AI agents help teams solve problems 19% faster
Teams communicating exclusively through personal AI agents solved collaborative problems 19 percent faster than teams using traditional group chat, a Carnegie Mellon University and Unanimous AI study found.
Teams using Hypercom AI reached correct solutions in an average of 11.6 minutes, compared with 14.3 minutes using group chat. Accuracy was nearly identical at 93 percent and 92 percent, respectively.
The reduction in solution time was statistically significant and remained after researchers controlled for which simulated terror plot each team solved. The study, released by Carnegie Mellon’s Center for Collaboration Science and Unanimous AI, is available as a preprint.
The experiment involved 15 networked teams of seven or eight members. Each team had to identify a terrorist in a simulated scenario using evidence distributed among its members, combining relevant information while rejecting conflicting evidence and red herrings.
Each team completed problems using both traditional group chat, similar to Slack or Microsoft Teams, and Hypercom AI. In the latter, each participant communicated only with a dedicated AI agent rather than directly with teammates.
The agents, called Conversational Surrogates, passed relevant information to other participants through their respective agents. They coordinated communication without introducing new information or making deductions of their own.
The researchers described the findings as the first demonstration that a distributed team could solve a problem requiring every member’s contribution while communicating entirely through intervening AI agents. They expect benefits to increase with group size, particularly when participants are multitasking.
The research addresses communication bottlenecks in time-pressured settings: group calls allow only one speaker at a time, while simultaneous chat messages can bury important details. Research shows these problems worsen as teams grow and members perform other urgent tasks.
The researchers identified potential applications in emergency response, disaster relief, cybersecurity, medical triage, military command and critical operations in large organizations. A first responder, for example, could continue working while an agent shares observations and returns relevant field updates.
Co-author Anita Williams Woolley, professor of organizational behavior at Carnegie Mellon’s Tepper School of Business, said integrating distributed expertise through AI could benefit organizations making rapid decisions.
“There could be additional opportunities to facilitate completely asynchronous coordination and dramatically reduce the need for meetings that currently clog everyone’s calendars and slow down progress,” Woolley said.
The preprint, “Evaluating an Agentic Hyper-Communication Framework for Optimized Human Deliberation Across Distributed Teams,” was written by Louis Rosenberg, Hans Schumann and Gregg Willcox of Unanimous AI, and Woolley and Ganesh Mani of Carnegie Mellon.
The U.S. Air Force funded the research through AFWERX contract FA864926P0228, “Hyper-Communication for Optimized Command and Control.” An accompanying disclaimer said the author’s views do not necessarily represent the official policy or position of the Department of the Air Force, Department of Defense or U.S. government.
Hypercom AI is available as a beta feature within Unanimous AI’s Thinkscape platform.
Featured image credit
