When an AI-driven decision causes harm, responsibility can be scattered among vendors, data teams, IT, and managers. This diffusion makes it difficult to assign accountability, creating a dangerous "fog" where no single person or entity feels fully responsible for system failures.
Stating a "human is in the loop" is often symbolic. For oversight to be effective, the manager must have the time, competence, and organizational permission to genuinely challenge and override an AI's recommendation, rather than just serving as a liability shield for a pre-framed decision.
Excluding employees from AI adoption is a quality issue, not just a labor relations problem. Workers understand the gap between official processes and on-the-ground reality. Ignoring their knowledge leads to AI systems that seem rational centrally but fail in practice because they are based on flawed assumptions.
The most significant change AI brings to management is not tool proficiency. It's the shift to becoming a governance actor who must interpret machine outputs, ensure procedural fairness, challenge unreliable recommendations, and explain decisions, acting as the human interface for algorithmic systems.