Tom White’s Perception Engine presents the outcomes of a five-year experiment with artificial intelligence, undertaken to explore the parallels and contrasts between human perception and its simulation by computers.
White started by training early neural networks on thousands of images depicting various household objects or common animals. Over a lengthy series of iterations, the neural network refined its understanding of its dataset, producing outputs that other neural networks could recognize for what they intended to represent. Most, however, still seem abstract and impressionistic to human viewers. Electric Fan (2018) is one such output that never attained human recognizability. White collected these imperfect iterations into _The Treachery of ImageNet _subseries, whose title invokes Rene Magritte’s famous 1929 painting La Trahison des images. More successful versions from 2019 like _Rabbit _and _Banana (trial proofs) _more clearly evoke their intended subjects. Printer (2023), meanwhile, animates the neural network’s creation process, showing every layer of abstraction until full representation is achieved.
Related Works
Homage to the pixel: FeelingTom White2022NFT/Digital
Circuit (Atlas of Perception)Tom White2025NFT/Digital
Roots (Atlas of Perception)Tom White2025NFT/Digital
Rabbit Trial Proof 5Tom White2022Print
Homage to the pixel: CaveTom White2022NFT/Digital
Eye of the EagleTom White2022NFT/Digital
Banana Trial ProofTom White2022Print
PrinterTom White2023NFT/Digital
Email Newsletter
Sign up to receive the occasional news and stories from the world of computer art.