This paper presents new and alternative avenues for regulating AI, understood in the broad sense of state intervention that directly or indirectly affects how AI is developed and deployed. In doing so, it also identifies two new regulatory axes to be addressed - Big/Small AI Impact, and Big/Small AI Teams - and the challenges associated with designing tools that are effective and tailored to these axes. Against this background, our proposals in this paper thus fall into two broad categories: 1. Forms of strict liability, as a means to incentivise operators of Small AI systems (where direct regulation is unlikely to be effective, and often unnecessary) to take steps preventing or at least minimising harm resulting from their AI. 2. A mechanism of constant monitoring through random sampling with human testing and determination, as well as associated cumulative “penalty points”, to control the impact of Big AI systems on individuals and society before they cause large-scale, and in particular systematic, harm to society. Our proposals aim to have the maximal possible impact on the behaviour of AI operators and on where and how AI is used, so as to prevent harm and encourage responsible development and use of AI. In our view, these proposals are implementable, policeable, enforceable, cost-effective, and technologically-neutral; they do not focus on what the AI is, but rather what it does. We consider different legal tools and mechanisms for implementing our proposals, and conclude that while elements of existing tort liability regimes, platform regulation as well as the EU AI Act can partially be relied on, law makers should consider adopting more tailored rules on both liability and constant monitoring.
Ruse-Khan et al. (Thu,) studied this question.