Yes, but that won’t be ChatGPT. It is just a language model. AI models for the actual problem domain of ATC are another matter - I’m sure that will get easier to do as cloud ML technology advances. Really it’s primarily a training and input issue. And with MSFS Microsoft could get a huge data set to train something on if it so chose. Going to an ML model as the core of a new ATC system for the sim would make a lot of sense.
Edit: (and slightly off-topic, so I’ll keep it short) equally, it should be possible to use an ML model to generate realistic evolving theme-based weather on demand. Imagine being able to say ‘I want realistic summer weather for this route assuming a flight date of 4th July’, for example. That would be awesome for those of us who want to fly with live-like weather but not the current live weather.