We’re currently looking at automated full game generation with three different systems: ANGELINA, Cillr and TekaTekiTech. “Teka Teki” is Javanese for Puzzle and TekaTekiTech is our system for producing puzzle games automatically. The games largely involve getting pieces onto a board, with things changing dynamically on-screen, and can be really challenging and engaging. The games were demonstrated at the 2015 Dagstuhl symposium on Artificial Intelligence and Games, and were well received. Development has moved from a desktop version of the software to an Android port, and a tablet version is now nearly finished.

Here is a paper that was presented at the Computational Creativity and Games Workshop at the 2015 International Conference on Computational Creativity:
The TekaTekiTech System for Automated Minigame Generation: First Steps

And here is the talk that we presented for the paper:

With TekaTekiTech, in addition to investigating how software can be used to generate entire mini-games, we will also look into building a community of players for these games. In particular, we are interested if gamers will play a puzzle in the knowledge that it might not be solvable. Obviously, this would be seen as terrible game design if done by a person. However, the software is really just finding these games in a space of puzzles, and can be seen as akin to how mathematicians find unsolved problems in pure mathematics. Mathematicians toil for years on conjectures that may turn out to be false, in the knowledge that they will be the first person to know a fundamental mathematical truth if they are successful in proving the theorem. Will gamers play a puzzle that might not be solvable, with the chance of glory in being the first ever to solve the puzzle?