View article

We are developing a generative art application for casual creation on handheld devices, where user enjoyment is prioritised over the quality and/or utility of the abstract images they make. In an attempt to increase enjoyment, we have enabled the app to generate image titles, and we describe here the approach we take, along with details of some experiments we undertook to optimise it for efficiency and variety. The approach employs words relating to a machine vision classification and colour breakdown of the images, wrapped in the language of International Art English. When there is no suitable input from the machine vision analysis of an image, the approach generates text which is not necessarily related to the image. Such texts have been described linguistically as pseudoprofound bullshit statements, in work on the psychology of human judgement that we survey. We evaluate the title generation approach with a curation analysis, and end with a discussion of some potential benefits of employing bullshit in a Computational Creativity context.