RIVAL is a deep learning, neural network based AI that I am training, from scratch, on my previous artwork and image collecting practice(s). The images are associated / "weighted" with a variety of keywords, (based on imagery, color, theme, groupings, etc.). These become the categories / "things" that RIVAL learns to identify and associate. In later phases these associations become the core of projects I develop into lectures, videos, books, printed compositions, etc.
For Phase 3 of my work with RIVAL; images totally unrecognizable to human eyes that deep neural networks identify with high confidence as familiar objects* were used to group images from RIVAL's previous knowledge/data sets (my previous artwork and image collecting) that most closely resemble the misrecognized imagery, taking into account both the imagery (pattern/arrangement of pixels) and keywords.
Through this process(ing) the abstract (classic AI misrecognized imagery) is pulled into something simultaneoulsy more representational and more poetic. The images in these groupings sometimes resemble the original identification, and just as often further complexify the associations that the AI (RIVAL) is making, beginning to point toward the possibility of the appearance of "meaning(fulness)".
*( Source: Anh Nguyen, Jason Yosinski, and Jeff Clune, "Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Images," cv-foundation.org, 2015 → [PDF]. )
