Engineering the Mind

What separates someone who understands something from someone who can merely reproduce it?

True intelligence is the ability to see directly to the heart of things - to feel it from the inside, whether through empirical knowledge or some unexplained primordial intuition. There are many forms of intelligence (academic, intuitive, emotional) but it’s the combination of these forms that gives us the insight and wisdom to tackle the really big problems.

What I admire in Einstein is that childlike directness, an innocence of perception that gets to the heart of things before the layers get in the way. His discoveries can feel so fundamental in hindsight that you hit yourself on the head asking “why the hell didn’t I think of that?” Children have that quality naturally because they haven’t yet built the layers that filter raw experience into “what should be perceived.” The rest of life is spent trying to dismantle those layers to get back to it.

For most of us, by the time we reach adulthood, our psyche is so marred with unconscious patterning and coping mechanisms that the ability to be truly present is a distant memory. The irony is brutal: becoming more perceptive can demand as much unlearning as learning. A stripping back of the layers we’ve accumulated that prevent us from seeing clearly.

Michael Polanyi captured something essential in 1966 when he wrote:

“We can know more than we can tell.”

Michael Polanyi, The Tacit Dimension.¹

He was describing what he called tacit knowledge - the understanding that lives in the body, in experience, in intuition, and that stubbornly resists being captured in words or rules or formulas. A chef knows when dough is right by feel, not by measurement. A master musician doesn’t think about finger placement; the music moves through them. A veteran programmer doesn’t reason through architectural decisions step by step; they feel when something is off, long before they can articulate why.

This is the knowledge that matters most. And it can only be developed one way: through the slow, unglamorous process of doing, failing, adjusting, and doing again. There’s no shortcut.

The Role of Paradox

Paradox matters because it forces the mind to hold incompatible truths without rushing to resolution. That tension is one of the ways depth forms. Nothing great has ever been achieved without a significant measure of conflict and difficulty. Think of the paradox as fighting a battle in the mind in order to gain strength from the conflict. Without regular battle we all know the troops become restless and complacent, and the last thing you want is for your troops to turn to the booze and brothels - unless you so will it.

While paradoxical thinking plays an important role in asking the right questions, it’s also a trap in its own right. The mind ends up caught in an endless iteration until the scope changes or more information becomes available. Think of it like a computer program deferring intensive work to a background thread. Once you’ve fed your brain the right contradictions and raw material, step back and let it work. You can’t interrogate every stage of the process into happening faster.

Avoid laziness too. It’s hard with all this technology, I know. But the mind that becomes complacent forgets how to visit the extremes, and the internal voice that desires knowledge and action becomes nothing more than the whisper of self-loathing that drives self-destructive tendencies, and is eventually lost altogether.

The gong fu masters didn’t get there by thinking harder. They got there by training the body until the thinking became irrelevant. But the paradox is part of the path - you have to wrestle with the questions before you can let them go.

Recognising and Reprogramming Patterns

Thought patterns, both conscious and unconscious, govern the way we perceive sensory input and stored memories. In essence, reality doesn’t change, but the way we see it does. These patterns are learned and enforced on us from birth - by parents, friends, work, society as a whole. The deeper you look into your own life and the world around you, the more the matrix of patterns that comprise perceived reality will be exposed.

The more patterns we take on throughout life, the more disconnected we become from our source, and the less free we become. This metamorphosis prevents us from expressing ourselves truly, from simply being in the moment.

Patterns can change. Lally and colleagues studied people repeating everyday behaviours and modelled how long those behaviours took to become automatic.² The median estimate was 66 days, with estimates ranging from 18 to 254. That’s a study of habit formation, not a timetable for rebuilding your psyche. But it gives a practical point some weight: what you keep practising can become what you do without thinking.

The practice is straightforward, though not easy. Calm the mind, relax the body, and locate the pattern as something with weight, texture, emotional charge - not an abstraction. Watch how it bends your perception and drives behaviour you didn’t choose. Then explore it; follow the awareness wherever it goes, whether the feeling is pain, fear, rage, or loss. Sit in those feelings and let them be there. When a pattern loosens, the clarity can feel like weight lifting off.

Then act in direct contradiction to the conditioning. The more often the better. The rule-followers earn their pat on the back, their sub-achievement, and their nice retirement fund. The pattern-breakers earn something else.

No machine can do this for you; no tool, however sophisticated, can sit inside your nervous system and feel a pattern from the inside. This is the domain of the conscious human being willing to do the uncomfortable work of knowing themselves: embodied cognition, tacit knowledge, the one capability that becomes more valuable, not less, as the machines get smarter.

What the Machines Are Missing

We’ve built the most sophisticated pattern-matching systems in human history and we’re calling them intelligent. They can produce any shape you ask for - every form of text, every style, every pattern of reasoning that’s ever been written down. The shapes are extraordinary, so good they fool most people into thinking there’s a state behind them.

I don’t mistake that performance for the thing I mean by intelligence. A machine can hand me an answer. It cannot hand me the years of experience that would let me know what to do with it.

State, not shape. The shape is the movement you can copy. The state is the condition from which the movement works. A language model can produce a beautiful explanation of balance. Stand in front of someone who knows how to move you and find out how much of that explanation you actually possess.

This is what I want the debate about artificial general intelligence to confront. They argue about benchmarks - can it pass this test, can it score in this percentile? Those measure performance. The harder question is whether a machine can develop state: a felt understanding of the problem, an internal condition from which appropriate action emerges. The masters made that distinction real for me through a body, a practice, and years of contact.

Polanyi saw it from the other direction in 1966: we know more than we can tell. A chef’s feel for dough. A programmer’s architectural intuition. There is understanding in those acts that the explanation never quite contains. You can scrape every description of a skill from the internet and still leave the person reading them unable to do it.

The danger isn’t that the machines will bridge that gap. The danger is that we’ll forget it exists.

Developers are accepting output because it looks right without understanding it, skipping the struggle that builds understanding, and the understanding isn’t forming.

Every time you accept an answer instead of sitting with a question you need to understand, you skip a chance to develop judgement. The education system was already failing at this - cramming minds full of facts and figures rather than teaching system, technique, and strategy. AI gives that failure a much faster delivery mechanism.

What Are You Actually Practising?

There is a difference between using expertise and acquiring it. I can ask a model to work on something familiar and have a reasonable basis for judging the result. In an unfamiliar area, the same fluency can make it much harder to see what I don’t yet understand.

An experienced engineer can recognise a state boundary or question an abstraction because they’ve had to live with the consequences of those choices. A learner may produce working code without getting that opportunity. Neither person is permanently an expert or a beginner. The distinction changes with the task.

Anthropic’s January 2026 experiment put 52 developers through a learning task using an unfamiliar Python library. The reported immediate quiz averages were about 50% with AI assistance and 67% without, a 17 percentage-point difference. The completion-time difference was not statistically significant. The study tested learning immediately after that task; it leaves the longer-term question open.⁵

A different study makes the question more interesting. A purpose-built AI tutor in an introductory physics course produced stronger immediate learning than the compared active-learning lessons. Of 233 enrolled students, 194 met the study’s inclusion conditions. The tutor used structured problems and pedagogical prompts.⁶

Those results don’t cancel each other out. They describe different activities. Getting a completed solution, working through a problem, predicting an outcome, and explaining an error are different things to practise. Calling all of them “using AI” throws away the part that might explain the result.

This is familiar from training. You can copy an instruction without being able to use it under pressure. A good teacher can make the difficult part clearer; they don’t have to make the student struggle with every irrelevant confusion first. Difficulty on its own isn’t education.

For something I intend to understand, I want to be able to explain the mechanism without the answer in front of me. Change a requirement. Predict what should happen. Recognise when a plausible variation is wrong. Those are different checks from whether the original task finished.

For something I already understand and simply need done, delegation can be entirely sensible. I don’t need to reconstruct every abstraction every morning to prove I deserve to use it. The responsibility is to know which situation I’m in and what understanding the decision requires.

The Override

Sitting still and thinking while the rest of the world rushes to automate every cognitive act? That’s the most radical thing you can do right now. It doesn’t mean rejecting the tools; refusing them outright is just hiding from the era you happen to live in. It means insisting on understanding before acting, choosing depth over speed, training your mind with the same discipline a martial artist trains their body, because the mind is the instrument, and an untuned instrument produces nothing worth hearing.

Nobody develops that kind of depth by reading about it. It comes through decades of practice, failure, frustration, and slow, painful refinement.

The mind is programmable. The question is whether you’re writing the program, or whether you’ve handed the keyboard over and stopped paying attention to what’s being written.

Remember that everything considered normal today was once considered outrageous. At best people will think you’re a savant, and at worst they’ll think you’re an idiot. Maybe a little of both.

But who really gives a shit what people think anyway.

References

1. Polanyi, M. (1966). The Tacit Dimension. University of Chicago Press. ↩

2. Lally, P. et al. (2010). How are habits formed: Modelling habit formation in the real world. European Journal of Social Psychology, 40(6), 998-1009. ↩

3. Kosmyna, N. et al. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant. Preprint, version 2, revised 31 December 2025. ↩

4. Becker, J. et al. (2025). Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity. See also METR’s February 2026 methodological update. ↩

5. Anthropic, How AI assistance impacts the formation of coding skills (January 2026). ↩

6. Kestin et al., AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting (2025). ↩