Memo Akten — Learning to See

2017

Artists: Memo Akten · Year: 2017 · Medium: Real-time AI piece — camera input re-interpreted through trained neural network · Substrate: Pix2Pix neural network trained on specific image domains (waves, landscapes, flowers); real-time webcam input; continuous translation from input to output · Place: London, UK / Los Angeles, USA · Institution: Independent; later UC San Diego Memo Akten's Learning to See, begun in 2017 and refined through several versions, is a real-time installation in which a camera-facing display shows whatever is in front of the camera — hands, faces, everyday objects — re-interpreted through a neural network trained to see one specific thing. In one version of the piece, the network has been trained to see only waves; whatever the camera captures, the display shows as an interpretation of what-if-this-were-a-wave. In another version, the network sees only flowers; another, landscapes. Each trained network is narrow-minded, and its narrow-mindedness is the subject. The piece is one of the clearest demonstrations in AI art that the trained network's bias is the network's aesthetic. A wave-seeing network shown a face does not produce a good translation of the face; it produces a face-that-the-wave-network-had-to-do-something-with. The viewer's own face, in real time, is being reinterpreted by a machine that has never seen a face and does not know what one is. The uncanniness is specific. Akten has argued extensively in writing and in lectures that this narrow-minded quality is what AI systems actually do in the world — systems trained on specific data cannot see outside their training, regardless of how universal they claim to be. The political implications of the argument are substantial. He is based currently in Los Angeles, teaching at UC San Diego, and continues to produce real-time AI work. Lore Akten trained as an engineer and moved to artistic practice through openFrameworks and the creative-coding community. His work has been widely exhibited at Ars Electronica, Barbican, and major new-media festivals. His 2020 doctoral thesis from Goldsmiths was on real-time deep learning in creative practice — one of the first PhDs written on the subject by a practicing artist. Sources • Akten, Memo — memo.tv archive • Akten, Memo — Deep Visual Instruments (PhD thesis, Goldsmiths, 2020) • Leonardo — Akten essays