Leading legal disruption

Establishing liability for damages caused by AI used to be rather straightforward when only one or few stakeholders are involved, or when the AI could only take a limited range of pre-defined decisions in accordance with specific parameters defined by a human programmer. However, AI usually involves several stakeholders and components (e.g. sensors and hardware, softwares and applications, data itself and data services, connectivity features) and recent forms of AI are increasingly able to learn without human supervision which makes it difficult to allocate liability between all stakeholders. This contributions maps various possibilities, identify their challenges and explore lines of thought to develop new solutions or close the gaps, the whole from a global perspective. Existing liability regimes already offer basic protection of victims, to the extent that specific characteristics of emerging technologies are taken into account. Consequently, instead of considering new liability principles (solutions that require certain amendments of the current liability regimes), one should consider simply adapting current fault-based liability regimes with enhanced duties of care and precisi