Plastic Dinosaur is a fictional robotic velociraptor guided by a set of machine-learning neural nets. David Clement and I are using him as a foil in our exploration of the current state of the art in machine learning, a way to make some interesting corners of the space more accessible to a general audience. After all, what's more accessible than an AI predator with teeth? This episode is the narration of the first article in the series, on the physical structure of the robot. Other articles deal with the neural net architecture, training the robot to play goalie and perceptual gaps we have in understanding what neural nets are doing.