I use modified versions of BigSleep and other text-to-image GANs, submitting lines of poetry to see how neural networks trained on classification data respond to poetic imagery. I have a degree in poetry and by day work on NLU and ML for mobile messaging. I have a degree in poetry and have been a career technologist for humanitarian work for 15 years.
About the current collection
I'm exclusively using the works of Emily Dickinson for this first collection. Emily Dickinson was, in many ways, a poet before her time, and the form and imagery lends itself well to this space (and it's... also in the public domain). Other than an appreciate for poetry itself, I'm hoping to create equal spaces about questioning if computers can truly understand poetry, or why not; and about the nature of the data we feed these networks. With so many examples of the training data producing racist or harmful results, what more could we achieve with systems designed for perception, not simply accuracy.