Sophisticated generative models have been some of the most obviously impressive achievements of AI in the past decade. Indeed, AI has approached or achieved human-level performance in a variety of domains, with a particular focus placed on artistic disciplines. Because writing a value function for this kind of model is effectively impossible, these models are notoriously difficult to evaluate. My aim was to create such a model, and in doing so elucidate the characteristics of successful generative AI and explore how generative models can be adapted successfully to new domains. And more broadly, I hoped to create a model capable of producing good sculpture, although I have no intention of evaluating what it means for a sculpture to be good.
Downey, Nick, "Generating 3d Sculptures Using a Recurrent Neural Network with Long Short-Term Memory" (2019). IPHS 300: Artificial Intelligence for the Humanities: Text, Image, and Sound. Paper 21.