When a Brown University professor noticed nearly half his students had aced a take-home midterm in a course that usually averages grades in the 70s, he got suspicious. So he made the final exam in person and told the class their midterm grade would only stand if their final exam performance held up.

Twenty-seven students didn’t show up for the final. Of the students who did, 19 failed. The class average fell from 96 to 48. They didn’t lack the necessary resources to master the material. The difference was they didn’t have access to artificial intelligence during the final.  These students will be headed into the workforce, and the gap between looking competent and being competent is something employers will have to pay close attention to.

“The higher the stakes, the more important it is to preserve judgment, verification and accountability.”

That gap is exactly what concerns Joey Huang, an assistant professor of learning, design and technology at NC State, whose research examines AI-supported learning and AI hallucinations. “Overreliance begins when we stop using AI to extend our thinking and start using it to avoid thinking,” she said. “The warning sign is that someone can no longer explain, evaluate or take responsibility for what the AI produced.”

AI has become a fixture in the modern workplace — drafting emails, untangling spreadsheets and even brainstorming when the ideas won’t come. Used well, it can be a genuine timesaver. When used on autopilot, it can quietly do what researchers say GPS did to our sense of direction and smartphones did to our memory for phone numbers.

Research backs that up. A Microsoft and Carnegie Mellon study of 319 knowledge workers found a telling pattern: the more people trusted AI, the less they questioned its output. An MIT study found that people who rely on AI to write showed weaker brain engagement and struggled to explain their own work — a pattern researchers call idea flattening.

Not every task carries the same risk, Huang notes. Low-stakes work, such as reorganizing notes, drafting a headline and writing routine emails, frees up mental energy for bigger decisions. But for substantive work, “the higher the stakes, the more important it is to preserve judgment, verification and accountability,” she said. 

Despite its drawbacks, there are no widespread calls to ditch AI. The problem arises when it is used as a substitute for your own thinking. 

For that kind of work, Huang recommends a simple five-step habit for checking AI’s output before you trust it.

1. Pause. Form your own initial take before asking AI so its answer doesn’t frame your thinking from the start.
2. Identify. Flag what actually needs verification — facts, stats, quotes, dates and citations.
3. Trace. Find the original, credible sources behind the claim.
4. Compare. Check what that source really says against what AI claims.
5. Judge. Decide for yourself. “A citation is not evidence until you open the source and confirm that it actually supports the claim,” Huang said.

Here are a few more common-sense habits you can adopt to preserve your brain’s muscle memory:

• Draft first, ask second. Take a first crack on your own before asking AI for help. It keeps the start-from-scratch muscle strong.
• Ask for options, not answers. Let AI offer a few directions, then use your judgment to pick and shape one.
• Keep one task AI-free. Pick something to do, such as writing meeting notes, writing a first draft or doing a quick calculation, and don’t ask AI to assist you. 

The move toward using AI to strengthen skills rather than replace them is showing up at the organizational level, too. IBM’s SkillsBuild program uses AI to recommend courses and certifications for employees transitioning to tech-driven roles. The NC State University Artificial Intelligence Advisory Group offers guidance on responsible AI use across university roles and functions, helping to strike that important balance between leaning on AI without losing the judgment behind it.

Think of AI as a smart, new colleague, not a stand-in for the thinking everyone still needs to do. As Huang noted, “the AI system you are using today may be among the least capable systems you will use during your career.” In a nutshell: The brain is still the best tool in the building.

Joey Huang, who holds a doctorate in learning and development sciences, is an assistant professor of learning, design and technology at NC State’s College of Education. A learning scientist whose research covers computational thinking, collaboration and AI-supported learning, she studies how people can use AI thoughtfully without losing the thinking skills it’s meant to support.