Why You Keep Doing the Wrong Thing Even When You Know Better

Learning updates what you know, understanding updates your reasoning, and doing requires your automatic responses to change, and these are governed by different systems. Insight alone does not install a new default. Research on habit formation shows automaticity grows through repetition in a consistent context, often taking eighteen to two hundred and fifty four days, and the old behaviour keeps winning whenever real triggers arrive before the new one has been rehearsed.

Why the hardest system I have ever tried to upgrade is myself

Over the past three years I have been getting a lot of help with something far more important than technology: how to be a better dad and a better husband. Here is the strange part. Almost from day one, what I was being taught made sense to me. I understood the logic, I could see why certain behaviours were unhelpful, and I could even see the positive effect when I managed to respond differently. So the obvious question was this: if I understand it, agree with it, and can see that it works, why don’t I just change? That question bothered me for a long time, because I genuinely wanted to change, and yet when the actual moment arrived I often couldn’t remember what I was supposed to do, or I remembered it five minutes later, or I knew exactly what I should do but somehow still did the old thing.

Eventually I realised I had been confusing three completely different things: learning something, understanding something, and doing something. They are not the same, and there may actually be a fourth stage beyond them, which is becoming someone for whom the new behaviour is the default. That final stage is much harder than I ever appreciated.

1. Learning changes what you know. It doesn’t necessarily change what you do.

I have spent most of my career in technology, and technology has changed constantly. New programming languages appeared, physical infrastructure became virtual infrastructure, virtual machines gave way to containers, data centres moved into cloud platforms, architectures changed, security models changed, mobile changed everything, and AI is now rewriting assumptions that were considered almost sacred a few years ago. I have always found this kind of change relatively easy. Sometimes I have had to throw away years of accumulated knowledge because the underlying technology moved on, and that has never particularly bothered me. If the new thing is demonstrably better, I learn it, experiment with it, and move on.

So I assumed human behavioural change worked in roughly the same way. Old model discovered to be defective, install new model, restart, continue operating. Unfortunately humans appear to have a somewhat less elegant deployment pipeline than that, because learning often updates our knowledge without updating our automatic responses. Psychologists describe something closely related as the intention-behaviour gap, and research makes that gap surprisingly visible. Thomas Webb and Paschal Sheeran reviewed forty seven experimental tests and found that interventions producing a medium to large change in what people intended to do produced only a small to medium change in what they actually did. Changing intention mattered. It just mattered considerably less than we intuitively expect.

Imagine discovering that an important piece of software has a design flaw. You rewrite it, the old version disappears, and the new version takes its place. Human beings don’t seem to work like that, because the old code remains surprisingly available. A response that has been repeated hundreds or thousands of times becomes associated with particular circumstances, and the environment itself can begin triggering the response with relatively little conscious deliberation. Being a good husband or father is not an examination you sit while calm, well rested, and holding a textbook. The test arrives unexpectedly, when someone is upset, you are tired, you feel misunderstood, and you are trying to solve three other problems at once. Suddenly the old behaviour receives production traffic, and the old version has an enormous advantage, because it has been running for years while the new version was only explained to you on Tuesday.

Technical learning usually gives me the opportunity to consciously select the new approach. I can read documentation, I can stop, I can experiment, I can inspect what happened, and if something fails I can debug it before it matters. Relationships don’t provide that luxury, because the behaviour happens while the system is running and there is no maintenance window. You have to notice the trigger, regulate whatever reaction has appeared, remember the better response, and execute it before your old response gets there first. I had treated behavioural knowledge as though understanding the architecture meant the migration was complete, and it wasn’t.

One of the most useful pieces of research on habit formation comes from Phillippa Lally and colleagues at University College London. Participants chose a new behaviour and repeated it in a consistent context over twelve weeks, and automaticity gradually increased with repetition, though there was enormous variation between individuals and behaviours, with the time required to reach ninety five percent of the modelled automaticity plateau ranging from eighteen to two hundred and fifty four days. Interestingly, missing one opportunity did not materially derail the habit forming process.

Worth being precise about what that study actually covered. Lally’s participants were forming simple daily habits, such as doing fifty sit ups after morning coffee or eating a piece of fruit with lunch. It is not evidence that changing something as complicated as defensiveness or emotional regulation under stress takes eighteen to two hundred and fifty four days. What it does show is that there is no magical universal number, and that behaviour becomes available through repetition in the specific context where it needs to operate. That matters, because we love epiphanies since they feel transformational. You hear something profound and think you have got it, and perhaps you do, but insight may only be the beginning of the installation process. The brain still needs repetitions. The old behaviour has been rehearsed thousands of times, and the new behaviour may need to be deliberately practised dozens or hundreds of times before it becomes readily available when you actually need it. That changed my interpretation of failure, because if I responded badly to something it did not necessarily mean that I had failed to learn. It often meant that I had not yet practised the alternative enough.

2. This is one reason I find DBT so interesting

Dialectical Behaviour Therapy, or DBT, was developed by psychologist Marsha Linehan as a clinical treatment, originally particularly associated with borderline personality disorder, and it should not be reduced to a handful of productivity tricks, since people who need clinical treatment should obviously work with appropriately qualified professionals. But there is an idea embedded in DBT that I think is fascinating far beyond therapy, which is that knowing a skill is not enough, and you have to be able to use the skill in the situation in which you normally lose access to it.

DBT teaches skills across four broad areas, mindfulness, emotion regulation, distress tolerance, and interpersonal effectiveness, and comprehensive DBT even includes between session coaching designed specifically to help people generalise skills from therapy into real world situations. That word, generalise, jumped out at me, because that was exactly my problem. I could understand something in one context, but I couldn’t necessarily retrieve it in another. I had learned the skill in development, but I hadn’t deployed it reliably into production.

3. DBT also contains an extraordinarily useful debugging tool, chain analysis

One of the things behavioural approaches such as DBT encourage is examining the sequence surrounding a problematic behaviour. Rather than saying you shouldn’t have done that, you investigate what actually happened. What state were you in beforehand, what was the prompting event, what did you notice, what did you think it meant, what emotion appeared, what happened physically, what urge followed, what did you do next, and what happened because you did it. This is much closer to incident analysis than moral judgement.

Suppose you want to stop becoming defensive during difficult conversations. Telling yourself to stop being defensive is almost useless as a behavioural instruction, but a chain analysis might reveal something much more actionable, something like this. You were already tired, someone challenged something you had done, you interpreted it as criticism, you felt an immediate need to explain yourself, you interrupted, you started producing evidence, the other person felt unheard, and the conversation escalated. Now we have telemetry, and the target might not actually be to stop arguing. The critical intervention point might be the three seconds between feeling criticised and starting to explain yourself, and that is where the new behaviour needs to live.

4. Don’t stop at goals. Create responses to triggers.

Another field of behavioural research offers an extremely useful technique here, called implementation intentions. Instead of saying you will listen better, you specify that if X happens then you will do Y. Peter Gollwitzer and Paschal Sheeran reviewed ninety four independent tests involving more than eight thousand participants and found that these kinds of implementation intentions had a medium to large positive effect on goal attainment compared with simply having the goal.

That distinction is enormous. Wanting to be calmer with your family is a goal, while noticing the urge to interrupt and choosing instead to stop speaking and ask one question first is executable code. Needing to be more empathetic is a goal, while trying to understand what someone is feeling before offering a solution is a behaviour. Not becoming defensive is a goal, while waiting ten seconds before replying when you feel the urge to explain why you were right is a behaviour. One lives in your aspirations. The other can eventually become part of your programming.

5. DBT’s STOP skill creates a tiny execution window

One DBT distress tolerance skill is conveniently called STOP. Stop. Take a step back. Observe. Proceed mindfully. The basic idea is to interrupt the automatic sequence before acting, create a little psychological distance, notice what is happening, and then choose the next action deliberately. I increasingly think this tiny window is where much of behavioural change happens. You do not necessarily need twenty minutes of contemplation. You sometimes need three seconds, three seconds in which the old behaviour has fired but has not yet been executed.

Take the defensiveness chain from earlier. The critical moment landed at the point where you felt an immediate need to explain yourself, just before you interrupted and started producing evidence. That is exactly where STOP lives. You feel the urge to explain building, and instead of acting on it immediately you stop, take one breath, notice the urge without obeying it, and then choose to ask a question instead of mounting a defence. That tiny gap is tremendously valuable, because once you can see the behaviour arriving you have something you can work with. Before that, you are simply inside it.

6. Sometimes you need to practise the opposite behaviour

DBT also includes an emotion regulation technique called Opposite Action. The principle is more nuanced than simply doing the opposite of whatever you feel. It is used when the emotion, or the intensity of the emotion, does not fit the facts, or when following the emotional urge would be ineffective, and the person deliberately performs behaviour inconsistent with the unhelpful urge. Again this is behavioural rather than intellectual. If your instinct when challenged is to talk more, perhaps the practice is to talk less. If your instinct is to withdraw, perhaps the practice is to remain present. If your instinct is to immediately solve someone’s problem, perhaps the practice is simply to listen. You are not waiting until you naturally feel like behaving differently. You are using behaviour to train behaviour, which is a very different proposition.

7. Measure latency, not perfection

This may be the biggest mistake I made, because I kept measuring myself against the finished behaviour, asking simply whether I got it right, and that makes progress incredibly difficult to see. A better measurement is how long it takes you to notice the trigger. If last year I noticed a mistake two days later, and six months later I noticed it twenty minutes later, and today I noticed it while I was doing it, that is not zero progress. The detection latency is collapsing, and eventually detection can move ahead of execution, which is where choice starts to appear.

The first sign that you are changing may not be that you stop doing the wrong thing. It may be that you notice yourself doing it sooner. Don’t dismiss that. The shrinking gap between execution and detection is where control is being built.

8. Practise when nothing is wrong

There is another lesson here that seems obvious in retrospect. You cannot expect your first serious attempt at a new behaviour to happen under maximum emotional load. We don’t train emergency procedures for the first time during an emergency. Pilots use simulators, sports teams rehearse plays, technology teams run disaster recovery exercises, and cyber teams run incident simulations. Yet with behaviour we often hear something on Wednesday and expect ourselves to execute it flawlessly during an emotionally charged conversation on Thursday, which seems unreasonable. It helps to rehearse the response, visualise the trigger, say the sentence out loud, practise the pause, write down the if then rule, and use the new behaviour during low intensity situations until it becomes familiar.

DBT research provides an interesting clue here. In a study of one hundred and eight women with borderline personality disorder undergoing treatment, increased use of DBT skills, rather than merely being exposed to them, mediated improvements in several measured outcomes. That study is specific to a clinical population and should not be casually generalised to ordinary relationship problems, but the underlying observation is important, which is that the skill has to be used. Reading about push ups does remarkably little for your chest, and behavioural skills appear to suffer from the same inconvenience.

9. Build behavioural observability

Technology taught me another useful lesson, which is that systems improve faster when they are observable, so perhaps behaviour should be observable too, not obsessively, and not as another opportunity to punish yourself, but simply enough to reveal patterns over time. The single most useful question at the end of a difficult interaction is what you will do when this exact trigger appears again. That question matters most, because it converts yesterday’s failure into tomorrow’s implementation intention, which is learning from telemetry rather than simply logging it.

10. Acceptance and change can both be true

This may be DBT’s most powerful idea of all. The dialectical part of Dialectical Behaviour Therapy is partly about holding things that appear contradictory together. I can accept that a behaviour exists and work extremely hard to change it. I can understand why I developed a particular response and decide that it is no longer acceptable or useful. I can acknowledge that change is difficult without using difficulty as an excuse not to change. That balance matters, because if every mistake becomes evidence that you are hopeless you create shame rather than learning, but if every mistake is excused because that is just how you are, nothing changes. The useful position lies somewhere between them, which is that this is your current default, it makes sense that it is difficult to change, and you are still responsible for changing it. That is a demanding position, but I think it is also an effective one.

11. The hardest upgrade is yourself

I spent much of my life believing that once I understood something, change should follow fairly quickly. Technology reinforced that belief because knowledge and behaviour are often closely connected, since you learn the new API, understand the new architecture, and use it. Human beings are different. We carry old implementations around with us. They are triggered by context, strengthened by repetition, and particularly difficult to replace because they often execute before conscious intention has fully entered the conversation.

So I now think behavioural change looks something like this. You learn the principle, understand why it matters, identify the situations in which the old behaviour fires, define the replacement behaviour, create explicit if then rules, practise them while calm, notice failures without pretending they didn’t happen, analyse the chain, repair quickly, and repeat, and repeat again. Eventually something interesting starts to happen. The behaviour that once required enormous conscious effort begins to arrive a little sooner. The pause becomes easier. The new response becomes more available. And eventually, perhaps, what started as something you were taught becomes something you understand, then something you deliberately practise, and finally something that is simply part of who you are. Learn, understand, practise, default.

That last transition is the one I underestimated. Learning can happen in an afternoon, and understanding can arrive in a moment, but changing your default settings can take months or years. That doesn’t mean the learning failed. It means the deployment is still underway.

Research referenced