Civil responsibility for damage caused by artificial intelligence

Abstract

The aim of this thesis is to determine whether the legal frameworks currently available in Malta are sufficient to allocate responsibility and to cater for damage caused by artificial intelligence which acts autonomously from the wills of its creators, users and owners. The thesis starts off by examining the current technological scenarios as regards robots, expert systems and artificial intelligence. This shows how the development of forms of artificial intelligence poses critical questions as to who should be held liable for damage caused by the operation of these machines. The Product Liability regime is explored to determine whether this form of liability can be applied in cases where the AI causes damage to a third party through its conduct. Given that product liability can develop only once there exists a defect in the product, it follows that if the AI functions properly as was intended by its creators, product liability cannot subsist. Therefore, the thesis explores other possibilities for allocating liability through the application of various other paradigms of legal responsibility. The provisions concerning direct and indirect liability under tort law are examined in detail, and a parallel is drawn between persons who may be held liable for damage caused by AI and the responsibility placed on parents for damage caused by minors on owners/users of animals, on employers for acts of their employees and the French concept of liability of the custodian of the thing. Subsequently, the thesis also explores certain possible scenarios of how an AI could enter into contractual obligations. An analogy is drawn with the conduct of artificial agents and mandataries under Civil Law. It was determined that similarly to the obstacle encountered when trying to apply direct liability to an AI, an obstacle to the application of agency law to that AI is that an AI is not considered to be a person in the eyes of the law. Hence, the prospect of granting personhood to the AI itself is also investigated in detail. If the legislative framework remains unchanged, it is likely that most actions to recoup damage resulting from AI injuries are likely to fail. Therefore the thesis concludes that should AI not be granted personhood, legal amendments to our tort law need to be undertaken in order to introduce the concept of liability of the owner of a thing, which would adequately cater for injury caused by AI.