Its running a recurrent neural network that describes the scene in front of the webcam…in real time.
For some examples of what this sort of neural network can do, go here: (http://www.contriving.net/link/gw). You'll find a bunch of images with captions wrote by a computer.
That's amazing. I never saw the 'black and white photo of a plane in a wreck' 😉 but overall that was fantastic. It only struggled when there was no obvious object in the frame such as when panning or looking at the street/floor.
0_o