Interesting Idea

I have an interesting idea for someone into chess who has knowledge of AI. My idea is to get a database with millions of games and analyze these games using AI looking for those positions that repeat most often. Take, say, the top thousand of these positions, make a quiz out of them, ask the person taking the quiz to figure out the best moves to play in each of the positions and then in the answers to the quiz publish the moves that the original master players played. This is something that someone might publish in chess.com or make a separate website containing them. If someone does pick up this idea, do let me know. If the person who does this has access to super computer it would even be more fun!

I think you should be asking "What are the top 1000 chess tactics".

The interesting thing about chess tactics is that there can be multiple board positions that are essentially the same tactic. You can take a minimal tactic from the endgame and just add more and pieces to the board in order to raise it's difficulty of finding the tactic. -Not that it would that that much harder, but technically it would be slightly more difficult to find if there were more of pieces on the board that wasn't part of the tactic.

Oh, that's right, there is a training program that does JUST THAT.

I'm not looking just for tactics. I know there are tactic trainers available on chess.com as well as other websites and I do use them.  I am also interested in middle game positions, what middle game positions arise most frequently in actual master level chess games and how do chess masters respond to them?

That is so true. 

Perhaps the OP needs to filter by length of the variation. Perhaps look for variations with a minimum of 20 moves.

Some lines, like the Roy Lopez, have book moves close to 30 moves deep.

I don't about SCID or Chess Assistant,  but Chessbase has their Opening Encyclopedia database.  

I am looking for middle game positions. I cannot think of any specific variations off hand but I do know that  often different opening variations lead to the same or similar middle game positions. It is these repeating middle games that interest me. Also often entirely different openings lead to the same endgames. I am also interested in those endgames. That is what I would look for using Artificial Intelligence. Unfortunately my own computer knowledge is not great enough for me to know how to program a computer to do this kind of research. It would be interesting to see how various masters handle these positions.