This seems interesting to me in the context of ML research (not that I know anything about that). We have generative AI and this is sort-of the reverse doing compression like turning an interactive movie into an interactive book. I know it's reading the code of the graphics, but that could serve as labelled data for training image to text generation. Image to text isn't new, but if the sequences of images to text could tell a coherent story that could be interesting. Perhaps we need many people playing Doom with running commentary on the evolving story for data.
Or we could do the same for an adventure game which has more story than Doom.
Or we could do the same for an adventure game which has more story than Doom.