Unsupervised learning - is learning based on predicting future, present or past data from current data.
Both use immediate feedback.
Reinforcement learning is based on learning from dynamic datasets as opposed from static datasets.
Games are like dynamic datasets. Of course game theory has a lot to do with deep learning. If the signal you are learning is dynamic, it's related to game theory.
Reinforcement learning is learning from sequences of states/actions and a reward signal. The sequences could be totally deterministic and similar across different runs of the environment, and hence what I'm guessing you're calling "static" ("static" really has no meaning in the ML field). Or not. Go is totally deterministic, for example, but RL is still applied there.
Unsupervised learning is kind of vague, but it's supervised learning except the "desired output" (label) is generated from the input, or some desired statistic is inferred from the input. Inferring past/present or predicting future data is only a subset of the first case.
I'm not really familiar with the literature in this area, so sorry if this sounds dumb. But what you're saying here seems to make sense and sounds almost profound - like there's a vast amount of research (game theory) that's immediately related in some way. Is there somewhere I can read more about the relationship?
Now, it can teach you something about your data, but it is then up to you to turn that into something that can let you predict about your data. (So, typically cluster analysis.)
Right?