Dr. Leonardo ROMANO

Thesis : “Approaching the Unknown: Regulating Artificial Intelligence and Filling Liability Gaps in the Financial Market Domain”

Artificial intelligence (AI) is everywhere and its development, deployment and use are moving forward rapidly and contributing to the global economy. One of the first domains where autonomous applications have taken off is in financial markets. Markets are well suited to automation, as they are essentially built on information, which requires algorithms to digest. The effects on markets of AI systems is at present unclear, however, in part because the factors at play are unprecedented, and there is ample reason for concerns that the financial system is vulnerable to AI agent’s misbehavior, whether accidental or purposeful, legal or illegal, ethical or unethical. This study focuses on unintentional unethical behavior in agents used to trade in financial markets. Although today’s autonomous agents operate within a relatively narrow scope of competence and autonomy, there is little doubt that trading agents will become increasingly capable of operating at wider levels of initiative without human oversight, and that regulation is now needed to prevent societal harm.

On these premises, potential impacts of AI on financial systems will be considered, with a view to understanding and mitigating AI threats to financial system and to building in accountability and control. The analysis will principally converge on three main points. As first, it will lay out some legal issues presented by the introduction of autonomous trading agents in financial markets, both specific to the financial domain and as a case study for autonomous agents in general. Then, it will examine the effects of autonomy on questions of lawful or ethical behavior that hinge at all on intent, focusing on questions about how to map ethical and legal concepts delineating acceptable trading behavior from the human to computational realm. Eventually, it will explore the possible legal solutions that might reconcile immediate autonomy with ultimate human authority. Finding effective solutions to the regulation of autonomous agents in tis domain is important in its own right, and may also prove illuminating for addressing the broader problem of AI control