Artificial Intelligence, Administrative Law and Financial Regulation
Artificial Intelligence, Administrative Law and Financial Regulation
In this conference paper, Dr Will Bateman presented a technically-embedded analysis of doctrinal legal issue which arise in the use of artificial intelligence (AI) by regulators, government administrators and other legal actors. The paper was delivered to the collected Justices of the Supreme Court of New South Wales, with special guest Justices from the High Court of Australia and the Supreme Court of the United Kingdom.The paper commenced by providing a short history of the development of binary computers (hardware and software) and their application in government, industry and academia. Computer science and AI grew throughout the 20th century as a response to the administrative burden and economic cost of administering government, large industrial enterprises and financialised western economies. The principal benefits of automation and algorithmic decision making in those fields are reduction in production costs and minimizing human error (or malpractice). Despite those advantages, the use of AI raises difficult issues when it is deployed in social domains which are governed by legal rules that assume the existence of a particular type of decision-making agent.Buying and selling securities through algorithmic trading involves exercising legal powers conferred by the law of contract and property. Those areas of law, along with the law of trusts, will apply to automated processes used by asset managers for distributing dividends or financial information to beneficiaries. Statutory law hovers over virtually all legal powers, but has particular prominence when government is acting in a service-provision role, such as paying welfare benefits under a statutory scheme.It is radically unclear whether algorithmic decision-systems (whether deterministic or probabilistic in nature) match the type of decision making assumed by those legal regimes. Most legal regimes assume that decision-making agents exercising legal powers are capable of grasping the immense semantic complexity of legal texts and the complex institutional frameworks (including parliaments and judiciaries) in which those texts are given meaning. They also assume that legal agents can respond to highly dynamic external environments. The paper concluded by highlighting the challenges presented to computer scientists in building algorithmic systems which fit those fundamental assumptions of most legal systems.