Humans Reign Supreme: ChatGPT Falls Short on Accounting Exams

Image name: Accountant-vs.-Machine.jpg

In a test of ChatGPT’s ability to handle accounting assessments it still couldn’t compete with the student’s level. Credit: Nate Edwards/BYU Photo

The massive study was crowd-sourced with 327 co-authors from 186 institutions in 14 different countries.

Last month, OpenAI introduced its latest AI chatbot product, GPT-4. The company claims that the bot, which employs machine learning to produce text in a natural language style, performed exceptionally well on various exams. Specifically, it achieved a score in the 90th percentile on the bar exam, passed 13 out of 15 AP exams, and received a near-perfect score on the GRE Verbal test.

Academicians from Brigham Young University (BYU) and 186 other institutions were curious about the performance of OpenAI’s technology on accounting exams. Hence, they subjected the original ChatGPT model to testing. The researchers have stated that although ChatGPT still requires improvement in the field of accounting, it has the potential to revolutionize the way people teach and learn, for the better.

“When this technology first came out, everyone was worried that students could now use it to cheat,” said lead study author David Wood, a BYU professor of accounting. “But opportunities to cheat have always existed. So for us, we’re trying to focus on what we can do with this technology now that we couldn’t do before to improve the teaching process for faculty and the learning process for students. Testing it out was eye-opening.”

Since its debut in November 2022, ChatGPT has become the fastest-growing technology platform ever, reaching 100 million users in under two months. In response to intense debate about how models like ChatGPT should factor into education, Wood decided to recruit as many professors as possible to see how the AI fared against actual university accounting students.

His co-author recruiting pitch on social media exploded: 327 co-authors from 186 educational institutions in 14 countries participated in the research, contributing 25,181 classroom accounting exam questions. They also recruited undergrad BYU students (including Wood’s daughter, Jessica) to feed another 2,268 textbook test bank questions to ChatGPT. The questions covered accounting information systems (AIS), auditing, financial accounting, managerial accounting, and tax, and varied in difficulty and type (true/false, multiple choice, short answer, etc.).

Although ChatGPT’s performance was impressive, the students performed better. Students scored an overall average of 76.7%, compared to ChatGPT’s score of 47.4%. On 11.3% of questions, ChatGPT scored higher than the student average, doing particularly well on AIS and auditing. But the AI bot did worse on tax, financial, and managerial assessments, possibly because ChatGPT struggled with the mathematical processes required for the latter type.

When it came to question type, ChatGPT did better on true/false questions (68.7% correct) and multiple-choice questions (59.5%), but struggled with short-answer questions (between 28.7% and 39.1%). In general, higher-order questions were harder for ChatGPT to answer. In fact, sometimes ChatGPT would provide authoritative written descriptions for incorrect answers, or answer the same question in different ways.

“It’s not perfect; you’re not going to be using it for everything,” said Jessica Wood, currently a freshman at BYU. “Trying to learn solely by using ChatGPT is a fool’s errand.”

The researchers also uncovered some other fascinating trends through the study, including:

• ChatGPT doesn’t always recognize when it is doing math and makes nonsensical errors such as adding two numbers in a subtraction problem, or dividing numbers incorrectly.

• ChatGPT often provides explanations for its answers, even if they are incorrect. Other times, ChatGPT’s descriptions are accurate, but it will then proceed to select the wrong multiple-choice answer.

• ChatGPT sometimes makes up facts. For example, when providing a reference, it generates a real-looking reference that is completely fabricated. The work and sometimes the authors do not even exist.

That said, authors fully expect GPT-4 to improve exponentially on the accounting questions posed in their study, and the issues mentioned above. What they find most promising is how the chatbot can help improve teaching and learning, including the ability to design and test assignments, or perhaps be used for drafting portions of a project.

“It’s an opportunity to reflect on whether we are teaching value-added information or not,” said study coauthor and fellow BYU accounting professor Melissa Larson. “This is a disruption, and we need to assess where we go from here. Of course, I’m still going to have TAs, but this is going to force us to use them in different ways.” 

© Brigham Young University