When AI Amplifies the Biases of Its Users

"A widely discussed concern about generative AI is that systems trained on biased data can perpetuate and even amplify those biases, leading to inaccurate outputs or unfair decisions. But that's only the tip of the iceberg. As companies increasingly integrate AI into their systems and decision-making processes, one critical factor often goes overlooked: the role of cognitive bias."

"Grace Chang , PhD, Associate Director of Behavioral Science & Insights at Ernst & Young LLP, is a cognitive neuroscientist who bridges science and practice to design effective learning programs. Previously, she was Chief Scientific Officer at The Regis Company, taught global management professionals through the NeuroLeadership Institute, conducted assessment research for UCLA CRESST, and lectured at UCLA. She frequently presents at major conferences and publishes in academic and practitioner outlets."

Generative AI trained on biased data can perpetuate and amplify existing biases, producing inaccurate outputs or unfair decisions. Companies increasingly integrate AI into systems and decision-making processes, yet they often overlook how human cognitive biases influence AI development, deployment, and interpretation. Cognitive biases affect data selection, labeling, model training priorities, evaluation metrics, and user trust in outputs, compounding technical biases. Behavioral-science-informed strategies can identify bias points and design interventions such as structured decision protocols, debiasing training, diverse review teams, and monitoring frameworks. Integrating cognitive-bias mitigation into AI governance improves fairness, accountability, and reliability of AI-driven decisions.

Read at Harvard Business Review