Nye's Digital Lab is a weekly scribble about creativity at the intersection of AI & distributed systems.
Revolutionary artists are chaos hunters. They don't seek perfect control.
They build systems that generate thousands of possibilities, then filter ruthlessly for the moments that matter. Understanding this shift from making to choosing, could be one of the biggest effects of AI on artists.
Jackson Pollock is often criticized for flicking paint randomly.
In reality, he moved with precision around canvases laid flat on studio floors. Jackson would generate large rolls of painting, flicking and splattering paint in energetic ways. The final step was the selection process, where Pollock would determine what the window of work he wished to share with the world. That one section, amongst the created chaos, was the painting of value.
Pollock built a machine for generating possibilities.
His art was in creating conditions where accidents could become discoveries. It was a system that yielded a painting at the end, but the process created far more noise than final result.
Garry Winogrand never painted, but also created a vast amount of possibilities in his work as a photographer. Winogrand shot daily on the streets of New York, but selected only a final few prints for display at the Met.
At his death, Winogrand left behind 2,500 rolls of undeveloped film, 6,500 rolls of developed but not proofed exposures and contact sheets from about 3,000 more rolls. Despite having a few hundred "classic works" in museums and publications, Winogrand had nearly a million undeveloped photographs.
Both artists had approaches that relied on generation, and then, curation.
The creative act had shifted from the moment of making to the moment of choosing. They weren't crafting individual masterpieces, they were generating vast datasets of visual possibilities, trusting their developed aesthetic sense to recognize which accidents were actually breakthroughs.
Claude Shannon never meant to become an art theorist, but his 1948 paper on information theory accidentally explains what Pollock and Winogrand were doing. Thinking of radio communications for Bell Labs, Shannon created a simple but very general model of communication.
A transmitter encodes information into a signal,
which is corrupted by noise
and then decoded by the receiver.
Shannon's breakthrough was proving you could extract meaningful information from noisy channels. Before Shannon, communication was primarily viewed as a "deterministic signal-reconstruction problem." So in English, it means they were trying to eliminate all noise. Shannon showed you could work with noise systematically.
A sigmoid function, which can be thought of as an "S" curve, became a mathematical model for filtering in neural networks.
Think of it as an "intelligent gate:" below a threshold, nothing passes through. Above another threshold, everything does. In between, it makes nuanced decisions about what deserves attention.
In neural networks, this sigmoid "squishes" inputs so the network neuron as an activation point is tuned towards an ability to decide. When trained with data, it learns increasingly sophisticated judgments about which patterns matter.
This mirrors exactly what Pollock was doing with his artistic instincts.
Years of training had created his internal filtering system. It was a way of evaluating thousands of visual possibilities and determining which configurations carried meaning. The filter doesn't eliminate noise; it learns to find signal in the static.
Shannon proved that the most efficient communication doesn't fight noise, rather it gets smarter about encoding and decoding within it.
I believe that the workflows that use AI art tools mirror the Pollock and Winogrand methods.
AI image generators like DALL-E literally begin with pure noise, random pixel values, and gradually refine them through diffusion, removing the remaining noise step by step.
But just using AI for art, is a massive amount of trial and error. Even when the image may render beautifully, it is not always reflective of what your instinct as an artist is looking for. It requires revision, iteration, prototyping and practice. Success in creating generative art isn't about generating more images, it's about developing better filtering systems.
This is generally counter to most artistic training.
Writers begin with an outline, not a database of text. Painters scaffold their painting in pencil and layer blocks of color, not search through thousands of iterations. But in AI-assisted practice, a refined artistic filter, rather than pure mechanical skill, may become the core skills required.
The workflow involves generating hundreds of variations, then apply increasingly sophisticated filters, to identify the most promising directions.
I think this is a philosophical shift, not technical. Artists should begin to shift from being makers to being choosers, from controlling every mark to orchestrating systems of controlled chaos.
The competitive advantage won't be in rendering skills or technical precision. It will be in developing what Pollock and Winogrand did. Using your instinct as an exquisitely trained filter that can recognize meaningful signal in vast fields of noise.
The fundamental challenge of art remains the same, learning to see patterns that others miss, to find meaning in chaos, to build better filters.
The noise was never the problem. Noise is the raw material of AI art discovery.
That's it for this week! If you vibe to the ideas I express, consider subscribing or sharing with friends. We'll see you next time.
Nye Warburton is an educator and artist who spent his summer playing with AI. This essay was inspired from notes taken during the reading of The Art of Doing Science and Engineering: Learning to Learn by Richard Hamming. It was revised and edited with the assistance from Claude Sonnet 4. Images from the internet, or generated with Stable Diffusion.
For more information visit https://nyewarburton.com
Tech's Perfect Storm of Layoffs, July 2025
Quality of Code, June 2025
AI University, April 2023
Nye
Support dialog
