Contracts and contracting have changed dramatically in the past fifty years. We have moved from negotiated paper contracts to standard form contracts to digital contracts presented in various ways. The next fifty years promises even more dramatic changes, and not just to the form of contracts. Technological innovation and marketplace needs will undoubtedly disrupt contracting in ways that don’t exist today. John Linarelli’s article Advanced Artificial Intelligence and Contract addresses one of the biggest anticipated disruptions – a not-quite human contracting party. In this article, Linarelli asks the provocative question, “How might contract law adapt to a situation in which at least one of the contract parties could, from the standpoint of a capacity to engage in promising and exchange, be an AGI?”
Linarelli states that artificial intelligence will bring about transformational changes in the law. He uses the term “artificial general intelligence” or “AGI” to refer to an advanced form of artificial intelligence which has a cognitive architecture of its own, unlike the artificial intelligence that currently exists. He invites us to consider the “feasibility of investing an AGI, from a legal point of view, with the power to enter into contracts, either with humans or other AGIs.”
By referring to AGI as a contracting party, Linarelli is not talking about as an agent for a human, nor is he talking about “smart contracts” which he states is “not a legal concept” and which is simply a substitution of “algorithms for human contract performance and enforcement.” In his words, “(t)he difference between contracting with or by an AGI and smart (or algorithmic) contracts is that human or legal persons in the form of entities such as corporations are the actual parties to smart contracts, whereas contracts with AGIs involve at least one contract party that is neither a human nor a currently recognized legal person such as a corporation, limited liability company, or partnership.” (P. 333-34.)
So should the law grant AGIs legal status with all the rights and obligations that such a designation confers? And how should contract law respond to AGIs as potential contracting parties?
The issues pertaining to the first question have been discussed before in the context of limited liability firms, such as corporations. As Linarelli notes, “(a)n artificial person with limited liability, however, still must operate through human agents” unlike the AGIs that he is anticipating which would operate independently of humans. Linarelli argues that AGIs should be granted legal status as contracting parties only when AGIs develop the cognitive architecture that humans possess. His rationale is simply that “(c)ontract law exists to meet human needs in human societies for voluntary exchange.” (P. 343.) Accordingly, AGIs should exhibit a “sufficient level of mutuality in terms of autonomy, interaction, and adaptability” before granting them contract rights and liabilities otherwise “cooperation with humans will fail.” (P. 343.) Thus, he makes a simple yet profound observation — that one of the primary societal benefits of contract law is to encourage cooperative activity — and cautions that to recognize a contracting party without the relevant human qualities poses serious risks to society: “Worse, an alien system of AGI values and cognitive abilities, based on norms humans do not understand and cannot reasonably accept, may be considered by humans to be harmful or pernicious or may cause harm to humans.” (P. 343.) (Of course, one could argue that by recognizing limited liability entities as contracting parties, we have done just that….) Linarelli notes, “It will be difficult for AGI to engage in contracting or to be subject to contract law if it cannot interact with humans since humans have evolved to interact and to be regulated by institutions that humans have constructed to reflect the intentions they hold in common.” (P. 342.) Thus, before we recognize AGIs as contracting agents, they should exhibit cognitive capabilities that are human-like as it will be “far more practical to align their cognition to contract law and values it represents than to change contract law and the humans who invented it.”
But are those values reflected in modern contract doctrine? While the “smarter-than-us” AI scenario is what tends to capture the public imagination and terrify entrepreneurs like Elon Musk, it is the second question – how should contract law respond to AGIs as a contracting party? — that is perhaps more intriguing to contracts scholars. Linarelli notes that the objective theory of contracts “coincides closely to the Turing test for assessing whether AI exists” and therefore, may “at least partly, and with some adaptation…answer the question of how to treat an AGI as a contract party.” (P. 334.) The objective of the Turing test is to determine whether a human interrogator can distinguish the answers given by a machine from those given by a human. The Turing test may be useful for “weak AI” but in order to qualify as “strong AI,” the machine must also “simulate thinking and intentions.” (P. 335.) Linarelli observes that contract law, however, would be satisfied with weak AI: “Weak AI is sufficient for the purposes of determining whether an AGI could be a party to a contract in terms of understanding the question as one that is internal to contract law…The Turing test has been effectively embedded into Anglo-American contract law in the objective theory of contract.” (P. 336.) In other words, contracts and contract law — capitalism’s vaunted tools of autonomy — require very little of contracting parties; they require only outward appearances regardless of mental states: “The objective theory of contract tells us that the intention to be bound to, or form, a contract is determined by evidence external to the actual intentions of the parties.” (P. 336.)
Linarelli isn’t the only scholar to point out the ways that contract law seems to permit – even encourage – automated, mechanical behavior from contracting parties. Notably, Brett Frischmann and Evan Selinger questioned the way that wrap contracts and in particular, clickwraps, have conditioned humans to act like machines. But Linarelli’s questions are directed more pointedly to contracts scholars who may not have considered the way technology changes contract law – and the way that contract law alters our expectations and standards for human and human-like behavior.