인공지능은 회계업무에 (얼마나) 적합한가?: 공인회계사 자격시험을 중심으로

Artificial intelligence (AI) is rapidly replacing human labor. Physical labor has already been largely replaced. In contrast, cognitive labor, which requires expert judgment, intuition, and insight, has been slower to be replaced. This paper examines the extent to which AI can effectively perform the tasks of accounting professionals, one of the most representative forms of cognitive labor, based on the performance on the Certified Public Accountant (CPA) exam. When AI (ChatGPT 4o1) was tasked with solving the 2024 CPA exams, it achieved a correct answer rate of 90% on the first-tier (accounting) exam. The score is higher than the average score of 53% in the accounting subject obtained by first-tier passers of the 2024 CPA exam. Specifically, the scores of non-computational questions were higher than those of computational questions in each sub-category in the accounting subject. The results suggest that the current level of AI learning is sufficient to meet the basic accounting professional competence required by the qualification exam. The second-tier CPA exam requires applicants to answer multiple essay questions by presenting complex material on a single accounting topic. Therefore, the second-tier financial accounting subject requires more multifaceted and complex cognitive skills than the first-tier accounting subject. When we had ChatGPT 4o1 solve the second-tier financial accounting exam, it resulted in a score of 51%. While the score is below the 60 % absolute cut-off for passing the second exam, it is higher than the 40% cut-off for the minimum requirement applied to the relative assessment. The result that the second test scored lower than the first suggests that even the best AI in 2025 will not be able to fully solve complex and multifaceted accounting problems, which might be more pronounced under the principles-based K-IFRS. Combined, our results suggest both the potential and limitations of AI for accounting expertise. This phenomenon is not limited to Korean qualifying exams. When the initial version of ChatGPT was announced in late 2022, there was much skepticism about the capabilities of AI. This was due to concerns about imperfect accuracy and risks to information security. In a January 2023 experiment solving accounting problems, ChatGPT slightly outperformed human students. However, the AI’s performance improved steeply with each subsequent training session. In May 2023, ChatGPT 3.5 was still unable to pass the US CPA exam. However, since its latest iteration, GPT-4, it has passed the Certified Public Accountant, Certified Management Accountant, Certified Internal Auditor, and Enrolled Agent exams with an average score of 85%. Large accounting firms are accelerating the development and adoption of AI from a strategic perspective. In addition to automating simple repetitive tasks in the external audit process, machine learning is also considered to replace complex judgments required in audit work. In other words, the recent use of artificial intelligence in the auditing process has evolved from simple reliance on existing IT tools to automate the cognitive tasks of auditors. Nevertheless, the acceptance of AI by the accounting workforce is still low. Research shows that while 85% of accounting professionals feel they need to speed up their adoption of technology to remain competitive, many are reluctant to explore data analytics tools beyond Excel. A survey of Korean CPAs reports that CPAs recognize the importance of AI but perceive insufficient investment and training. If AI can perform as well as the accounting expertise of CPAs, however, it could possibly change the career prospects for the profession. More specifically, the social and economic value of the CPA qualification would be diluted when AI successfully delivers accounting expertise. In this regard, our results provide an opportunity for the accounting profession to become specifically aware of the cognitive capabilities of AI and to gauge its potential impact on the CPA labor market. To be specific, our findings imply that the scope of CPA expertise must immediately evolve to encompass more advanced contextual and critical thinking. This study also calls for a fundamental review of the educational objectives and curricula of public and private education institutions in the accounting field. This study serves as an exploratory assessment of AI’s current accounting capabilities through CPA exams and provides a basis for further discussions on the role of AI in the accounting profession.