Rozhodování uživatele s pomocí AI v doporučovacím rozhraní: vliv vysvětlení a možnosti override na důvěru a autonomii

This bachelor thesis examines how selected elements of an AI recommendation interface influence user decision-making in a digital environment. It focuses on the relationship between recommendation explanation (why-explanation), the possibility to change the recommended option (override), and the level of system autonomy. The thesis is based on the assumption that user experience with a recommendation is shaped not only by the content of the recommendation itself, but also by the way it is presented in the interface, which may affect user trust, perceived transparency, and the sense of control over the final decision. The theoretical part defines the issue of human–AI interaction in the context of recommender systems and focuses on the relationship between explanation, trust, transparency, user control, and system autonomy. It also shows that a recommendation interface should not be understood only as a technical environment for displaying model output, but as a design space that actively structures the user’s decision situation. The empirical part is based on testing a clickable prototype called Streamio, which simulates the selection of a subscription plan in a streaming service. The prototype was designed in Figma and tested in Maze. The research design compares two interface variants, low autonomy and high autonomy, which differ in the intensity of system guidance. Data collection combined a prototype walkthrough with a follow-up questionnaire focused on trust in the recommendation, perceived control, and interpretation of the decision situation. The results suggest that the more autonomous variant more often supported acceptance of the recommended but unsuitable option, while the less autonomous variant more often led users to compare the parameters of the available options. At the same time, the findings indicate that why-explanation can strengthen trust in a recommendation without automatically leading to a deeper understanding of the decision situation, and that the mere presence of an override option does not in itself guarantee active user control. The contribution of the thesis lies in connecting theoretical insights with the empirical testing of a specific recommendation interface.