Come to think of it, the thoughts below suggest a fun new compression scheme.
First, some background. JPEG is a common lossy image format. It has a problem: JPEG artifacts, noticeable image distortion as a result of having thrown away some details. All lossy compression schemes lose information, and since a compression scheme never really knows what information is safe to discard and what isn't, they all have artifacts. But it seems that if a compression scheme had an idea of what chaos is, and what it looks like, then chaotic details could be discarded without losing anything a human would ever notice.
It would amount to including a macroscopic charactarization of some parts of the image. JPEG, for example, has a particular problem with tree branches. But if the file format included a way to encode, “ all right, the whole background of this image is tree branches,” and a few parameters such as the colors and thickness of the branches, then JPEG wouldn't even need to bother trying to encode all those sharp lines and angles. The JPEG viewer would fill in the background with exquisitely rendered random branches.
It's fun to consider how this idea could be taken to ridiculous extremes. The brightly colored background of a sports photograph could be filled in with randomly generated fans. As JPEG has trouble rendering corporate logos (too many sharp edges and fields of solid color), you could randomly generate those too. They all look more or less the same anyway. A dias could be filled with randomly generated politicians; a movie poster with explosions; a gallery with randomly generated modern art.
Anyway, this would result in much better compression: either fewer artifacts for a given file size; or much smaller image files; or a combination. But the resulting image would be part authentic and partly fabricated based on a high-level description of the original.
Arguably this is the direction our media culture is going anyway.