Joshua Borsman
A machine is taught to read a human hand. What you are watching is the learning itself — a network in the hours of its becoming, settling toward fluency, and beginning again.
Sound on. Press F for fullscreen.
A network learning to read, in real time.
A machine is taught to read a human hand. Settling shows the lesson itself — not its result but the hours of becoming. A network is seeded knowing nothing. It guesses, and is wrong, and is corrected, and by degrees the corrections grow rare. Its first layer, at the outset only noise, resolves into the strokes and edges of a hand; the figures it once mistook, it comes to know. When it has learned it does not stop — for a long while it reads, figure after figure, everything it has come to know, and the searching gives way to something settled and sure. Then it dissolves, and another begins. No two descents are alike. The work is built to run for hours, and does not repeat.
Read the field from left to right. A handwritten figure enters at the left. In the middle is the machine's forming mind: each small tile is an actual receptive field — the weights leading into a single unit of the network — drawn exactly as it stands, resolving over time from noise into the marks of a hand. At the right is the reading. While the network is uncertain, several figures hang together in a wavering cloud; as it learns, the cloud closes onto a single answer, and holds. While it learns it is often wrong, and now and then sure and wrong at once; the warmth of a reading tells being certain from being right.
Sound is continuous and made in the moment. Each figure newly read strikes a single note, and each kind of figure keeps its own voice, so the work is heard to settle: many notes while it is uncertain, then fewer and longer as it comes to know. Nothing is sequenced and nothing is recorded. What is heard is a function of what the network is learning, right now.
Joshua Borsman is an artist working with sound, light, and real-time systems. His generative works give form to processes that ordinarily pass unseen — the orbits overhead, the weather of a coast, the machinery beneath the internet — and render them as continuous, non-repeating fields of image and sound. Settling turns that attention inward, to the moment a machine learns. It is the first work in Hidden Layers, a series on the unseen interior of machine learning — the place where a system comes to know.
© 2026 Joshua Borsman. All rights reserved.