How can Artificial Intelligence be Defective?

Abstract

While it is not always clear how the Internet of things (IoT) and artificial intelligence (AI) should be defined precisely, it is a fact that the new tech‐ nologies associated with algorithms and the Internet attract an increasing amount of attention from lawyers, and rightly so. The issues that are raised are very diverse, but liability is definitely one of them. Who can be held liable for damage associated with the IoT or AI, and under what condi‐ tions? This is a central, and very complex question. This short contribution obviously will not attempt to give a complete and final answer to it, but will rather focus on the element that can or should trigger liability associ‐ ated with the IoT or AI. The element that triggers liability, and which French lawyers sugges‐ tively call fait générateur (literally: the generating fact, but the expression is often translated into English as the ‘event giving rise to liability’), de‐ pends on the liability regime that is being applied. It is often a human con‐ duct, as is the case when liability is based on fault, but it can also be a cer‐ tain type of event or a given characteristic in a thing, as is the case when liability is based on a product’s defect. A preliminary question is there‐ fore: what are the liability regimes which may or should apply in case of harm caused by, or in connection with the use of, the IoT or AI? Depending on the circumstances and on the legal systems, different regimes may be applicable. A general distinction can be made, however, between sector-specific regimes and non-sector-specific regimes. The for‐ mer are relevant every time that the IoT or AI is used in a field of activity covered by such a specific regime. For example, in many countries, if an autonomous car or vehicle is involved in a traffic accident, a specific (strict) liability regime designed to cover such accidents will apply1. Sec‐ tor-specific regimes will not be considered at this stage, since they come