Explainable Artificial Intelligence

The rapid development of machine learning methods, particularly deep learning, has led to the deployment of artificial intelligence systems in areas with a significant impact on human lives, such as healthcare, finance, and transportation. While these models often achieve very high predictive accuracy, their decision-making processes are difficult for humans to interpret. This lack of transparency raises technical, ethical, and legal questions related to the trustworthiness, fairness, and accountability of algorithmic decision-making. The field of Explainable Artificial Intelligence (XAI) addresses this challenge. The aim of this bachelor’s thesis is to analyze the reasons for the need for explainability in current artificial intelligence systems, systematically define the basic approaches to explaining their decision-making, and critically evaluate their benefits and limitations. The thesis is theoretical and analytical in nature and is based on a systematic literature review and synthesis of specialized sources in the fields of machine learning, computer science, ethics, and law. The results of this study show that explainability is an essential prerequisite for the safe and socially acceptable deployment of artificial intelligence (AI) systems; however, current approaches to XAI face fundamental limitations. In particular, additional post-hoc methods may provide explanations that are understandable but not always technically accurate representations of the model’s actual decision-making process. The paper also highlights the problem of spurious correlations and the limited ability of current methods to distinguish between correlation and causation. The conclusion discusses promising directions for the further development of XAI, particularly mechanistic interpretability and causal approaches, which have the potential to overcome some of the fundamental limitations of existing methods.