Main Article Content
It is well established that AI has a bias problem; however, black-boxed machine learning systems render it difficult to even understand and visualize the nature and extent of the problem, let alone find solutions. This paper discusses an artistic research approach toward highlighting AI bias and explores the aesthetic potential of machine learning through a case study of an AI artwork called #RiseandGrind. The artist trained a recurrent neural network on a dataset extracted from Twitter hashtags (#Riseandgrind and #Hustle), which were selected to represent a specific filter bubble (embodied neoliberal precarity) in order to produce a biased AI that generates tweets for a Twitter bot. This paper unpacks how this artwork makes visible the processes of machine learning in a playful and poetic way. The work reveals how the original filter bias is consolidated, amplified, shaped, and ultimately codified through this machine learning process. The AI is found to reproduce a cohesive world view that, while reflecting the original data bias, further amplifies that bias through a process of flattening.