by the state, with the goal of punishment.
• A breach of contract is the failure to fulfill a private agreement, with the goal of enforcing the promise.
• A tort is a breach of a general duty owed to society (e.g., the duty not to be careless), with the goal of compensating the victim for their losses, typically through monetary damages.

Tort liability can be organized into three main categories, based on the nature of the defendant's conduct.

As the video explains, the three categories are:

1. Intentional Torts: The defendant desired to cause the harm or knew with substantial certainty that it would occur (e.g., battery, defamation).
2. Negligence: The defendant acted carelessly, causing unintentional harm. This is the most common type of tort.
3. Strict Liability: The defendant is liable for harm even if they were not at fault. This applies in specific situations, such as abnormally dangerous activities or selling defective products.

While intentional torts could apply to AI (e.g., programming an AI for a malicious purpose), most legal questions surrounding AI failures fall under negligence and strict liability. Let's explore these in detail.

Negligence: The "Reasonable Person" Standard

Negligence is the workhorse of tort law. To succeed in a negligence claim, a plaintiff must prove four elements.

Let's break down these elements, which form the prima facie case for negligence.

The Four Elements of Negligence

1. Duty of Care: The defendant must have owed a legal duty to the plaintiff to act with a certain level of care. As the Palsgraf case illustrates, the dominant view (Cardozo's) is that this duty extends only to those within the "zone of foreseeable danger." An AI developer clearly owes a duty to the intended user, but do they owe a duty to a bystander harmed in a freak accident? The scope of this duty is a critical legal question.

2. Breach of Duty: The defendant's conduct must have fallen below the standard of care. This standard is typically that of a "reasonable person" under the circumstances. In the context of AI, this becomes the "reasonable AI developer," "reasonable data scientist," or "reasonable doctor using an AI tool." Proving a breach would involve showing that the defendant's actions (e.g., using a biased dataset, failing to test adequately) were not what a reasonably competent professional in their field would have done.

3. Causation: The defendant's breach must be the legal cause of the plaintiff's harm. This is a two-part test:

   • Factual Causation ("But-For" Test): "But for" the defendant's careless act, would the plaintiff have been injured?
   • Legal Causation (Proximate Cause): Was the harm a foreseeable result of the defendant's act? This element is designed to cut off liability for bizarre, unforeseeable chains of events. Notice how Justice Andrews in Palsgraf preferred to analyze the foreseeability of the victim here, rather than under the 'duty' element.

4. Damages: The plaintiff must have suffered a legally recognized injury (e.g., physical injury, property damage).

Strict Liability: Liability Without Fault

In some cases, the law imposes liability even if the defendant was as careful as possible. This is called strict liability.

The two most relevant applications for AI are:

1. Abnormally Dangerous Activities: Activities that involve a high degree of risk of serious harm, which cannot be eliminated by exercising reasonable care (e.g., blasting with dynamite). There is a legal debate about whether deploying certain types of highly autonomous AI could be considered an abnormally dangerous activity.
2. Product Liability: A manufacturer or seller of a product is strictly liable for harm caused by a product that is sold in a defective condition. A product can be defective in its manufacture, design, or because of a failure to warn of non-obvious dangers.

This is a major avenue for AI litigation. If an AI system is considered a "product," its developer could be strictly liable for harms caused by its "defective" design—for example, an algorithm that is inherently biased or unsafe.

The choice between a fault-based (negligence) and no-fault (strict liability) regime is one of the most significant policy questions in AI governance.

The Challenge: Proving the Case Against an Opaque AI

We now have the legal frameworks: negligence and strict liability. But applying them to AI presents a profound practical challenge: opacity, or the "black box" problem.

How can a plaintiff prove that a developer was negligent or that an AI was defective when it may be impossible, even for the developer, to explain exactly why the AI made a specific decision? Your technical background in data analytics gives you a head start in grasping the depth of this issue, which stems from the non-linear and statistical nature of complex models.

This challenge of "explainability" is where law and computer science collide.

As Fraser et al. argue, proving a case requires explanations that are specifically tailored to the legal elements. For example, to prove causation in a negligence case, a plaintiff needs to show that 'but for' the developer's negligence, the harm would not have occurred. As the paper suggests (and explores in later sections you can skim if interested, like Section 6), one could theoretically use counterfactual XAI methods to do this—by demonstrating that had the developer taken a reasonable precaution (like using a more balanced dataset), the model's error rate would have been significantly lower, and the accident would have been avoided.

Case Studies in Practice

Let's see how these principles play out in plausible scenarios, drawn from the AIKC report.

• Case 1: Medical Negligence (Resource LINK, Section 4)
  A doctor uses an AI diagnostic tool that misdiagnoses a patient, leading to harmful treatment. The AI was trained on a dataset that under-represented the patient's demographic.

  • Negligence: A case could be made against the doctor for medical negligence for relying on the AI without seeking a second opinion, breaching the standard of care for a reasonable physician.
  • Product Liability: A case could be made against the AI developer under strict product liability, arguing the AI tool was a "defective product" due to its biased design (the unrepresentative training data).

• Case 2: Autonomous Vehicle Collision (Resource LINK, Section 5)
  An autonomous vehicle with a malfunctioning sensor causes a minor collision.

  • Product Liability: The owner could sue the vehicle manufacturer for providing a defective product. The manufacturer might, in turn, sue the sensor's component manufacturer, illustrating the complexity of liability in modern supply chains (the "component part" doctrine).

These examples show that tort law does not provide one simple answer. Instead, it offers a set of tools to analyze fault and attribute responsibility, even in a world of complex, interdependent technological systems.

Conclusion

This lesson provided your first formal introduction to Tort Law, a critical pillar of civil liability. We've seen how centuries-old principles, forged through judicial precedent, are now being adapted to address the novel challenges of artificial intelligence.

Key Takeaways:

• Tort law allows individuals to claim compensation for harms caused by others' civil wrongs.
• Negligence is the primary framework for harms caused by carelessness. It requires the plaintiff to prove four elements: duty, breach, causation, and damages. The foreseeability of harm and the "reasonable person" standard are central to the analysis.
• Strict Liability imposes liability without fault, most notably for defective products. This is a powerful tool for holding manufacturers accountable and is a focal point of the debate on AI liability.
• The greatest practical challenge in applying tort law to AI is the opacity or "black box" nature of complex models, which makes proving elements like breach and causation incredibly difficult. This has spurred the field of Explainable AI (XAI) as a potential bridge between technology and legal proof.

Preview of the Next Lesson:

We have explored the duties we owe to people generally (torts). In our next lesson, we will turn to Contract Law, which governs the duties we voluntarily assume through agreements. This is essential for understanding the legal relationships defined by terms of service for AI platforms, API agreements, and the burgeoning world of automated "smart" contracts.