Auburn researcher explores how AI changes productivity and human incentives
While most businesses have introduced AI in the workplace to enhance employee productivity,
an Auburn University information systems researcher cautions that under certain circumstances, AI deployment can inadvertently
zap the motivation of its highest-performing workers, which could diminish the company’s
overall profit.
In a recently published study, Harbert College of Business faculty member Di Yuan found that implementing AI may do more harm than good in a pay-for-performance workplace,
where employee compensation — pay, bonuses and raises — is linked to measurable outcomes
rather than a flat salary.
In the study, she and her research colleagues created a game-theory model to analyze
how AI-driven knowledge sharing changes competition among workers.
“We looked at the specific organizational dynamics that interact with employee incentives
and found that the reward system is very relevant,” Di said. “If a company doesn’t
actually think about its existing employee incentive scheme and just introduces the
AI bluntly, it might actually backfire and hurt the overall productivity level.”
To illustrate the findings, Di describes a tech firm that pays its software engineers
based on their productivity. Some of them are extremely good at writing code, while
others are less proficient programmers, but have other valuable talents like good
communication and critical thinking skills.
Further, she explained, the AI tool being deployed was trained on the work of the
best-performing software engineers, meaning their expert knowledge is being given
to their less-skilled engineering colleagues.
“When the company introduces a productivity-enhancing AI-powered tool like copilot,
the poor programmers will now be better at writing code than their colleagues,” she
said, noting that an incentive shift has occurred because the less skilled programmers
benefitted greatly from the AI and will ultimately be paid more than the highly skilled
programmers.
“This reduction in the highly skilled programmers’ performance ranking and pay level
will have a demoralizing effect that lowers their productivity,” she said.
Di and her colleagues examined ways to avoid the demoralizing effect on top performing
employees and determined that the best solution is to introduce AI tools that raise
the productivity of the less skilled workers, while allowing the highly skilled employees
to retain their superior productivity levels.
“We have examined a few intuitive methods and found that choosing a less-costly and
less-powerful AI [tool] was the most effective solution,” Di said. “Other methods
we considered didn’t work because they had strong additional constraints.”
“Di’s paper is the first study by a faculty member from our department to be published
in Management Science. Being published in such a prestigious journal reflects the
exceptional quality of her work.”
Uzma Raja, chair of the Department of Business Analytics & Information Systems
Di conducted the study with University of Maryland Assistant Professor Manmohan Aseri and University of Pittsburgh Professor Narayan Ramasubbu. Their paper, “Backfiring AI: AI deployment in the workplace,” was published in Management Science, a premier journal for management, business
and operations research.
“Di’s paper is the first study by a faculty member from our department to be published
in Management Science,” said Uzma Raja, Benoski Professor and Department Chair of
Business Analytics & Information Systems within Harbert College. “Being published
in such a prestigious journal reflects the exceptional quality of her work.”
Degrees & Programs
Faculty & Staff
Career Development
Recruiters & Industry
