Dream Machine
The view from a window is a promise. A promise of the world outside, continuous and ever-changing. But a photograph of that view... that is something else entirely. It is an interruption. A moment arrested, severed from the flow of time. We have come to trust these interruptions, to call them memories.
I have spent some time with this idea. In 2018 I captured thousands of moments from a single vantage point out my window. Not to possess them, but to provide them as a kind of visual nourishment for a machine. I then asked the machine to dream.
And the machine, in turn, began to offer back images that were not memories, for nothing was remembered. They were echoes. Plausible ghosts of moments that never occurred.
This prompts us to question the original photograph itself. We have always assumed its authority. We say: this happened. But a photograph is not the event. It is a slice, a selection. The camera always interprets.
Look at these grids, each a representation of a Deep Convolutional Generative Adversarial Network (DCGAN) training epoch. At first, the machine sees only whispers of color and form. A suggestion of pink, a smudge of blue. Then, slowly, a horizon begins to assert itself. A phantom city takes shape. The machine is not copying; it is learning a way of seeing. It is developing its own grammar of light and shadow, based on the vocabulary it was given.
What does it mean when a machine can generate an image that feels authentic, that might evoke a sense of nostalgia for a past that never took place? Perhaps it reveals that the camera’s claim to authenticity was always a fragile one. The act of looking, whether by human or by machine, is always an act of construction.
And we, the spectators, where do we stand now?