Humans Do Not Record Reality Perfectly: The Errors We Bring to Every Decision

I remember a board meeting, some years ago, in a company that had asked me to observe rather than advise. The proposal before the board was an expansion into a neighbouring market, and the room was serene. The finance director remembered the due diligence as thorough. The commercial director remembered the partner as committed. The managing director remembered the market research as conclusive. The vote was unanimous. Eighteen months later the venture had cost the company eleven million euros, and when I was asked back to help with the review, I watched the same six people, with the same sincerity and, in two cases, almost the same words, reconstruct a meeting that the minutes did not quite describe. Nobody lied. That is the point.

Every serious critique of artificial intelligence contains the same silent comparison. The machine invented a fact, stated it confidently, misled its user. Fair criticism, as far as it goes. But notice what the criticism implies: that the human alternative would have remembered correctly, weighed evidence evenly, resisted its own preferences, and known the limits of its knowledge. In thirty years of advising decision makers, I have met perhaps a handful of people who do all four consistently, and none who do them naturally.

This is not an attack on human intelligence, which remains the most remarkable reasoning instrument we know. It is an argument about baselines. The reliability debate about AI is conducted as though the reference point were a rational, objective, well-calibrated human observer. That observer is a fiction. The real human observer remembers selectively, reasons from identity as often as from evidence, and experiences false confidence as a normal state, not a malfunction. My purpose in this article is not to lower the estimate of humans. It is to raise the honesty of the comparison, because decisions made on false baselines are reliably bad decisions.

Start with the faculty we trust most. Human memory feels like an archive: events are filed when they happen and retrieved intact when needed. The research says otherwise. Memory is reconstructive. Each time we recall an event, we rebuild it from fragments, and the rebuild is silently influenced by everything that happened since: what we have discussed, what we have concluded, what would be convenient to believe. The memory updates to fit the story, and the story updates to fit us.

The consequences for business are not abstract. The post-mortem that blames the partner, remembered vividly and in detail, may be accurate. It may also be a reconstruction shaped by the outcome and by who is in the room. The market entry that failed in the retelling was often a more ambiguous event in the living. I have sat in reviews where every participant remembered the decisive meeting differently, sincerely, with total confidence, and I have learned to treat that confidence as evidence of sincerity, not of accuracy. A memory that vivid is a memory that has been rehearsed, and rehearsal edits.

The second error type is better known and less absorbed. Confirmation bias, the tendency to seek, prefer and remember evidence that supports what we already believe, is not a weakness of the unskilled or the careless. It is a feature of normal cognition, documented across every population studied, and it operates most effectively precisely where we feel most expert. The experienced executive does not escape the bias. The experience supplies better material for it.

This is the uncomfortable mechanics of expertise. A career builds a library of patterns that let the expert see instantly what a novice must reason through slowly. The same library pre-loads conclusions. Ask the veteran of three successful ASEAN entries whether the fourth will work, and the pattern library answers before the evidence has been examined. Sometimes the pattern is right, that is what expertise is for. Sometimes the market has changed in ways the pattern predates, and the veteran’s confidence is the last thing to update. Expertise is training, and like all training, it teaches us what to see and, invisibly, what to stop seeing.

The deepest layer is the one few decision makers examine at all. Each of us was trained by our culture, our religion, our education, our family and our political environment, long before we made our first professional decision. This training shapes what counts as obvious, what counts as respectable evidence, what a trustworthy person looks like, and what an acceptable risk feels like. Two competent, honest professionals can examine the same market data and reach different conclusions, not because one reasons badly, but because the lens through which each reads reality was ground in a different workshop.

In our advisory work across Malaysia and ASEAN, this layer is impossible to miss. A decision maker trained in one commercial culture reads hierarchy, relationship and contract through one set of defaults, a counterpart trained in another reads the same signals through a different set, and both experience their own reading as simple reality. Most cross-border business friction lives in this gap, and it is sustained by the fact that each side experiences its lens as neutrality. We do not see our training. We see through it.

The final error type ties the others together. Confidence feels like information. When a professional states a conclusion with calm certainty, we treat the certainty as a signal that the conclusion has survived scrutiny. Sometimes it has. Very often, it has simply been rehearsed in a friendly environment. The correlation between how confident a person feels and how accurate they turn out to be is far weaker than our decision processes assume, and the gap widens precisely on the questions where the stakes are highest and the feedback is slowest.

This is why the most dangerous decision meeting is not the one with loud disagreement. It is the one with serene agreement, where every senior voice confirms the plan and the confidence in the room compounds. A room full of confident people sharing the same reconstruction of the past, the same pattern library, the same training and the same preferences is not a verification mechanism. It is an amplifier wearing the costume of one.

If this article reads as pessimistic, it has failed. The point is almost the opposite. Human error is not a scandal to be concealed. It is a constant to be designed for, and constants can be engineered around.

It is worth saying plainly that none of this makes the human instrument dispensable. Context, consequence, ethical weight, the judgement of what matters at all, these remain ours, and no framework I have built pretends otherwise. The argument is narrower and, practically, more useful: the parts of reasoning we do badly are known, documented and predictable, which means they are exactly the parts a well-designed process can carry for us, whether the partner in that process is a colleague, a structure or a machine.

Build the reconstruction problem into your reviews: ask what the record shows, not what the room remembers. Counter the pattern library deliberately: assign the strongest possible case against your own experience to someone senior. Surface the lens: when a conclusion feels obvious, ask which part of the obviousness is training. And treat confidence as a datum about the person, not about the claim.

Only then does the AI comparison become useful. Once we stop grading the machine against a human standard that never existed, we can ask the honest question, which is how the two error profiles differ and how to arrange them so each side checks the other’s failures. That question, and the framework built to answer it, is the subject of the next article in this series, Human versus Artificial Intelligence: Who Really Hallucinates?, where the comparison is made properly, with measurement rather than conviction. It begins, as all honest work does, with the author examining his own instrument first.

If your organisation wants decision processes designed around how humans actually reason, rather than how we like to believe we do, TAPiO works on exactly this, from board protocol to field verification.

Let us talk.

TAPiO Management Advisory is an international advisory firm headquartered in Kuala Lumpur, supporting companies, government agencies and institutions with market entry, investment facilitation and strategic projects across Malaysia and ASEAN.