Campbell Brown warns that AI’s confident answers can hide mistakes

An AI answer can be easy to read and hard to challenge—even when it is wrong. Forum AI founder and CEO Campbell Brown warned about that mismatch in an October 28 interview with CNBC’s Scott Wapner at the network’s Technology Executive Council Summit in New York.

When clarity conceals an error

Brown’s concern was not simply that AI produces mistakes. It was that the way an answer is written can make those mistakes difficult for a reader to recognize. She said AI-generated responses arrive with such confidence and clarity that users can struggle to question them or locate the errors.

The remarks extended an argument Brown made in a Wall Street Journal opinion piece published in June. There, she wrote that AI was already good enough that its mistakes did not look like mistakes, and called for a more critical approach to its output as reliance on the technology grows.

Human expertise in domains is essential to getting this right.

Campbell Brown, Forum AI founder and CEO, speaking with CNBC’s Scott Wapner

Political information raises the stakes

Brown singled out political and election information as an area of growing concern as people turn to AI for answers. Her criticism covered both how information is presented and where it comes from, putting source quality alongside the difficulty of spotting an authoritative-sounding error.

She also directed criticism at AI developers. Brown said labs were not engaging with real subject-matter expertise as much as she believed they should. Her argument placed people with knowledge of a particular field at the center of improving AI’s answers, rather than treating clear presentation as enough to earn trust.

The interview was part of CNBC’s eighth annual summit, a members-only gathering. The summit’s agenda linked questions of AI trust with the practical decisions facing technology leaders: returns on computing investment, cybersecurity, regulation and the costs of building out AI.