The Hidden Cost of Generative AI Adoption

Generative AI offers tremendous potential for productivity gains, but new research suggests a concerning downside: over-reliance on these tools may be eroding human knowledge and critical thinking skills. Experts warn that organizations need to address this risk proactively before it undermines core capabilities.

The Knowledge Decay Phenomenon

According to a recent analysis by Matthias Holweg (University of Oxford) and Thomas H. Davenport, the widespread use of generative AI can lead to “knowledge decay” at an organizational level—where processes and outputs gradually deteriorate as people become less reliant on their own expertise.

This occurs through several mechanisms:

  • Verification challenges: Disentangling accurate information from AI-generated errors requires time-intensive fact-checking that often negates the efficiency gains of using AI in the first place.
  • Validation gaps: Demonstrating actual human intellectual contribution becomes more difficult when AI produces standard outputs (like reports and slides) that clients are effectively paying for expert insights on.
  • Iterative distortion: As knowledge passes through multiple AI iterations, it drifts further from the original “ground truth”, especially with LLMs that operate based on probabilistic next-word predictions rather than factual understanding.

Specific Risks Across Functions

These risks manifest differently across departments:

Recruiting: Candidates may use AI to optimize resumes and even generate interview responses, potentially selecting for those skilled at prompting algorithms rather than genuine qualifications. Consulting: Firms risk delivering AI-generated reports without clients receiving the intended human expertise. Performance management: Managers may generate generic evaluations instead of providing meaningful feedback based on real observations.

Mitigating Strategies

Rather than outright restriction, experts recommend a balanced approach:

  1. Strategic application: Limit AI use to scenarios where it demonstrably adds value rather than replacing essential human input.
  2. Structured inputs: Require specific details and factual responses that AI struggles to generate (e.g., asking candidates about particular projects or outcomes).
  3. Clear accountability: Define who is responsible for the accuracy and validity of AI-generated content.
  4. Human oversight: Implement review processes where subject matter experts validate outputs before distribution.

By addressing these challenges, organizations can harness generative AI’s power while safeguarding their most valuable asset: human knowledge.