I agree. On the last point, I'd start with even lower level - objects, how many, how large, etc, and only then try to figure out what those objects are. Also

I agree. On the last point, I'd start with even lower level - objects, how many, how large, etc, and only then try to figure out what those objects are. Also, I think it's interesting that it appears that the first successful AI will come from analysis of visual information (computer vision) not textual (like the recent Twine, or semantic web in general).