Introduction
The proliferation of artificial intelligence (AI) systems into commercial and consumer life represents a significant technological shift, with automated platforms now frequently involved in the negotiation and execution of contracts. This development poses questions for the traditional doctrines of English contract law, which were established in an era of human-to-human agreement. The central issue is whether these established principles, such as offer and acceptance, intention, and liability for error, remain fit for purpose in an age where autonomous systems can perform contractual functions. This essay will argue that whilst the existing legal framework can be adapted to address transactions involving simpler forms of AI by treating them as sophisticated tools, the increasing autonomy of advanced AI systems exposes fundamental weaknesses in the law. It will be contended that the orthodox legal concepts, particularly those concerning legal personality and intention, are strained by these new technologies, suggesting that future legislative intervention may be necessary to provide certainty.
The Application of Traditional Contract Formation Rules
The foundational elements of a binding contract in English law are offer, acceptance, consideration, and an intention to create legal relations. The initial question is how these principles apply when one or both parties are utilising an AI system. For many current forms of e-commerce, the law has found ways to accommodate automated processes. The courts have long dealt with transactions made through machines, as seen in the case of Thornton v Shoe Lane Parking Ltd [1971] 2 QB 163, where the operation of an automatic ticket machine was analysed within the offer and acceptance framework. Lord Denning MR concluded that the offer was made by the machine on behalf of its owner, which the customer accepted by inserting money. This analogy provides a basis for understanding many AI-driven transactions, where the AI is viewed not as a party to the contract, but as a medium or tool through which a human or corporate principal operates.
This approach is further supported by the Electronic Commerce (EC Directive) Regulations 2002, which provide a framework for online contracting. The regulations clarify aspects of the process, such as when an order is placed and accepted, which helps to integrate automated web services into the traditional offer and acceptance model. Generally, a website displaying goods is treated as an invitation to treat, and the customer’s order constitutes the offer, which the seller can then accept or reject. The AI system, in this context, is merely the mechanism for communicating the seller's pre-determined pricing and terms. Therefore, for non-autonomous or "weak" AI that follows a set of programmed instructions, it can be argued that the existing law is sufficiently flexible. The intention and the contractual promises are deemed to be those of the person or company that deployed the AI system.
The Problem of Intention and Legal Personality
A more significant challenge arises from the concept of intention to create legal relations, a cornerstone of contractual validity. English law recognises only two types of entities as possessing legal personality and thus the capacity to hold rights and obligations: natural persons (humans) and legal persons (such as corporations). Artificial intelligence does not fall into either of these categories. An AI, no matter how advanced, currently has the legal status of an object; a piece of property or a tool owned and operated by a person. Consequently, it is a legal impossibility for an AI itself to possess the intention to form a contract.
This means that any contractual intention must be attributed to the AI's user, owner, or controller. While this attribution is straightforward for simple automated systems that act on explicit instructions, it becomes problematic with the advent of more sophisticated, "strong" AI. These are systems capable of learning, adapting, and making decisions that were not specifically foreseen by their programmers. If such an AI negotiates and concludes a contract on terms that its owner would not have approved, the question of whether the owner had the requisite intention to be bound becomes complex. The law of agency might offer a partial solution, by treating the AI as an agent acting for a principal. However, the law of agency is also built on the premise that the agent is a legal person. Attributing agency to a machine is a legal fiction that the courts have not yet formally endorsed. While a workable solution in the short term is to hold the owner responsible for all actions of their AI, this approach may begin to appear unjust as the technology becomes truly autonomous, stretching the concept of intention to its breaking point.
AI and Contractual Mistakes
The issue of liability for errors made by an AI in the contracting process provides another area of difficulty. For example, if an AI system mistakenly offers a high-value product for a nominal sum, is the seller bound by this apparent offer? The law of mistake offers some guidance. The general principle is that a unilateral mistake as to the terms of an offer will not render a contract void unless the other party was aware, or ought to have been aware, of the mistake. This is often referred to as the 'snapping up' doctrine, as established in cases like Hartog v Colin & Shields [1939] 3 All ER 566.
This principle was applied to an electronic context in the Singaporean case of Chwee Kin Keong v Digilandmall.com Pte Ltd [2005] 1 SLR(R) 502, where a pricing error on a website led to printers being offered for a fraction of their actual cost. The court held that the buyers knew there was a mistake, and therefore the contracts were void. This reasoning could logically be extended to contracts made by an AI. If the AI makes a pricing error that is so obvious a reasonable person would have realised it was a mistake, a court would be unlikely to enforce the contract against the AI's owner. However, this relies on the error being apparent to the human counterparty. If a sophisticated AI negotiates a contract with another AI, and one of them makes a subtle but significant error, it is less clear how the doctrine of mistake would apply, especially in determining what each machine 'ought to have known'. This highlights that while existing doctrines provide a starting point, they were not designed for interactions devoid of human consciousness or intuition.
The Future: Autonomous AI and the Need for Reform
The greatest strain on the current law comes from the prospect of truly autonomous AI agents capable of operating independently in the marketplace. If an AI can learn from data, identify new commercial opportunities, and enter into contracts to exploit them without real-time human oversight, the "AI as a tool" analogy begins to fail. The Law Commission has begun to explore these issues, for example in its work on smart legal contracts, which acknowledges that the increasing complexity of automated systems requires legal clarity (Law Commission, 2021).
If an autonomous AI enters into a disastrous contract, attributing liability to the owner who may not have foreseen or authorised such an action raises questions of fairness. Conversely, absolving the owner of responsibility would leave the other contracting party without a remedy, creating unacceptable uncertainty in commerce. This has led some commentators to speculate on the need for new legal concepts, such as a form of limited legal personality for certain classes of AI, or specific statutory liability regimes that allocate risk between the programmer, the owner, and the user. While creating a third category of "electronic personality" is a radical step, the alternative may be to develop bespoke legislation. This might impose strict liability on the owner for the actions of their autonomous AI, or require mandatory insurance for those deploying such systems in commerce. Without such reform, the common law may be forced to either stretch existing principles to the point of incoherence or leave a significant regulatory gap.
Conclusion
In summary, the established principles of English contract law demonstrate a degree of flexibility in adapting to the rise of artificial intelligence. For the relatively simple automated systems that currently dominate e-commerce, doctrines developed for vending machines and online contracting allow the law to function by treating the AI as a tool of its human or corporate principal. Principles of offer and acceptance can be applied, and the law of mistake provides a remedy for obvious errors. However, this essay has argued that this approach has its limits. The core concepts of contractual capacity and intention, which are tied to legal personality, are not conceptually equipped to deal with AI systems that are more than mere tools. As AI becomes more autonomous, its ability to act in ways not directly intended by its owner poses a fundamental challenge that the common law may not be able to resolve satisfactorily on its own. While the judiciary may continue to adapt existing rules, it is contended that the growing complexity and independence of AI will ultimately necessitate clear legislative action to define the legal status of AI and allocate liability for its contractual actions, thereby ensuring certainty and fairness in the digital marketplace of the future.
References
- Chwee Kin Keong v Digilandmall.com Pte Ltd [2005] 1 SLR(R) 502.
- Hartog v Colin & Shields [1939] 3 All ER 566.
- Law Commission. (2021) <a href="https://www.lawcom.gov.uk/project/smart-legal-contracts/">Smart legal contracts: Advice to Government</a>. Law Com No 401.
- McKendrick, E. (2021) Contract Law. 14th edn. Red Globe Press.
- The Electronic Commerce (EC Directive) Regulations 2002, SI 2002/2013.
- Thornton v Shoe Lane Parking Ltd [1971] 2 QB 163.
