Not real-time, nor a shooter, but when developing a car combat game we did use an evolutionary process to build the AI. Enemy cars could see the edges of the track and locations of other vehicles.
However things like how hard to break, when to start, how hard to steer around opponents, when to target opponents vs. focus on driving were all setup as variables. Then the AI was setup to run laps all night long slowly tweaking the variables. Each morning we'd look at the previous nights data, find the ones that had the best lap times/kills and then mutate that again. The end result was an AI that could be dumped into any track and be able to compete.
This was all then baked into the game, so it didn't learn from playing against the player.
There are several sports games that track player tendencies and then work to anticipate those choices, not sure if anyone has shipped adaptive AI in a shooter though.