New research suggests relying too heavily on AI could erode the firsthand experience managers need to make sound decisions
Managers increasingly turn to generative AI to draft emails, summarize meetings and answer questions in seconds. But new research suggests that convenience carries a hidden cost, the practical judgment managers build through years of firsthand experience. That reliance is only growing, with about one in five U.S. workers already saying some of their job is done with AI, according to the Pew Research Center.
A study published in the Academy of Management Review, led by researchers from the University of Bath, Ohio State University, the University of Lausanne and Cardiff University, examined how generative AI tools like ChatGPT reshape the way managers think and decide.
Dirk Lindebaum, the study's lead author and professor of management and organization at the University of Bath's School of Management in the U.K., said the team grew concerned that AI captures a fraction of the knowledge managers actually rely on to make good decisions.
"Artificial intelligence can only capture knowledge that has been codified," Lindebaum said. "It is based on the zeros and ones that are statistically predicted to generate an outcome. But there are other forms of knowledge, affective knowledge, embodied knowledge, tacit knowledge."
Tacit knowledge, the kind built through firsthand experience with people and problems, is what the study says erodes when managers reach for AI as a shortcut instead of working through a problem themselves.
What epistemic deskilling looks like
Lindebaum and his co-authors coined the term "epistemic deskilling" to describe a gradual erosion of decision-making ability that sets in when people outsource too much thinking to generative AI, particularly under deadline pressure.
"Epistemic deskilling for us is likely to happen when people work under time pressure," Lindebaum said. "It is very tempting to use the technology then because it can give you a quick fix. However, the concern that we raised is that it stops us from engaging with the real world."
A recent study from Microsoft Research and Carnegie Mellon University, which surveyed 319 knowledge workers about 936 real-world tasks, points to the same risk more broadly, finding that heavier confidence in AI tools tracked with less critical engagement with their output. The risk isn't unique to managers, either. A recent HRD America report on how skill decay is eroding workers' ability to think found researchers warning that outsourcing decisions to AI builds up what one researcher calls "cognitive debt," the gradual loss of capability that comes from letting AI handle decisions people used to work through themselves.
He pointed to two moments where this shows up most: when a manager turns to a chatbot instead of brainstorming with colleagues, and when a real problem surfaces and a manager asks AI for a solution instead of talking to the people involved. Over time, both habits mean managers stop collecting the firsthand experience that builds what Lindebaum calls practical wisdom, the ability to diagnose a problem accurately and land on a solution that fits.
A recent HRD America report on whether staff are leaning on AI a little too much found HR teams already wrestling with how deeply the technology has embedded itself in daily workflows.
Where accountability changes the equation
Lindebaum's paper also describes a more hopeful path he calls "epistemic upskilling." It happens when managers know they'll be held accountable for a decision. That accountability, rather than time pressure, is what pushes them to think a problem through instead of accepting an AI's answer at face value.
Pete Dusché, founder and principal consultant at Hesion Leadership Consulting in Nashville, Tennessee, reviewed the study and said it lays out two possible paths, deskilling and upskilling, depending on how AI gets used.
"So far, evidence in this realm does lead to what I'll call the erosion path," Dusché said. "The open question for leaders isn't whether to use AI. It's whether your organization is building in the accountability that pushes people toward the better path, the better judgment."
Dusché said it's a mistake to treat time pressure and accountability as a single trade-off. They're separate conditions that can show up together, and the dangerous combination is high time pressure paired with low accountability, which is what reliably drives deskilling.
"It's when you have to justify your own reasoning to other people and to yourself that you're able to say, these are the choices I made, and here's why," he said.
A recent HRD America piece asking whether AI's time-saving benefit is real found much of that saved time gets eaten up fixing AI output anyway, adding pressure to move fast rather than check the work.
Designing work that protects judgment
Lindebaum pointed to redesigning how work itself is structured, rather than restricting AI outright. Organizations need to build in regular moments where employees interrogate whether a routine still holds, rather than assuming AI's answer is automatically correct.
"You give managers intermittent learning experiences to understand the moments when a routine is changing, the why and how behind it," he said. "Use the technology when you feel the routine is stable, but keep sensitizing employees to moments when it's about to change."
Dusché said transparency about AI use, not secrecy, keeps everyone honest. It's fine for an employee to tell a manager they ran an idea past a chatbot, he said, as long as they can still walk through their own reasoning behind it. He also recommended having junior employees attempt tasks unaided before turning to AI, so they build the underlying skill instead of skipping straight to the shortcut, the same way children learn math before they're handed a calculator.
Lindebaum said the stakes extend beyond any single decision. If organizations chase AI's productivity gains without protecting time for hands-on experience, he warned, the newest managers could end up highly skilled at prompting a chatbot and far less able to handle a real-world problem when one arrives.
Knowledge, he said, has a codified element, but it also has affective, tacit and embodied dimensions that AI cannot touch. "If we don't understand that, we will likely impoverish our decision-making at work," Lindebaum said.
