"Knowledge That Works" is more than a UBalt tagline—it's the driving force behind everything The University of Baltimore stands for and everyone we serve. Through our faculty's insights and expertise, we invite you to explore what that knowledge looks like in action.
Professor of Accounting Mikhail Pevzner, Ph.D. shared his perspective on how AI is transforming competitive advantage in scholarly research and shows through is own experience how colleagues can utilize their talents to create an advantage when collaborating.
Artificial intelligence is usually described as a technology that automates work. That is obviously part of what it does. But the more I use it, the more I think that one of its most important effects may be different. AI can make it much easier for people with vastly different forms of expertise to work together.
That matters because many important problems do not fit neatly within a single field.
They need people who understand different functional areas, but those people often
do not have the time or background to become experts in one another’s disciplines.
My recent research experience inadvertently showed how AI can help bridge that gap.
I needed to collect detailed financial accounting data from corporate filings for
a research project I am working on. The project required substantial accounting and
disclosure knowledge because of heterogeneity of how companies present this information
in their annual reports. In the past, I’d have had to hand-collect this data which
would require months of work. This time around I decided to experiment with Claude
Fable 5 model which was very recently released.
My comparative advantage is in accounting. I know financial reporting, U.S. Securities
and Exchange Commission (SEC) filings and the kinds of disclosures we needed to identify.
What I do not have is advanced Python programming ability that would be needed to
extract this data.
Using Anthropic’s Claude Fable 5 model as a “black box,” I developed a process for identifying and extracting the relevant information by feeding it the examples of corporate disclosures I wanted to extract. But I could not independently verify the programming. What if Claude did something wrong that I did not notice?
My co-author, Dr. Cong Zhang, assistant professor of information systems at the University of Baltimore’s Merrick
School of Business, whose focus is data science, reviewed the technical process. He
could evaluate the code, the data structure, and the overall design in ways I could
not. At the same time, he did not begin with my knowledge of financial reporting,
SEC disclosures, or the institutional details of corporate filings. I verified the
accounting disclosure output. He verified the coding.
Without AI, I would have needed to acquire programming skills far beyond what was
practical for this project, or Dr. Zhang would have needed to spend a great deal of
time learning accounting rules, filing structures, disclosure practices and the context
surrounding the data. Instead, each of us was able to contribute what we knew best.
A process that might once have taken months took approximately a week.
This experience led to me to an insight of the nature of comparative advantage, one
of the key concepts in economics that goes back to 19th-century British economist,
David Ricardo. People and organizations become more productive when they concentrate
on what they do well, work with others who have different strengths, and trade on
their outputs in which they are inherently more efficient. In practice, however, collaboration
across fields is often expensive.
An accountant working with a programmer must learn enough programming to explain the
problem and evaluate the result. The programmer must learn enough accounting to understand
what is being asked. Similar barriers arise when attorneys work with economists, physicians
with engineers, or business professionals with computer scientists. The problem is
not simply that people lack knowledge. They often lack a shared language.
AI can reduce some of that cost. It can help a subject-matter expert develop a preliminary
technical solution without first becoming an expert in the entire technical field.
It can also help a technical expert understand the structure, terminology, and logic
of an unfamiliar professional problem.
That does not eliminate the need for specialists. It changes how their expertise can be combined.
The accountant does not need to become a professional programmer before meaningful
engagement in a data-intensive project. The programmer does not need to spend months
mastering accounting before beginning to contribute. But each person still needs enough
knowledge to understand their own levels of proficiency and to know when another professional
must verify the work.
That is the crucial point. AI did not make our expertise unnecessary. It made the combination of our expertise much more productive.
In our project, I could recognize problems in the accounting output that might not
have been obvious to a programmer. Dr. Zhang could identify technical weaknesses that
I would not have recognized. Neither of us could responsibly verify everything alone.
The project also showed me that AI can shift time away from the most repetitive and
frustrating parts of work and toward the parts that are more intellectually valuable.
This has important implications for higher education. Universities are still organized
largely around separate departments, majors, and professional schools. Students need
strong disciplinary foundations, but they also need opportunities to work across those
boundaries.
Business students should work with data science students. Technology students should
learn to understand the organizational settings in which their tools will be used.
Students in law, public policy, ethics and the humanities should be part of conversations
about how AI is designed and applied.
Students need to learn much more than how to use AI tools. They need to learn how
to define problems, contribute their own experience, work effectively with people
from other disciplines, verify AI-generated output, and take responsibility for the
result. That kind of education cannot happen entirely through lectures or conventional
written assignments. It requires projects, presentations, internships, consulting
engagements, and other forms of experiential learning built around real problems.
This is especially important in business education, where AI is not simply a computer
science issue but one that also involves accounting, finance, management, law, ethics,
cybersecurity, and organizational behavior.
A strong Artificial Intelligence for Business program, like the one at the University of Baltimore’s Merrick School of Business, should teach students not only how AI works, but also how to integrate it responsibly into organizations and use it to collaborate more effectively across different areas of expertise. Ultimately, AI may be remembered less for replacing human labor than for making specialized knowledge easier to combine. But that works only when genuine ability remains present.
Dr. Pevzner holds the Ernst & Young Chair in Accounting and brings deep expertise in accounting, auditing, finance, and economics to his work as a business-school professor, higher-education leader and consultant. He provides valuation and economic-damages estimation for corporate litigation, contributes economic analyses to SEC rulemaking, and has published more than forty peer-reviewed articles on financial regulations and accounting research.