Hallucination Frailty - Leadership to AI

There is a quiet strength in three powerful words - ‘I don’t know’. Look back centuries, and one will find that the annals of great leadership, whether it be in sporting arenas, boardrooms or battlefields, is filled with those who were not afraid to utter them. 

On the 1934 Ashes tour of England, Sir Donald Bradman leaned heavily on Bill ‘Tiger’ O’Reilly’s experience with field placements. Bradman was new to captaincy, and instead of doing the wrong thing, he freely admitted that he didn’t have all the answers. So he sought help from his state senior, and the team’s most experienced bowler. By doing this, he didn’t surrender authority. He deepened it. By acknowledging what he didn’t know, he created space for knowledge to flow freely around him. Decades before it became a buzzword, Don Bradman was harnessing the power of collective intelligence. 

Contrast that with the many leaders - corporate, political, sporting or tech geniuses, who mistake false certainty for strength. The need to appear infallible too often drives them to craft responses on the fly, making up  answers to preserve the illusion of omniscience. They win the room in the moment, but lose trust and self-confidence as they navigate the long game. They do this because they don’t get a simple truth, that great leadership is not about having every answer ready. It is about creating an environment where “I don’t know yet, but let’s find out” becomes a rallying cry for harnessing that collective intelligence, rather than a confession of ignorance.

AI - The new Pretenders of Certainty

The fear of being seen as clueless is a deeply held insecurity in the human psyche. And somewhat bizarrely, it has found new inheritors in the digital world. 

The latest generation of AI tools, whether it be ChatGPT, Perplexity, Grok, Gemini, or others, were built to reflect the best of human reasoning. And go beyond it. But entirely counterintuitively, with growing sophistication they have inherited an uncomfortable human trait. It is the compulsion to always pretend to know something. Even when they don’t. 

Ask these large language models a question for which there is no clear answer from existing digital sources or indeed its continual learning thus far, and watch what happens. Instead of admitting “I don’t know,” the tools confidently produce elaborate, entirely plausible sounding fabrications. As if programmed with human pride, they prefer to be wrong rather than be labeled ignorant.

Recently, I put the same query to multiple AI tools. It was to help me find the university syllabus of an academic course my spouse had taken almost three decades ago at the University of Calcutta, a document that had not been digitized on the university website. Two of the tools threw up impressively official looking documents which on closer scrutiny revealed content that was entirely disconnected with the query. It had nothing to do with the queried course or university. The third came back with what seemed to be the actual syllabus, except that included it were courses on usage of advanced technology and AI, none of which had been invented, much less taught at universities in the 1990's. On rabbit holing down the cited sources, I found that despite a specific query, the syllabus was a replication of the 2024 syllabus from a completely different university.

This phenomenon, which engineers call “digital hallucination” is really just digital overconfidence. It is a manifestation of our own human insecurity, which our digital avatars have adopted, nothing but a mirror of our own uncertainties. Like a manager who tries to bluff through a question in a meeting that he is unprepared for, the AI tool fills the silence with something that sounds credible, hoping you won’t notice the holes in its logic. Dig deeper, go back to the sources, and the deceit is laid bare without much difficulty. 

In trying to be infallible, the makers of AI have caused their creations to become victims of their own human insecurities. And now that is being perpetuated to the masses who accept the answers with unquestioning faith in its infallibility, as they would religious beliefs. 

Admitting Ignorance Builds Strength

The irony is that both humans and machines earn more respect when they admit uncertainty. When a leader owns up to not knowing, she invites collaboration, learning, and transparency. When an AI tool is allowed to simply say, “Insufficient information found,” it builds trust and encourages curiosity and triggers deeper questions, rather than eroding faith in its abilities.

Great leaders learn early that credibility comes not from omniscience but from intellectual honesty. The courage to admit ignorance signals self-assurance rather than weakness. It tells the team that she is confident enough in her competence to acknowledge its limits. Vulnerability humanises her. It signals that learning and growth are shared responsibilities.The same principle applies to technology. The next evolution of AI must not only know more but must also know when it doesn’t know. That humility could prove to be the ultimate mark of intelligence. It would make the artificial, real.

Leadership, at its core, is about truth-telling. Whether on the sporting arena, the factory floor, or the global stage, followers quickly sense who is bluffing and who is learning. The former lose credibility fast; the latter earn enduring respect. It is time both leaders and their digital protégés embrace what great sportspersons, scientists, and innovators have always known - that knowing when you don’t know, that is where wisdom begins.