AI errors: can employees really speak up?
An employee may spot an error in an AI-generated summary and hesitate to report it if doing so means questioning a tool backed by senior management. Within SHR’s programme on managing in the age of artificial intelligence, the question is specific: how can managers make speaking up genuinely possible?
What the literature establishes
Amy Edmondson (1999, Administrative Science Quarterly) identifies an association between team psychological safety and learning behaviour. Psychological safety refers to a shared belief that the team is safe for interpersonal risk-taking, including acknowledging an error or asking for help. Applied to AI, this framework directs attention to the interpersonal cost of reporting a problem, without providing evidence specific to these tools.
Managerial openness, beyond words
James R. Detert and Ethan R. Burris (2007, Academy of Management Journal) show that managerial openness is associated with employees speaking up to improve organisational functioning. Their analyses identify psychological safety as a mechanism in this relationship. The issue is therefore not simply inviting teams to report problems, but making it credible that they can do so without excessive interpersonal risk.
The premature conclusion
One might conclude that encouraging everyone to speak up is enough to bring AI errors under better control. Yet this research establishes neither that every objection is justified nor that a team reporting more problems experiences more failures. An increase in reports may reflect greater freedom to speak rather than deterioration in the tool. (our executive and employee training programmes)
What the evidence cannot establish
These studies do not examine generative AI and cannot quantify the effect of a reporting process in a Zurich-based company. Their field findings also do not justify promising that a managerial instruction alone will cause change. Finally, an empty register cannot distinguish the absence of errors from the absence of reporting.
A practical check in Zurich
For a Zurich-based team in financial services, insurance, tech or a European headquarters, one practical application of the SHR programme is to test a register of reported AI errors for four weeks, without copying confidential data into it. For each case, record the date, the observed problem, the person responsible for reviewing it and the reasoned response, then calculate the proportion of reports receiving that response within five working days. This timeframe is a proposed test rule, not a scientific threshold: the indicator checks how reports are handled, not, on its own, psychological safety or AI reliability. To go further: explore the Managing in the Age of Artificial Intelligence training in Zurich, or browse our executive and employee training programmes in Switzerland.
In pictures: Managing in the Age of Artificial Intelligence in Zurich
- managing in the age of artificial intelligence
- Zurich
- research
- training zurich