On the Transparency of Artificial Intelligence Algorithms from a Legal Perspective

Abstract

In the modern era of active practical development of artificial intelligence (AI), lawyers are facing the question — how to resolve the ‘black box’ problem, i.e. the incomprehensibility and unpredictability of decisions that artificial intelligence makes. Developing rules that maintain the transparency and comprehensibility of AI algorithms enables artificial intelligence to be incorporated into conventional legal frameworks, thereby eliminating the threat to the concept of legal liability. In private law, protecting consumers from major online platforms makes algorithm transparency a key issue, changing the obligation to provide information to consumers, which can now be described by the formula ‘to know + to understand’. Similarly, states are unable to adequately protect citizens from the harm caused by their dependence on algorithmic applications when public services are provided. The only way to counter this is through knowledge and understanding of how algorithms work. Fundamentally new regulations are required to bring the use of AI within a legal framework, which should include requirements for algorithm transparency. Experts are actively discussing the development of a regulatory framework to establish a system for observing, monitoring and provisionally authorising the use of AI technologies. Measures are being developed for an ‘algorithmic accountability policy’ and a ‘transparency through design’ framework, which address issues throughout AI development, with an emphasis on ongoing stakeholder engagement and organisational openness, as well as the implementation of explainable AI systems. Overall, the proposed approaches to regulating AI and ensuring transparency are quite similar ones, as are the predictions regarding the mitigating role of transparent AI algorithms in building trust in AI. Of interest is the concept of ‘algorithmic sovereignty,’ that refers to a democratic state’s ability to govern the development, deployment and impact of AI systems in accordance with its own legal, cultural and ethical norms. This model is designed to promote the harmonious coexistence of different states, which in turn leads to the harmonious coexistence of humanity and AI. Overall, although the use of AI differs ideologically in the private and public spheres, transparency of algorithms is equally important and ultimately increases the likelihood of regulation.