Monday 30 October 2017

beyond the uncanny valley of the dolls or playable-character

Neural networks have trained themselves, dreaming themselves from an age of nightmare to a liminal, lucid slumber, to produce absolutely convincing, authentic-looking images of people that only exist in the machine’s circuitry.
Second generation applications of learning software are already outpacing human engineering by crafting their own versions and this visage-generator works as good as it does by bouncing ideas off of a parallel, adversarial programme constantly critiquing its choices and adjusting accordingly. Visit Sploid at the link up top for a video demonstration which includes entire computer-created environments—although not uncanny like a mannequin any longer, the results and transformations are still eerie and different than the transitional figures of the face-morphing sequence of Michael Jackson’s 1991 single “Black or White,” who had a warmth in their slipping away.  There’s something a touch dejected, I think and it’s different that the impermanence of a dream since one dreams of those one knows or remembers, in watching these characters glide past that seem surpassingly real yet are a chance configuration that may never be grounded as existing in any sense.