RNN Drawer

In 2019, generative artificial intelligence had not yet become the widely accessible infrastructure for image production that it is today.

Systems at the time were smaller in scale, narrower in capability, and usually designed to perform a limited single task. RNN Drawer was only a small neural network. Its limitations remained clearly visible. It had not yet acquired the ability to conceal the instability of its own operation.

What these images present is not mature technique, but the moment immediately before technique became possible.

Failure as a visual state. The model’s poor convergence did not generate a complete visual language. Its output remained suspended between structure and noise, intention and accident, resemblance and abstraction.

The work did not treat these results as technical waste to be discarded, but preserved them as evidence left by an unresolved process.

The early studies of a painter not yet known, or the drawings of a child not yet disciplined by established rules.

These drawings belong to a brief historical interval:

At that moment, human technique still appeared to possess a clear technical advantage. Drawing ability, visual judgement and artistic control remained difficult to reproduce through computation.

Today, technique, ability, and even the appearance of talent are gradually separating from the trained hands that once gave them their distinctiveness.

RNN Drawer preserves an earlier threshold: a moment when the machine was approaching the image but had not yet occupied it.

In 2023, the surviving generated results of RNN Drawer were reorganised as the virtual exhibition The Vintage Drawings.

“Vintage” points to relics left behind before generative systems. They are traces of the moment when the machine was still learning what an image might be.