Úloha a dôležitosť etických pravidiel v systémoch umelej inteligencie.

Abstract. In the field of artificial intelligence, its ethical rules represent a significant part, which is understood as a set of values, principles and techniques using generally accepted standards of „good and bad“ and regulating ethical behaviour in the process of developing and using AI technologies. Ethical rules create a delicate balance between technological progress and the protection of human values such as privacy, justice and transparency. The real application of ethical values and principles is however possible only after their implementation in AI systems, in each phase of the „AI life cycle“ from the initial design through its practical deployment to monitoring. Currently, there is neither general nor uniform practice about what method or what mechanisms should be used to implement ethical rules into AI systems. Common to the implementation procedures is however the fact that they focus on AI algorithms that immediately generate final decision of AI system containing an ethical rule used (applicable) in a specific situation. For this purpose are algorithms trained on a large amount of input data, which, however, may contain bias caused by humans in the preparation and the application of AI. If they are not properly addressed, the practical application of such „infected“ algorithms can lead to discriminatory consequences and the strengthening of social inequality. These shortcomings should be identified by AI audit, the aim of which is to assess whether individual AI systems fulfil the expected functions while respecting the corresponding ethical and legal rules. However, there are currently no internationally accepted standards of procedure in the field of AI auditing. This is caused by e.g. insufficient adjustment of AI (since its very definition is still ambiguous), so the creation of common standards for its audit seems problematic. Although they can be identified on the basis of and through an audit, this alone is not capable of actually eliminating such algorithms or to block their application. System biases found their way into AI because, like all technologies, they were created by humans. As a result, the final results of audits reflect and preserve human biases and errors in society. Therefore, it is generally accepted that the only way how to “combat” AI bias is currently by rigorously testing and evaluating its data and algorithms, and using best practices for collecting and using data and creating AI algorithms.