One more thing I think is relevant to the discussion on the meaning of ‘roguelike’ is the Berlin Interpretation, though I personally think it’s a touch too narrow to be a usable, non-academic definition. Plus roguebasin (where that link is) could probably be placed in the Additional Resources section, being a wiki dedicated to roguelikes.
Yeah, opinions on roguelikes/-lites are definitely very divisive, a problem I think that mostly comes down to prescriptive vs descriptive linguistics. Given that, I think you’ve done a perfectly good job in the OP.
My personal definition of ‘roguelike’ is a game that is turn based, with perma-death and procedural generation, and ideally is also grid-based. A ‘traditional roguelike,’ to me, is more a specific set of games (Angband, NetHack, etc.), rather than a genre, but if you did want to use ‘traditional roguelike’ as a genre, it’d have all of the above, plus be a fantasy dungeon-crawler RPG. I also do think roguelikes and rogue-lites are meaningfully distinct, or atleast should be, even if most people don’t consider them to be. Rogue-lites can be very fun games, but when I want a roguelike, I want a roguelike, not a fast-paced bullet hell whatever. The best roguelikes I’ve played thus far are Cataclysm: Dark Days Ahead (CDDA), and Cogmind. Plus I’ve been thinking of picking up Jupiter Hell and Dead Cells when I can, though AFAIK Dead Cells is more of a rogue-lite than a roguelike.
It’d be dead easy, actually. Don’t even have to actually ban it: For image generating models, every artist whose work is included in the training data becomes entitled to 5 cents per image in the training data every time a model generates an image, so an artist with 20 works in the model is entitled to a dollar per generated image. Companies offering image generating neural networks would near instantly incur such huge liabilities that it simply wouldn’t be worth it anymore. Same thing could apply to text and voice generating models, just per word instead of per image.