Segmenting observations from an input stream is key to human cognition. Evidence suggests that humans refine this ability through experiences with the world. However, few models address the unsupervised development of event segmentation in artificial agents. This paper presents presents a computational model of how an intelligent agent can independently learn to recognize meaningful events in continuous observations. The agent's segmentation mechanism starts from a simple state and is refined. The agent's interactions with the environment are unsupervised and driven by its expectation failures. Reinforcement learning drives event boundary identification by reasoning over a gated-recurrent neural network's expectation failures. The learning task is to reduce prediction error by identifying when one event transitions into another. Our experimental results support that reinforcement learning can enable detecting event boundaries in continuous observations based on a gated-recurrent neural network's prediction error and that this is possible with a simple set of features.