It looks that way to me, less "AI generated gameplay" and more "AI generated minecraft let's play"
For actual AI gameplay, this video was interesting to me

You've got it right. There was already a video of someone "playing" Doom and it was mess with objects and enemies not reacting to weapon fire, randomly disappearing, health obviously not being tracked etc.

This was fucking Doom, a 30 year old game that will run on anything, and they were trying to brag about the energy guzzling treat printer making a horribly mangled version of it.

AFAIK the premise is basically "given the last frame and the current input, what is the next frame?" so there's no object permanence or structure behind it it's just making a video that for any given span of two consecutive frames looks like a video of minecraft.

If the most likely next frames from the dig action being taken is the block removed and the item appearing in the inventory bar then that is what would occur

You'll notice it's not actually got as far as the item appearing in the inventory bar - it doesn't manage to pick up any blocks in the video, and when placing blocks the item counter just jumps around randomly.

I suspect, like with the Doom video BelieveRevolt mentioned, it can only make about 3 seconds of footage at a time too, which is the reason it keeps jumping to new scenes.