Understanding the Risks of Artificial Intelligence as a Precondition for Sound Liability Regulation

Abstract

Not all AI risks are new. Risk of traffic accidents that self-driving cars generate is already reality in today’s traffic. Physical injuries a patient may suffer during a medical treatment occur regardless of whether the damage is caused by an autonomous agent or a human doctor. Modern societies are already familiar with the aforementioned risks. This chapter explores whether liability regimes, traditionally designed to deter physical risks and compensate an injured person when they occur, have rules apt for tackling the social risks that AI represents. In the European Union, the European Parliament has adopted a text of a Regulation on AI liability. The text is a clear step forward in adjusting liability rules to the challenges of AI. It sets out a position on who should be responsible and on what basis and provides injured persons with procedural devices in order to enhance their position and tackle the black-box issue. It, thus, for better or worse, deals with well-known fundamental issues surrounding AI liability. However, while social risks have been previously recognized by the European Commission in the White Paper and by some scholars, the adopted text omits to address them specifically. This chapter presents the nature of AI risks that liability rules should regulate. It seeks to address whether the traditional liability concepts are apt for regulating the novel types of risks. Just like in the case of safety regulation, this chapter attempts to demonstrate that a proper understanding of AI risks is a basis for sound regulation of liability.