AI adoption risks eroding both human expertise and long-term knowledge

Companies adopting AI to cut costs and improve efficiency may face an unexpected risk: long-term knowledge loss. While automation offers clear benefits, a new study warns that both human expertise and AI-driven insights can degrade over time if not properly managed. The problem begins when AI systems, trained on outdated data, start to lose accuracy—a phenomenon known as concept drift. In fast-moving industries like journalism, this drift happens faster than in more stable fields. Without regular updates, AI models fail to reflect current realities, leading to unreliable outputs.

Human expertise faces a similar threat. Layoffs, retirements, or simply not using critical skills can erode institutional knowledge. An automotive company, for example, replaced a veteran production worker’s expertise with an AI system—only to risk losing both the worker’s insights and the AI’s effectiveness as conditions changed. The study, blending management research and machine learning, highlights a *double threat*: AI drift and the loss of human oversight. While reviewing AI outputs can help maintain accuracy, this process slows down operations, conflicting with the short-term efficiency gains that drive many firms to adopt AI in the first place. Companies like Amazon and SAP have cut jobs after implementing AI, often removing the very employees who could spot errors or update systems. Without safeguards, this cycle accelerates knowledge loss across the organisation.

The research underscores a key challenge for businesses relying on AI. Without ongoing investment in model updates and human expertise, organisations risk losing critical knowledge permanently. Balancing efficiency with long-term reliability will require both resources and a shift in how companies manage their AI systems.