In a recent episode of Utilizing AI, recorded live at Qlik Connect, host Stephen Foskett sat down with Frederic Van Haren, CTO and Founder of HighFens, and Sidney Drill, Qlik Global Solutions Director for Data, Analytics, and AI, to discuss a looming threat in the age of AI: large language models that sound authoritative even when they are completely wrong.
The panel explored how this tendency creates a modern twist on the classic "garbage in, garbage out" problem. Instead of merely passing along bad data, AI systems can now generate errors that feed back into trusted datasets, creating a dangerous feedback loop. As AI adoption accelerates, data professionals are increasingly worried that AI-generated inaccuracies will corrupt decision-making processes.
“The risk is that we start to trust these models blindly because they present their answers with such confidence,” noted Van Haren.
The conversation highlighted the urgent need for robust governance, validation, and strong data practices. Without these safeguards, organizations may unknowingly build their strategies on a foundation of AI-confident falsehoods. The panelists called for a renewed focus on data quality and a healthy skepticism toward AI outputs, especially as enterprise reliance on generative AI grows.