for the sleepers in that quiet earth.: An Artbook Generated by a Deep Learning Agent

This paper describes an artbook created through the adaptive behavior of a deep learning neural network computational agent as it “reads” a novel. Through this process, the network as agent builds a model of the syntactic and stylistic principles behind the original text and is able to use this model to generate new, unforeseen text. There exists 31 printed editions of the artbook thus generated, each unique and consisting of a text that reveals the learning process of the agent itself as it goes through the adaptive process, starting from a state of randomness and gradually refining itself as it reads the novel. The artbook is examined by looking at generated excerpts and discussing how they relate to the state of the deep learning system at different steps through the adaptive-generative process. Practical and theoretical implications are discussed in the context of generative literature, machine learning, and behavior aesthetics.