← Back to Kortmann BioAdvisory
In Brazilian Jiu-Jitsu, as in science, you start by mimicking. You're taught techniques — how to escape a mount, apply an armbar, or defend a choke — and these become your footholds. For many early learners, it’s all about collecting these techniques, one by one.
But it takes years before you realize: those were just footholds.
🦶 (A quiet pun here — since "toe holds" and leg locks are real submission techniques. Once sidelined as fringe or even risky, leg locks have become a staple of modern BJJ. Why ignore 50% of the body?)
This shift — from memorizing moves to understanding systems — mirrors how we evolve as scientists. Early in your training, you follow protocols. Pipette like this. Run this assay. Follow the kit. It’s only later — sometimes far later — that you understand what not to do. What matters. What fails gracefully. And what never quite made sense but we all did anyway.
It’s the difference between technique and principle.
In BJJ, principles like base, pressure, angle, timing, and control become the invisible scaffolds. They’re what allow advanced practitioners to “teleport” into positions — skipping steps that were once drilled relentlessly. In science, this is akin to intuition: the practiced feel for where to probe, when to change variables, and when to let a result just sit.
The danger, of course, is oversimplification. One person’s "principle" is another’s bad habit. There are purists who argue that we must teach techniques correctly, from the start. And they’re not wrong — just incomplete.
Because what works for one person won’t work for another.
That might sound heretical in science — where reproducibility is king. If a protocol only works for one lab, it’s not a real protocol. But even in biology, variation is the rule. Multiple approaches can converge on the same truth. And insisting on a single technique — a single path — often creates rigidity, not reliability.
This is where the idea of situational training comes in.
In BJJ, instead of teaching a technique, you might drop students into a real position — say, top person in closed guard — and ask them to solve it. One person’s goal: break the guard and pass. The other’s: maintain guard or sweep. They go. They learn. They make mistakes. Then you talk.
It’s messy. Imperfect. Riskier.
But it creates real understanding.
Could science — and even AI training — benefit from this?
In training AI, we often face the same choice: teach through massive supervised data (akin to YouTube instructionals), or build situational learning environments that simulate feedback, failure, and pressure. One teaches breadth. The other builds depth.
And crucially: the latter can’t be faked. Just like jiu-jitsu.
Watching instructionals or matches helps — but only if you’ve felt the pressure. Grappling has to be done. That’s why even black belts study tape after they roll — to refine, reflect, and reframe.
High-level grapplers in Brazilian Jiu-Jitsu sometimes “teleport” into submissions — skipping the expected steps because they’ve internalized the system so deeply. Timing, pressure, angle — it’s all second nature.
Could something similar happen in science?
This isn’t about cutting corners — it’s about knowing which corners don’t matter anymore, based on prior hard-won pattern recognition.
Just like in BJJ:
❌ Don’t skip steps too early.
✅ But once you’ve internalized the system, you can skip steps — or recombine them in new ways.
Science, like grappling, isn’t about memorizing techniques. It’s about feeling the system. Pressure creates pattern. And the best discoveries — like the best submissions — don’t always follow the steps.
It’s also a return to first principles.
Before you can adapt, you need to understand the underlying structure — the base, the pressure, the leverage. The same goes for training scientists, or even machines. A vast library of data is only as useful as the framework you use to act on it.
Which brings us full circle: true learning begins where memorization ends.
— Jens