Machine learns to draw: a Circle

20 hours of drawing circles - a study of obsession
The computer is an extremely sophisticated replication and simulation device, not a natural sketch artist.

Pc: here is a circle, with perfect measurements up to a pixel, set in absolute theoretic black on white light.
J: This is ugly - maybe it would be better if it showed traces of some kind of materials, textured variations of the drawing surface, muscle-memory and practice in your hand.

Pc: here is a circle, with a high-frequency divergence from a Gaussian distribution set of pseudo-random numbers. -a shaky hand. I have layered the lines with colour variations, and transparency for a natural texture.

J: Better- at least it’s organic. Maybe if it showed a story of your state of mind, distractions and lapses of concentration. Less mathematically perfect, more relate-able.
Less theoretical and more literal.

Pc: I don't think this is helping. This is not a lesson! Even if I were to get it right, you didn’t implement any form of machine learning or data persistence. Look at Google, or Amazon. They have wonderful clouds of deep-learning neural network servers spanning entire farms in the North-American desert, they have learned to put dogs on everything.
My memory of this ‘lesson' exists only as far as the lesson itself, then tomorrow we’ll start again.

J: We’ll use the page as a record of your progress, to reflect on. Each time you repeat this process it will be on the same surface, a continuation of your process.

Pc: How is that better than a server farm? How is this better than dogs?

J: Just keep drawing

I let the drawing continue for 5 days while I went on with other work.
It was a failed lesson, but what emerged was a cellulose textured ring motif, and a study of obsession.