New study examines AI solution to leadership communication gaps
Can artificial intelligence-mediated conversations build interpersonal skills and improve confidence at work? Computer science researchers want to find out.
In filling out his profile for a new artificial intelligence (AI) software system, U.S. Air Force Maj. Lance Wilhelm was particularly candid about his past leadership challenges.
"I struggle to be confident when communicating with employees that have much more experience in the lab than me," he said out loud as he typed. "I would like to improve balancing empathy with directness, while maintaining confidence in the decisions I need to make."
He was giving a demonstration for study participants on how to use the software, but it was based on Wilhelm's past struggles as a new manager in a lab where many of the employees had spent years doing their jobs.
"This project came out of my passion, not only for computer science and artificial intelligence, but leadership," Wilhelm said. "A lot of the military and Department of Defense officer corps is centered around developing us as leaders, but employers may not be able to offer that. So can we use AI as a training tool?"
Building a learning environment
Difficult conversations crop up in every workplace — giving negative feedback, delivering bad news, resolving conflict — and most people just aren’t very good at them.
- According to one 2020 study, 86 percent of employees blamed ineffective communication for workplace failures.
- Seventy percent of employees avoid difficult workplace conversations altogether.
- Research highlighted by Harvard Business Review found that two-thirds of managers feel uncomfortable communicating with employees.
To find out if AI can improve communication at work, Wilhelm and his Ph.D. advisor, Assistant Professor of computer science Eugenia Rho, have created a hands-on AI tool called CommCoach that simulates tough conversations so workers and managers can practice and get coaching feedback in real time.
“The goal is to help users notice habits, try alternatives, and develop their own communication style with low stakes,” Wilhelm said.
The study has the potential to fill gaps in the research about AI-assisted learning and training as well as in the marketplace for AI tools.
What Wilhelm learns from this research will inform his future career. The project is part of his Ph.D. thesis in computer science. He's attending Virginia Tech through the Air Force Institutes Advanced Academic Degree and Special Experience and Exchange Duties program. The program offers qualifying noncommissioned officers three years of subsidized, full-time academic work in an area needed by the military. Wilhelm remains on active duty while in school and will return to the Air Force after finishing his coursework.
Scenarios and simulations
To use CommCoach, users describe a challenging workplace situation to the system.
“The system then builds a scenario that mirrors it, whether you’re preparing for a performance review with your manager, working through a performance improvement plan discussion, or navigating tension with a colleague,” Wilhelm said. “The system adjusts the way your simulated partner communicates, so your practice round feels grounded and useful rather than scripted.”
While the feedback isn’t scripted, CommCoach is optimized to support certain skills:
- Being clear and direct while staying attentive to the other person’s perspective.
- Reflecting, paraphrasing, and checking your understanding of what’s being said, rather than reacting or rushing to solve the problem.
- Noticing emotional or contextual cues and responding with curiosity rather than avoidance or escalation.
- Paying attention to your communication patterns across multiple practice conversations, rather than treating any single exchange as a pass/fail moment.
The system adapts its feedback to what a specific user tends to struggle with or prioritize, from listening to emotional awareness, making the coaching more relevant over time. As users continue to practice conversations, the researchers are assessing if participants report building both communication skills and leadership confidence.
Troubleshooting AI communication
But Wilhelm and Rho aren’t just developing a tool to solve a workplace problem. They want to understand how AI-mediated conversation practice is used, what if anything it changes, and where it may help or even hurt — before such tools are deployed commercially.
“My worry is that these AI technologies are being used to potentially train people with a limited understanding of their upsides and downsides,” Wilhelm said. “Are there blind spots that we’re missing? Are there things that users find helpful or unhelpful, and why?”
Relying on AI-mediated communications could have long-term consequences as well.
“If AI consistently handles the difficult parts of communication for people, we risk improving short-term outcomes while weakening long-term skill development,” Rho said. “This project is motivated by that concern.”
Wilhelm and Rho are seeking participants for a compensated study that will inform CommCoach's design and provide anonymized data on its strengths and weaknesses. Rho said that to gain in-depth knowledge about the usefulness of AI in communication, they need as many users as possible to join the project.
