Deep Learning as a Creative Collaborator, Or How I Learned to Stop Worrying and Love AI
Andy Lomas

Summary

Andy Lomas presents collaborative research with John McCormack's SensiLab at Monash University on using deep learning as a creative partner rather than a front-end image generator. He describes his long-standing practice of building simulation systems inspired by the ingredients of life, such as cellular division and growth, which produce remarkably organic forms that he then turns into prints, projections, 3D prints and virtual reality environments. The talk centres on Species Explorer, bespoke software he uses to rate, score and categorise thousands of outputs in order to steer the search towards interesting parameter regions. By training ResNet-50 and fully connected neural networks on his own aesthetic judgements, Lomas automates categorisation, predicts ratings from genotype parameters, visualises high-dimensional parameter spaces and automatically searches for transition points poised between order and chaos. He stresses that the machine is not replacing his taste but augmenting his ability to navigate complexity, keeping the human artist inside the loop while discovering emergent structures that would otherwise remain inaccessible.

Generated by AI using kimi-k2.7-code (2^nd September 2026).