PID controller is a very simple device described by a few math operations. It might be too simple to drive any sort of AI. One of the typical PID controllers is a thermostat in refrigerator that enables cooler when temperature goes higher than some threshold value. It just measures the difference between current and desired temperature (this value is called error) then does few math operations on this value and outputs “control signal”, which is again just a value. For cooler it can be an actual electric current that enables it.
Just PID on it’s own is not very useful, the challenge is in understanding how to measure error and what to do with the control signal, when you solve those, the PID code is trivial. The other problem is that parameters P, D and I are not intuitive, they don’t really have any relation to a real world and not transferable from one device to another. They are very low level too which means that only very basic behavior can be done using them. If you drop integral (I in PID) coefficient then it becomes a spring - how much AI can you do with a spring? 
For AI controlling physics based vehicles you can look into example of “Steering Behaviors” it’s a collection of simple algorithms, that are kind of like PID controller but made for more specific purpose:
https://www.red3d.com/cwr/steer/
Just google it if you need more articles or other examples.