I’ve always been mildly bothered by the term unstructured data. I finally realized why.

First of all, it’s not unstructured. It’s just not rigidly organized into rows and columns, files and fields, or according to a fixed schema. But that doesn’t mean it has no structure. White noise is unstructured. Entropy is lack of structure. If there is no structure, it’s not data. Unstructured data is an oxymoron, like inedible food If it’s really inedible, it’s not food. Right?

What people want to call structured data is that which is readily consumable by a computer. Interestingly, if they have a big fixed format file that’s full of errors, missing data, and broken pointers, they still call that structured data, even through it needs all sorts of algorithmic manipulation before it’s useful. So apparently, structured data is that which appears to be in a form readily consumable by a computer, even if it’s not really.

I think what is often called unstructured data is really data that is structured, but for a purpose or consumer different from the one at hand. The most obvious example of this is free form text. Human language is hardly unstructured. It’s highly structured in a way that works pretty well for human communication, but challenging for computers.

And the reason I’m going on about all of this is that the term unstructured data masks the huge advancement that is at hand. Computers are rapidly learning how to use data in forms that were, until recently, exclusively consumable and interpretable by humans. Text. Speech. Images. Video. Not unstructured. Human structured.