I got a kinda wonderful idea. Inspired by Gmic -Solidify filter found in GIMP image software. With that one can repair an image from very few pixels scattered randomly over the image.
Apply that to machine learning or AI. What if you call the pixels knowledge. It could then be possible to fill in the gaps in the knowledge with only a few percent of knowledge that you know.
My lightbulb moment.
I speculate further that knowledge can inspire knowledge. Then you can use any historic information there is. That is. Even if you are looking for physics information. Other seemingly non related knowledge from other fields could generate good inspiration data for generation.
Just some little information that is known
Calculate what the rest might sound or look like. The Gmic Solidify filter.
Original image for comparison