The Place of the Ethical System in the Architecture of Artificial Intelligence

The aim of the article is to analyze the existing views on the need to develop an ethical system for artificial intelligence and to identify methodological difficulties arising from the implementation of the idea of "ethical AI". The article presents arguments in favor of the relevance and importance of creating an ethical system or a set of rules of ethics and morality for an artificial agent. The stated ideas are based on the research of theorists and philosophers of AI, as well as on the results of the author's analysis of documents: current ethical codes of a number of companies and states (MIT Moral Machine, IBM Watson, Mercedes Benz autonomous cars' code, Code of Ethics for autonomous vehicles of the Government of Germany, analysis of the work of Tay and Zo chat bots from Microsoft). The article describes some methodological difficulties arising in the development of an ethical system for artificial intelligence. The complexity or impossibility of formalizing morality in a form accessible for processing by computer programs stand out among the difficulties. For example, it is challenging to describe morality in a quantitative format; put restrictions on the "purity" and amount of data for training the system-training on real data of people's behavior will lead to unpredictable results, since people do not tend to make decisions based on logic or rational arguments. There is a problem of data opacity or "black box"-the inability to trace the decision-making process of the neural network due to its extreme complexity. There is a need for an emotional system in the architecture of "ethical AI"; the presence of such a system makes AI more "humanlike" and brings in elements of "negligence" and errors, which is necessary when requesting to create an anthropomorphic AI. In addition, the issue of legal liability for the actions of AI and the legal status of AI in the future is raised. The levels at which AI and ethics are connected and interact with each other are revealed: the level of technical/algorithmic integration of ethics into the intellectual system; the level of regulatory and engineering methods for analyzing and evaluating ethical issues; the level of action codes, standards and certification processes, which allows developers and users to interact with AI in real time. As a conclusion, it is noted that due to the increasing involvement of AI in the everyday life of people, the development of an ethical system for AI is no longer just a matter of theoretical studies, but becomes an urgent problem with the ability to use research results or check ideas in practice. In the development of the methodology and architecture of AI systems, a great role belongs to philosophy as a source of many ethical programs and principles; therefore, philosophers need to participate in the interdisciplinary research of existing AI systems and in the development of new ones.

Keywords

Authors

References