Noise & Order
A lot of my most favorite AI products use diffusion models. Suno for music gen, Runway, Sora (Text-to-Video), Midjourney & DALL·E for image generation, 11labs with their voice cloning...
They all start from noise - literal static - and through learned steps, carve out structure. It’s powerful and poetic.
But you know what’s really cool?
In 1952, Alan Turing published a paper called "The Chemical Basis of Morphogenesis." It proposes a mathematical theory explaining how patterns in biological organisms - spots, stripes, body structures - can emerge spontaneously through chemical processes. He calls it a reaction-diffusion model.
70 years later, I’m sitting here thinking about the conceptual parallels between Turing's reaction-diffusion systems and modern AI diffusion models.
Turing’s system produces order from randomness through deterministic rules plus randomness-driven instability.
Diffusion models do the same: they begin with noise and, via learned transformations, recover structured data - images, audio, etc. Both systems rely on the transformation of disorder into structure.
While the math is different (Turing uses PDEs and linear algebra; DDPMs use stochastic differential equations and deep nets), the philosophical resonance is clear:
Both Turing’s models and AI diffusion models start from randomness.
Both follow step-by-step rules.
Both generate complexity - from nothing…
Time + noise + structure = beauty.
Kind of magic, isn’t it?
