I’m assuming you’re referencing Large Language Models (LLMs), trained on some or all of the BIONICLE text. Like Iylv said, It would be very good at imitating the style of Farshtey’s writing, but it might not make any sense. LLMs are trained to provide satisfactory outputs, not to be correct 100% of the time, so it would be prone to making careless errors and forgetting context.
I’ve experimented with semi-AI-generated BIONICLE content before - using Botnik (which is not an LLM; possibly more procedural pattern matching than any sort of machine learning). This was a “human in the loop” setup where I (the human) would be given a list of words that could follow the current word. People have told me that a lot of the phrasing it came up with was very Greg-like, even though I was intentionally writing anti-Greg content (namely a love triangle between members of the Federation of Fear.)