Externe Rechnungslegung und Steuern im KI-Zeitalter: Eine empirische Untersuchung zur Urteilskompetenz von Studierenden hinsichtlich inhaltlicher Richtigkeit

Abstract

This master's thesis analyzes the use of ChatGPT in the teaching of external accounting. Against the background of the increasing integration of artificial intelligence (AI) in both professional and academic life, the question arises to what extent the contents of language models such as ChatGPT are critically scrutinized and checked for their correctness when used in studies. AI tools like ChatGPT offer diverse application possibilities, which can provide considerable relief, particularly in the handling of repetitive tasks. At the same time, existing challenges are coming into focus, such as ethical issues, potential impacts on learning motivation, and the problem of generating misinformation. Since private use of such systems can hardly be controlled even in the presence of institutional restrictions, it appears all the more important to examine students' ability to critically evaluate the correctness of ChatGPT-generated content and to identify potential misinformation in order to enable responsible and smooth use in the higher education context. The thesis is divided into a theoretical and an empirical part. The theoretical section initially addresses the technological foundations, including key concepts of natural language processing, the training processes of large language models, and the development and functionality of ChatGPT. This is followed by a systematic presentation of the current state of research on the use of language models in the context of accounting education, highlighting key potentials as well as existing challenges associated with the use of ChatGPT in accounting teaching. The empirical part of the thesis applies various theoretical models. The focus is on signal detection theory, which primarily serves to measure students’ ability to distinguish between correct and incorrect AI-generated content. Based on these findings, a regression analysis is conducted to identify potential influencing factors on evaluation ability. Additionally, the Technology Acceptance Model is employed to capture subjective perceptions regarding the usefulness and ease of use of ChatGPT. Methodologically, the study is based on a standardized quantitative online survey of currently enrolled students in the Controlling, Accounting and Financial Management course. The results of the empirical analysis show that students who are familiar with ChatGPT and use the tool occasionally to regularly in their studies tend to classify the generated content as correct - regardless of its actual factual accuracy. This tendency is accompanied by a low to moderate ability to differentiate, indicating a limited capacity to detect incorrect content. This represents a significant risk in the context of higher education. At the same time, it was found that sound subject-specific knowledge as well as professional experience in the respective field have a beneficial effect on participants' sensitivity. These groups demonstrated significantly higher scores in terms of the ability to differentiate content. Despite these tendencies, the surveyed students expressed rather reserved views regarding their trust in the generated content and their overall satisfaction - this stands in contrast to the consistently positive evaluation of the perceived usefulness and user-friendliness of ChatGPT.

Date of Award	2025
Original language	German (Austria)
Supervisor	Susanne Leitner-Hanetseder (Supervisor)