You are right pb. So with the hype of AlphaGo and me being an amateur data scientist, I started exploring the use of neural networks in the realm of Go. I wanted to focus in the use of AI as a teaching tool, which I think is lacking in Go. The first thing I thought was how to go about extracting the knowledge a neural network had learn about playing Go. A Go engine tells you which move is good but not why. The idea that has been my top candidate is to create two networks, one to predict the ranking of a player based on move played in a certain position and another to play at a certain strength. I hope this can help players better understand their mistakes and what a higher up (2-3 stones) player would do. I think this small steps would be easier to digest for a player.
So why do I need a OGS game db? To create the two networks I need examples of gameplay from all sorts of players 25k to 9d, and that is what OGS has
Sorry I explained so lighty the ideas. I did it for brevity sake. Im happy to answer questions if anyone has. Im also way open to different ideas on how to use AI to help in learning.