You might recall the Twitter feed INTERESTING.JPG, from the Universty of Toronto's deep learning group, that features photographs analyzed and captioned by an artificial intelligence program, to the best of its ability. Samim Winiger has been doing similar research, testing algorithms by having software caption video footage as it plays.
Some of the scenes are decently close, but others are hilariously incorrect. Two captioned videos, made up of clips from movies and viral videos familiar to us, and the funniest screenshots are posted at Medium. Read an explanation of Winiger’s work at Gizmodo.