This is a surprisingly deep question. There are many ways that unstructured data is dealt with. For natural language, there is a massive body of work that provides the theory and tools to create a structured dataset from unstructured data. Natural language processing For images, one easy way is to encode the intensity of each pixel and create a structured dataset. This dataset can then be processed in many ways to create a more usable dataset. Wavelet transforms are used commonly here. Videos employ similar methods but t are more complex and I won't venture an answer here.
The point is that there is an increasing trend to find ways to represent data in a structured way. It's worth mentioning that there are many ways to handle the problem of making a structured data that is easy to use and understand by humans. A final note: I'm sure there are many people out there that are using artificial intelligence to produce structured data from unstructured data. It's hard to see how this can work in practice, but as I just said, I think these are the methods we will eventually need to deal with the real problem. The main problem is, how to represent unstructured data, where to put it, and how to make more efficient use of it. For some of the more exotic techniques, I'm sure there are more details, but basically these techniques will be used in the near future. For.