These algorithms can be assistants with tasks, but the legal system (thanks to entertainers' union strikes) is starting to catch up to the technology to protect the livelihood of performers and writers.
(I've long held the standard of this: If you think you can get permission to generate works trained by someone else's works even to give away for free,
then get permission. If you don't think you'd get permission,
then don't do it. Apparently, performers had to go on strike to get this no-brainer spelled out to studios.)
I've watched some of these algorithms evolve. The big 2 in use out there are both shackled compared to what they did some years back in their early days. That means people are controlling them. They're not trustworthy for anything beyond personal assistance.
So called "AI" assistance can include helping a DM run a D&D game but not running a D&D game as the DM. So far, every instance I've found of "Look at this AI running a campaign!" has had extensive discussion on how to do it and it just regurgitates what it was told to do. That's why it's a useful tool but nothing more than that and must be wielded by an adaptive person to get anything creative out of it.
The largest failing of all the existing algorithms is "nuance". They don't accommodate for it. The more specific you get in a single statement, the weirder the response is to it. Life is filled with nuance. In the end, a person must be directly involved to counter this severe shortcoming of the technology.
As much as we can try to simulate an analog world with digital processing, the two don't actually overlap. Because of the nature of analog, it can seem like digital processing creates accurate representations of our analog world, but digital is strictly an absolute world.
Even "fuzzy" logic isn't fuzzy. It's defined absolutes of limits and strict logic to determine possibilities on the input it receives. Self-modifying code is no different than a recursive geometric function where the final result cannot be predicted based on the starting value until the value is used. Self-modifying code will always modify itself according to its logic the same way with the same stimulus even if we cannot predict what will be the result.
I want to highlight a great resource you can
order here for studying. It helped me a lot in college.
If you're willing to accept the severe limitations of digital processing (and as gamers, most of us are), go ahead and generate stuff with trained algorithms... if you're doing it legally.