In producing the camera-ready version of our paper "Increasing the Action Gap: New Operators for Reinforcement Learning" I unfortunately left out Amir-massoud Farahmand's excellent 2011 NIPS paper, Action-Gap Phenomenon in Reinforcement Learning, which introduces the use of the term to reinforcement learning and provides important theoretical insights into the role of the action gap in practice.
Incidentally, the term is unrelated to this, this, and this.