OK, so you want to set a village on fire. Cool. It burns down, several NPCs are killed, and the ones that weren't killed scatter for a while and eventually come back to rebuild. What if one of the NPCs in that town is important to the main story somehow? What if you get to Whiterun and decide to just straight up murder the Jarl because you don't like the way he talks to you? In your perfect "choices matter" scenario, you should be able to do that - but what are the consequences? Do the guards in town mindlessly attack you for killing their leader? If you're strong enough, you could murder them all as well - then what? Does the town hold an election to decide on a new Jarl? Is there a line of succession? What role does their form of local government take? The town needs its Jarl, right? Maybe there is political turmoil or restlessness for a while, but eventually things go back to status quo. Functionally, how different is the game now that you've killed the Jarl? You've either cucked yourself on missing out on all the quests he would have given you, or the game would simply re-create those quests with a different NPC at the helm. And this is just a single decision you could potentially make, in a game where you are potentially making many many decisions like this every few minutes.
Dungeon masters in D&D do this kind of stuff all the time. They want to give their players the full freedom of choice and consequences for their actions, but doing so typically requires at least a few hours / days between campaign sessions where the DM can go off and form new narratives around the decisions being made by the player(s). Sometimes entire stories need to be re-done and rewritten. Spoiler alert, most DMs usually will just recycle an existing story and massage it to fit the new constraints the player made for them. And while computers can do this kind of stuff faster than humans typically can now, building an entire machine learning narrative around a single player's single game play through (while also building new art assets, voiced dialog, etc. on the fly) is incredibly expensive. To make something like this financially viable in 2024, we would have to go back to the "pay per minute" model of gaming - even a monthly subscription fee wouldn't be enough to cover costs, unless the monthly subscription fee were closer to $500-$600, maybe more.
The issue, even with a proper AI / ML implementation, is always going to be crafting a narrative. Games like Morrowind had no problem letting you kill important NPCs (though it would warn you that saving after doing so would ultimately doom your save file to a non-winnable state). Something like this is almost certainly a requirement for a game that has proper "world changing" consequences. If you alter the state of the game in such a way that you're no longer able to complete the original narrative of the story (and there are a LOT of ways of accomplishing this), then you have to be okay with the game you're playing either telling you Morrowind-style "hey dummy, you cucked yourself out of moving the story forward" or you have to be okay with the game not having a story at all in the first place. Even if you do control these elements through ML, as in the game "makes the story for you and crafts it around your choices", then you're still just playing a sandbox game - albeit a more fancy one that anything we have today.
We'll eventually get there, but where we are technologically right now is something more like a text-based "choose your own adventure" novel. Even by the time the technology catches up to let us play something like your perfectly envisioned version of Skyrim you describe in the OP, there will be people complaining that the gameplay is too ancient as it's basically a 2010 game, and they wished AI could give them something more modern.
The real question should be "how far do we have to go to trick people into believing that their choices have consequences?" For most people, we're already at that threshold. Add another 1-2 layers of complexity, and you've now captured up to 99% of people on the planet. Why do something difficult and expensive, when most people will easily be fooled by an imitation?
