The More You Know, the Less Certain You Sound: Why True Experts Often Seem the Least Confident

A Counterintuitive Phenomenon

In boardrooms, on social media, and across tech forums, we often notice a strange gap: the people who speak with the most absolute certainty aren't always the ones who know the most. Meanwhile, those who have spent years going deep in a given field tend to reach for phrases like "it depends," "in certain contexts," or "I'm not entirely sure." A thread on Hacker News captured this phenomenon well — "the more someone knows, the less certain they tend to sound" — and it points to a fascinating paradox between knowledge and expression.

This isn't false modesty on the expert's part. It's an inevitable consequence of how knowledge itself is structured. When you understand a problem deeply enough, you begin to see its boundaries, its exceptions, its preconditions, and the parts that remain unresolved. Speaking with absolute certainty, in a sense, is a simplification of complexity.

The Cognitive Mechanism Behind the Dunning-Kruger Effect

This phenomenon is closely tied to the well-known Dunning-Kruger Effect in psychology. Research shows that people with lower ability tend to overestimate their competence — precisely because they lack the metacognitive capacity to recognize their own limitations. When you don't know what you don't know, it's easy to fall into the illusion that you've got it all figured out.

Conversely, as expertise grows, a person becomes increasingly aware of how much remains unknown. The deeper the field, the broader its edges, and the more the questions branch out. True experts haven't mastered all the answers — they know precisely which questions are still open and which conclusions rest on unstated assumptions. That sensitivity to uncertainty is, paradoxically, the hallmark of deep understanding.

The Dunning-Kruger Effect was first proposed in 1999 by psychologists David Dunning and Justin Kruger at Cornell University. Their research focused on students who performed poorly in logical reasoning, grammar, and humor appreciation. The study found that these students not only overestimated their absolute performance, but also overestimated their standing relative to peers. The core mechanism is a "metacognitive deficit" — the cognitive skills required to assess one's own ability overlap heavily with the skills required to perform the task itself. In other words, a lack of competence simultaneously produces a blindness to that lack. It's worth noting that subsequent research has raised some questions about the effect's universality, suggesting that certain results may be influenced by statistical artifacts such as regression to the mean, and that cross-cultural studies show variation in effect size. Nevertheless, its central insight — the co-existence of low ability and low self-awareness — retains broad explanatory power in cognitive psychology.

Confident Tone ≠ Reliable Judgment

For anyone who relies on expert opinion to make decisions, this phenomenon carries a real practical warning. Humans are naturally inclined to trust voices that sound confident and assured — which makes sense from an evolutionary and social standpoint. But when it comes to evaluating information, this tendency can lead to systematic misjudgment.

An advisor, analyst, or self-proclaimed thought leader who only ever delivers "absolutely correct" answers may actually be the least trustworthy. Because most genuinely complex real-world problems — whether in technology selection, investment decisions, or medical diagnosis — are full of conditions and trade-offs. Answering with "it depends" is often closer to the truth than a clean, decisive "just do this."

How to Recognize Genuine Expertise

In practice, there are a few signals that can help you identify who is actually worth trusting:

• Willingness to admit not knowing: Someone who can comfortably say "I'm not sure about this" is usually more honest about the things they do know.
• Providing conditions and boundaries: Genuine analysis specifies "under what conditions this holds," rather than drawing blanket conclusions.
• Presenting multiple possibilities: Complex problems rarely have a single right answer. Someone who can lay out different trade-offs is thinking more completely.

Implications for the Age of AI

This topic is especially worth reflecting on as artificial intelligence develops rapidly. One of the most prominent issues with large language models is precisely this kind of overconfidence — they frequently output incorrect information in fluent, authoritative tones, a phenomenon known as "hallucination." The way models express themselves often lacks the appropriate hedging that a human expert would naturally provide, making it difficult for ordinary users to gauge the reliability of an answer from tone alone.

From this perspective, teaching AI systems to "express uncertainty appropriately" may be just as important as improving their accuracy. An AI assistant that knows how to say "I'm not sure" might actually be more trustworthy than one that always sounds completely certain. Calibration — aligning a model's confidence level with its actual accuracy rate — is becoming an important direction in AI research.

In statistics and machine learning, calibration has a precise technical meaning: when a model assigns 70% confidence to a prediction, that prediction should be true roughly 70% of the time in practice. If a model claims 90% certainty but is frequently wrong, it is called "overconfident"; the reverse is "underconfident." Most mainstream large language models currently suffer from overconfidence, partly because their training objectives optimize for the fluency of next-token prediction rather than the reliability of their outputs. Research directions include introducing uncertainty markers through supervised fine-tuning, using temperature scaling to post-calibrate probability distributions, and training models to actively output refusals like "I don't know." For users, proactively asking follow-up questions like "how confident are you in this?" or "under what circumstances would this answer not hold?" is a practical way to compensate for insufficient model calibration at this stage.

Closing Thoughts: Embracing Uncertainty

The maturation of knowledge ultimately manifests as a respect for complexity. The person who sounds the least certain may be the one who has thought most carefully. In an information environment that rewards punchy takes and craves definitive answers, relearning how to appreciate the honesty and depth behind "I'm not entirely sure" may be one of the most important critical faculties we can develop.

Whether in interpersonal communication or when evaluating AI output, staying alert to unqualified certainty — and listening carefully to cautious, hedged expressions — will generally bring us closer to the truth.