Losing good people is always a shock. Anna was creative, proactive and committed: all the things we needed in our team. Andrew was more of a wildcard. A recent hire from a different industry, he was struggling to adjust. I trusted that Anna would keep doing high quality work while I helped Andrew find his feet. ALSO READ: The antidote for office politics So, I was blindsided when Anna resigned. Her reasons for leaving were even more surprising. She was convinced we wanted to replace her with Andrew. Anna interpreted my trust as neglect and my support for Andrew as favouritism. She didn’t ask how I saw the situation. If she had, I would have said that I valued her work and was only focused on Andrew because he was new and needed support. ALSO READ: Automate admin, not trust I often see misunderstandings like this in my leadership coaching. My clients are consumed by situations where they have few facts and a lot of assumptions. They tell themselves stories to explain others’ behaviour but rarely ask what’s going on. This can spiral as they look for evidence to confirm initial suspicions: a delayed response to an email quickly becomes an act of sabotage, rather than someone just being disorganised; being left out of a key meeting is evidence of disfavour, rather than just an innocent oversight. The human brain is good at creating narratives with incomplete information. This helps us make sense of the world and feel in control. But it is also influenced by past experience. For example, someone who has been disappointed by someone in the recent past is primed to look for evidence of further disappointment, whether this outcome is likely in their current circumstances or not. Strangely enough, the errors we’re seeing with AI are similar. Large language models like ChatGPT appear to confirm people’s biases by anticipating the kinds of answers they want. This can have strange consequences. In 2025, UK police banned supporters of the Israeli Football Club Maccabi Tel Aviv from a match against Aston Villa. The police concluded that Maccabi fans were high risk due to violence that had apparently erupted during a previous game with West Ham United. Only, the match against West Ham never happened. It was hallucinated by an AI tool that contributed to a police intelligence report. If AI understands that users want stories about violent fans, it serves up stories about violent fans. Human intelligence works in a similar way. If the brain senses that a particular story might align with a deep-seated belief, it may interpret available facts to support it. The antidote for all this distortion is becoming curious about different perspectives. But fact checking personal stories feels vulnerable. It takes courage to say: I’ve been telling myself a story about this situation and I want to check whether it’s actually true. But the cost of avoiding these conversations is enormous as people react to imagined threats instead of real ones. The irony is that the discussions people fear most are often the ones that bring the greatest benefit. As we saw with Anna, checking facts can at least help you to make decisions that are informed by real rather than imagined concerns. In my next column, I’ll talk about how to make tough discussions feel less scary and more productive. Learning to test inner narratives is an essential skill that all leaders should master to prevent human hallucinations from clouding decision making and causing havoc at work. The writer is the CEO of Transforming Engagements Ltd, a Rwanda-based consultancy specialising in organisational culture, leadership development and performance improvement.
