Topics that shape the future
- Closer look at research results -

I actually started relatively recently, around 2019. I had no prior programming experience, so I started by setting up a Python environment. Fortunately, the late 2010s was an excellent time to begin research. There were three trends for this reason: ① increased awareness of preventing accounting fraud from around 2005, ② the advancement of big data in financial reporting from around 2010, and ③ the spread of text mining and machine learning technologies from around 2015, which led to an increase in related papers. Fueled by this background, I was able to continue studying and learn to write code, which brings me to where I am today. I continue to learn along with the development of the "accounting x AI" field.

The advantage of machine learning lies in its scalability, which, unlike traditional statistical research, allows for the incorporation of a wide variety of variables with virtually no limitations. Therefore, I am particularly focused on areas where diverse variables can lead to fraud detection.

Rather than saying "the resolution has increased," it might be more accurate to say "we can now see things from a different angle than before." In traditional statistical analysis, setting hypotheses beforehand is crucial, but in machine learning, you can input variables even without hypotheses, and sometimes unexpected effective variables are discovered from this. In overseas research, it has been reported that in the process of investigating the reasons why a particular variable is effective, features that humans had not noticed often come to light.

That's right. Combining methods from different fields, such as finance or language education indicators (like obfuscation) with accounting, expands the possibilities. For this reason, we are constantly exchanging opinions and gathering information with various experts. In collaborative research, dialogue with practitioners' on-the-ground experiences is also important. We discuss whether the variables identified by AI align with the on-the-ground experience of auditing and what they mean. If we can create a highly accurate and practical model, it will be useful for auditing practice and investors' decision-making. This kind of interdisciplinary research is very exciting.

Yes. Generative AI excels at contextual interpretation and can deeply evaluate and interpret the nuances (positive/negative, etc.) of complex financial documents. This gives us confidence that we can improve conventional metrics without building complex models, and we are currently researching how to integrate it into our models.

Even more interesting are the insights gained from collaborative research with auditing experts. When experts input their "hard-to-articulate sense of unease" as prompts, the AI provides hints for judgment from that perspective. We believe that AI is not merely "artificial intelligence," but "Augmented Intelligence," which extends human capabilities. By effectively utilizing this technology, we may be able to see the "humanity" behind those who commit fraud.

This research is in the field of "accounting x AI," a field that transcends the boundaries between humanities and sciences. In the coming era, there will be fewer jobs that can be completed with only a single specialty. Aoyama Gakuin University offers opportunities to learn while connecting with society through dialogue with companies and auditing firms. We encourage you to "multiply" your intellectual curiosity and challenge yourself in a new field. Research is not just about writing papers; it is about creating the future of society and yourself. We would be delighted if you could take that first step here.