[AgentSmith]

There are a lot of youtube videos showing how genetic algorithms can be used in order to generate images. The reason why this is quite cool and interesting is the fact that the image generating process based on the genetic algorithm has actually no knowledge about the target image to be generated. The only information is provided by some scalar fitness score. Only with such fitness scores complex target images can be generated over a number of evolution steps. If primitive scalar values can be used to produce images over time, they could basically also be used to produce mental pictures or dreams of an AGI. Maybe pictures a human perceives are not only encoded as images but also as fitness functions that are used to create images when they are loaded in the consciousness later.  This would mean that we have two completely different ways to represent/encode images. However, I cannot say exactly what the properties and relative advantages and disadvantages of these two different image representations would be and need to think more about this.

[Freddy]

I'd like to know how that one with the little dashes was done, the Eiffel Tower one 

[infurl]

This reminds me of technology that I first read about many years ago, but which seems to have gone away. Does anyone remember fractal image compression? It was software that could find formulae to reconstruct an image such as a photograph using fractal mathematics. The examples that I saw were able to compress very detailed photographs into a very small number of bytes. Even though it was technically lossy compression, the images could be scaled up to a very high degree without appearing grainy. Perhaps the technology wasn't as generally useful as it seemed at the time. It certainly seemed a bit miraculous.

https://en.wikipedia.org/wiki/Fractal_compression

Hmm. Skimming the wikipedia article it looks like the technology has been absorbed into other libraries. Perhaps we've been using it for the last 20 years without realising it.