The Law of AI Agents

Al labs may soon develop Al agents capable of working throughout the economy. This talk examines how we can use the law to govern such agents. First, Al agents will inevitably have accidents. Who should be held liable? We argue that some liability should be placed on Al agents themselves. Second, what kinds of legal rights and duties should be assigned to Al agents more generally? We approach these questions through the function of legal rights and duties in facilitating cooperation and economic growth. We tentatively argue that Al agents should be given property, contract, and tort rights. We also consider the case for political rights, such as suffrage. Third, the governance of Al agents faces difficult questions about how to count Al agents: how many Al agents are involved in a cluster of computations? We propose a solution, "A-corps," which creates a series of corporations managed by Al agents. While Al agents are difficult to count precisely, the corporations are not. Legal rights and duties in the first instance should be assigned to A-corps.

Simon Goldstein is an associate professor at the University of Hong Kong.

His research concerns a range of topics in Al governance and Al safety, surrounding optimal legal and economic institutions for governing Al agents.