An algorithm that allows an AI model to re-organise, re-use, and re-write features that diverge from the original dataset.

Although current approaches to exploring AI models remain partially in a blackbox (Bau et al., 2020), works from artists (Schultz, 2020, Som, 2020) have been attempting to bridge the gap between autonomous network decisions and computational creativity (Berns and Colton, 2020). Their works demonstrated unexplored potential of reusing features that diverge from the approximate distribution, further outlining the need for more predictable possibilities (Broad et al., 2020) and human intervention (Hertzmann, 2020) during inference. Motivated by this, we propose an alternative approach to reuse encoded knowledge in trained models and intervene in the generative process in pursuing of human-AI co-creation.