How the world is gearing up to handle harm caused by AI-powered products, and the path ahead for India
When it comes to product liability, the starting point of the discussion should be the 1932 "snail in the ginger beer bottle" case (Donoghue v Stevenson). The UK's House of Lords by a wafer-thin majority, laid the foundation of the law, which has since developed as an independent tort of negligence. As noted by Lord Denning in his Discipline of Law, the law laid in Donoghue (with its various extensions) has come to dominate the whole field of civil liability.
The House of Lords here held that a manufacturer must take care to avoid acts or omissions which can reasonably be foreseen as likely to cause injury. In a classic exposition of the law, Lord Atkin married law and morality: "The rule that you are to love your neighbour becomes in law: you must not injure your neighbour." A "neighbour" can be any person closely or directly likely to be affected: "You must take reasonable care to avoid acts or omissions which you can reasonably foresee would be likely to injure your neighbour."
Lord Macmillan added that the law is not concerned with carelessness in the abstract. The carelessness must have caused damage. Further, the grounds of action may be "as various and manifold as human errancy", and legal responsibility may develop in adaptation to altering social conditions and standards: "The categories of negligence are never closed."
Hence, liability in negligence is attracted when there is a duty of care situation; failure or breach of this duty; a causal connection between the careless conduct and the damage complained of and finally, the resulting damage should not be too remote or unforeseeable.
The law moves towards strict liability
While Donoghue v Stevenson became the foundation for an action in negligence, and there was a remarkable expansion of the principle to myriad situations, the liability was always rooted in fault (some careless or deliberate act or omission). It never expanded to a no-fault liability.
The law in the US took a different turn in 1963, with the landmark case of Greenman v Yuba (Supreme Court of California). Here, a hand power tool attachment flew off the machine and injured the plaintiff. The court held the manufacturer to be strictly liable: "A manufacturer is strictly liable in tort when an article he places on the market ... proves to have a defect that causes injury to a human being."
In other words, all the plaintiff needs to prove is that he was injured by a defective product used for its intended purpose. Liability is not to be governed by the law of contract warranties but by the law of torts. Greenman is seminal as it marks the beginning of a shift in the manufacturer's liability from "fault" to "no-fault".
Law of strict liability in UK
English law introduced the concept of strict product liability by the Consumer Protection Act, 1987. The act renders the producer liable where any damage is caused, wholly or partly, by a defect in the product. Certain exceptions apply. The most interesting (from the point of view of technology law) is that the state of knowledge at the relevant time was not such that the defect could reasonably be discovered.
Hence, the law acknowledges that a producer can only reasonably be expected to incorporate safeguards in line with the existing knowledge. In a sense, this reintroduces the element of negligence and fault, though in a limited way to defects which could not reasonably be safeguarded against, given the available state of knowledge.
Can a producer limit liability?
In common law, a person can contractually limit or remove liability for negligence, whether personal or vicarious (Charlesworth and Percy on Negligence, 14th ed, para 4-82). This, however, is subject to certain statutory limitations. The UK Unfair Contract Terms Act, 1977, negates any contractual term or notification which excludes or restricts liability for death or personal injury resulting from negligence. In all other cases, the exclusion or limitation clause must satisfy the test of reasonableness.
The Indian Consumer Protection Act was completely recast in 2019. The concept of no-fault liability has been introduced for any harm caused by a defective product or service. Any purported exclusion or limitation of liability can be challenged as an "unfair" term and must meet the test of reasonableness.
Limitations of Indian law
The Consumer Protection Act, 2019, is not ideal or even suitable in relation to technology-related disputes. This is due to a series of judgments holding that the consumer courts would not entertain cases involving complex facts or requiring expert evidence. The forum is meant for summary adjudication on the basis of uncomplicated facts. Any significant technology or AI-related dispute is not likely to be entertained by the consumer courts.
The civil courts, too, will be unequal to the task. They cannot simply adapt the no-fault liability principles crafted under the Consumer Protection Act, 2019, for the consumer courts. Under Indian law, strict liability can only be provided for by legislation. This was so held by the Supreme Court in Meenu Mehta's case (1977). Here, various high courts had held that "public good" required that anyone injured by a motor vehicle must get compensation, irrespective of the defendant's negligence or carelessness.
The Supreme Court struck this down, holding that the concept of owner's liability "without any negligence is opposed to the basic principles of law ... The proof of negligence remains the linchpin to recover compensation". It was in light of this that parliament was constrained to amend the Motor Vehicles Act in 1982 and introduce a no-fault liability compensation package for motor vehicle victims.
Given the factual complexity of technology disputes and the absence of a special regime, the normal civil courts are not likely to be suitable for AI-related disputes. (More on this later).
The evolution of product liability in AI realm
Considering the promise and potential impact of AI in all spheres of human life, regulatory bodies in developed jurisdictions have focused their attention on crafting an AI regime, including for civil liability for AI products.
The EU has taken the lead and made a concerted effort to formulate the relevant principles. This is in the shape of two directives: the AI Liability Directive and the Product Liability Directive (EU directives set out the principles and goals that the EU countries must aim to achieve and affords guidance to individual countries for devising their laws).
The EU AI Liability Directive sets out the reasons for formulating a special regime. It recognises that complexity of issues can make it difficult for a consumer to identify and establish the necessary constituents of liability. Expert evidence may not be readily available, and discovery run into confidentiality issues. It can become disproportionately costly to bring an action. Hence, the consumer needs protection.
Equally, the business also needs to be protected against an ad hoc approach or uncertainty, and consequent difficulties in procuring insurance. Public interest lies in legislative intervention so that the benefits of AI can be reaped.
Keeping these factors in mind, the EU AI Act was published on 21 May, 2024 after approval from all member-states and the EU parliament. Though the act has been published, the AI directives and product liability directive are presently in a draft form. The act classifies AI into four categories: prohibited AI; high-risk AI; chatbots and generative AI; and general-purpose AI.
Prohibited AI are those which cause an "unacceptable risk to individual fundamental rights" and are banned. High-risk AI is subject to compliance requirements. Chatbots and generative AI merely require compliance with certain transparency obligations. General-purpose AI products are subject to various specific requirements.
Civil liability
The EU AI Act requires proof of negligence and a finding of fault to fasten liability. Further, the fact of damages needs to be proved – (negligence without damages is not actionable). At the same time, it recognises the challenges a claimant may face because of the complexity of the AI system.
The act therefore creates a presumption of causality. For instance, if a claimant can show non-compliance with certain legal obligations relevant to the harm, or can demonstrate an act or an omission, the court will draw the causal link and presume that the lapse caused the damage complained of. The presumption is rebuttable. The act thus aims to strike a balance between the consumer and the business owner.
The path ahead for India
Indian policymakers are engaged in formulating an AI-related policy, but current efforts are in the form of broadly stated principles and guidelines. India needs to craft a sui generis policy, including in relation to civil liability for tech and AI-related disputes. The country should also consider a standalone technology court for all tech-related disputes. This would vastly facilitate India's rollout of AI and other tech-related initiatives and assist in the transition to a developed economy.
The author was assisted by associates Bhavana Chandak and Pratyush Khanna.
[This article was first published on 31st July 2024 by 'India Business Law Journal'.]
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.