The Error Beneath the Error - Double-Loop Learning

Sep 2

7 min read

Why serious improvement begins when a system stops defending its own assumptions.

Someone in the meeting points to the number, and for a moment the room relaxes.

It has moved in the right direction. That is enough to lower the tension. The awkward questions can wait because the evidence seems to have settled the matter before anyone has had to examine it closely. Nobody has to say that the work feels thinner than it did three months ago, or that people have become unusually careful with what they admit, or that the target now has a strange power over the way everyone describes reality. The number has spoken, and because numbers sound cleaner than unease, the room lets it speak for everyone.

This is how many systems protect themselves from learning. Not through stupidity, and not always through bad faith. More often, the failure is subtler: the organisation becomes attached to the surface that makes it look intelligent. It keeps correcting visible behaviour while leaving the deeper logic untouched. The same problem returns in a different form because the original conditions were never examined.

There is the error everyone sees, and then there is the error beneath it.

The first usually appears as a practical failure: a deadline slips, a hire does not work out, a campaign falls short, or a team repeats a mistake that should have been resolved long ago. But beneath that visible problem there may be a more consequential one, embedded in the system’s assumptions about what deserves reward, what must be avoided, which outcomes count as evidence, and which questions are too uncomfortable to raise. It may also live in the knowledge people carry privately but have learned not to voice because the organisation has given them no safe or legitimate way to do so.

Progress depends on whether a system can travel backwards from the visible mistake into the hidden logic that produced it.

When the measure starts giving orders

A metric begins its life as a servant. It is there to make something visible: sales, retention, speed, attendance, output, growth. The danger begins when the measure stops describing reality and starts commanding it.

Donald Campbell warned that social indicators become vulnerable to corruption once they are used to make decisions. Goodhart’s Law carries the same suspicion in a sharper form: when a measure becomes a target, it can lose the quality that made it useful as a measure. The insight is uncomfortable because modern organisations depend on measurement. Without numbers, people drift into taste, politics, and anecdote. With numbers, they may drift into a different distortion: the quiet substitution of the measured thing for the meaningful thing.

The problem isn’t measurement itself. It’s what happens when we start treating the numbers as if they can’t be questioned.

A company may begin tracking response time because customers deserve prompt attention, yet the moment speed becomes the prize in itself, the complicated customer starts to look like a problem rather than a person. The same drift can happen in education, where a school measures attainment in the hope of understanding learning, only to find that the measure gradually becomes a substitute for it and the parts of education that cannot be easily counted begin to feel expendable. A creator can fall into a similar trap: views may initially offer useful evidence about resonance, but over time the platform’s signals can become so persuasive that the work is shaped around whatever earns immediate attention, at the expense of the slower, less visible effort required to build a body of thought.

The metric has not become evil. It has become lonely. Cut off from judgement, context, and purpose, it begins to govern more than it can understand.

Steven Kerr’s classic essay, On the Folly of Rewarding A, While Hoping for B, belongs exactly here. Organisations often say they want one kind of behaviour while rewarding another. They hope for quality while rewarding volume, ask for courage while punishing bad news, praise initiative while making deviation costly. People learn the contradiction quickly. Human beings may ignore slogans, but they rarely ignore consequences.

An incentive is therefore never just a reward. It is a private education in what the system truly respects. This is why incentive design requires more humility than most leaders give it. Deci, Koestner, and Ryan’s research on extrinsic rewards showed that rewards can weaken intrinsic motivation under certain conditions, particularly when they feel controlling. Gneezy and Rustichini found that payment does not automatically improve performance and may even reduce it when the incentive changes the meaning of the task. The useful lesson is not that money, praise, recognition, or punishment should disappear. The lesson is more delicate: every incentive edits the story people tell themselves about why the work is worth doing.

Once that story changes, behaviour follows.

The system that cannot be offended cannot learn

Chris Argyris’s work matters here because he understood something that many organisations still avoid: learning is not the same as improving a result. An organisation can miss a target, adjust the process, push harder, and appear to be learning while leaving the deeper assumptions untouched. The visible correction may be real, but it stays shallow. The system becomes better at preserving its original logic.

Argyris described organisational learning as the detection and correction of error. At first, that sounds almost too obvious to be useful. The force of the idea appears when he distinguishes between two kinds of correction. Single-loop learning corrects an action so the system can continue pursuing the same goal in the same way. If the room is too cold, the thermostat raises the temperature. Nothing about the setting itself has been questioned. Double-loop learning goes further. It asks whether the setting, the standard, or the governing assumption deserves to survive.

That distinction is essential for any serious discussion of incentives, experiments, and preventable mistakes. Many teams are not short of effort; they are trapped inside the wrong lesson. If a company rewards speed and then complains about quality, single-loop learning tells people to work faster and be more careful. Double-loop learning asks why speed became the protected value in the first place. If a team keeps missing deadlines, the shallow response is to demand better time management. The deeper response asks whether the plan depends on a fictional version of capacity that only exists in meetings. If initiative becomes performative, the useful question is not only who is pretending. It is what kind of recognition system made performance more attractive than responsibility.

Argyris and Donald Schön also developed the distinction between an organisation’s espoused theory and its theory-in-use. The first is what the organisation says it believes. The second is what its behaviour reveals under pressure. This distinction gives the article a sharper human edge because most hypocrisy in organisations is not announced as hypocrisy. It appears as a gap between language and consequence. A team may say it values honesty while making disagreement costly. A founder may say they want independent thinking while rewarding people who bring reassurance. A company may speak about learning while quietly treating every mistake as an inconvenience to reputation.

The painful part is that people usually learn the theory in use faster than the espoused theory. They learn what can be said safely, which risks are worth naming, when silence is wiser than accuracy, and how much truth the room can tolerate before it becomes socially dangerous. Over time, the organisation may still sound intelligent, but its intelligence has become defensive. It knows how to explain itself. It no longer knows how to be interrupted.

This is the uncomfortable part of Argyris’s work. He did not treat intelligence as protection from defensiveness. In fact, he argued that successful professionals can be particularly poor learners because they are so unused to experiencing failure as useful information. Their competence becomes part of their self-image; when evidence threatens it, the mind does not calmly open for revision. It protects itself. It explains, redirects, rationalises, and moves the discomfort away from the centre of the room.

That pattern is not hard to recognise. A strategy meeting can discuss weak results for an hour without touching the decision-making habits that created them. People can speak fluently about accountability while avoiding the one question that would make accountability real. A team can keep improving the surface of its work while quietly preserving the belief that its way of seeing the problem is already correct.

The scientific method offers a different ethic because it lowers the status of the first answer. A hypothesis is useful precisely because it is not sacred. It gives thought a shape that reality can test. Observation, experiment, evidence, revision: the power of the process lies in its refusal to confuse conviction with knowledge.

The Plan-Do-Study-Act cycle carries that same humility into practical work. Its discipline is not simply to plan and act, but to study what happened before the next action hardens into habit. The word “study” matters. It asks for more than checking whether the result was convenient. It asks what the result reveals. A failed test may expose a bad assumption. A successful test may reveal a fragile condition that will not scale. Even a partial result may contain information that a more theatrical culture would miss because it is too busy deciding whether the plan looked good.

Real learning begins when evidence is allowed to alter more than behaviour. It has to reach belief.

That is why correction feels so socially expensive. A plan is easy to change before it has been defended. Once it has been named, praised, funded, scheduled, and repeated in front of other people, revision can feel like a loss of face. The organisation may say it wants learning, but the room may still punish the moment when learning becomes inconvenient.

So the deeper question is not whether a system collects feedback. Many do. The question is whether feedback is allowed to accuse the system of misunderstanding itself.

When it is not, improvement becomes cosmetic. The report changes, the meeting changes, the wording changes, and the same assumption survives underneath. When it is, the organisation develops a rarer kind of intelligence: not the confidence of always knowing what to do, but the capacity to let reality revise the map.