Memorising is not understanding

We have all met the student who knows every exercise in the book by heart and freezes the moment a question is phrased slightly differently. Machines catch exactly the same ailment, and the field has its own name for it: overfitting.

It happens like this. While training on a set of data, the model sometimes memorises too thoroughly, learning even the random details that carry no pattern at all. On exactly the material it studied, it barely makes a mistake. Give it new data it has never seen and the result drops sharply. In other words, it is memorising rather than genuinely understanding.

This is one of the central problems of the whole field, so people use every means to guard against it: holding back part of the data as an “unseen exam” to test with, making the model less complex, or showing it a wider variety of examples, all to force it to grasp the pattern instead of learning by rote.

Our organisations fall into exactly the same trap. A process that runs smoothly under the old conditions can still collapse when the market turns, simply because it was designed for a situation that has passed. Real capability, in the end, lies in understanding why we do it this way so that we can still flex, rather than knowing the steps by heart.

Which process in your organisation runs very well on paper, and may not hold when the circumstances change?

To your desk

Rote learning produces lovely results in familiar conditions and breaks down when things shift. This is as true of organisations as of machines.

A gentle test for a process that is running well: imagine customers or the market turning in one specific direction, and ask whether the process still makes sense. And if nobody on the team can explain why it works, we are probably running on habit rather than on understanding.