Noise is a Feature: Biology as Self-Refining Collective Intelligence

Biology and evolution do not operate on passive matter, but on an agential substrate: living systems are multi-scale competency architectures, nested layers of organization whose components exhibit problem-solving capacities in their own domains. This adds a crucial, often overlooked layer between genotype and phenotype: a collective, problem-solving substrate that coordinates organismal outcomes and compensates under novel developmental contexts. Evolution therefore does not optimize static traits, but acts on – and builds – systems with intrinsic problem-solving abilities. In this view, the genome is not a blueprint but a generative bowtie: a compression–expansion architecture interfacing with this agential substrate.

Collective intelligence enables decentralized coordination in living systems at every scale. Cells, tissues, and organisms achieve coherent behavior through local interactions despite noisy conditions and unreliable hardware. To study this, we use neural cellular automata (NCAs) as transparent models of morphogenesis to investigate how tissues self-assemble, maintain anatomical targets, and regenerate.

Using neuroevolution, we design in silico microswimmers with decentralized, embodied control. These generalize across body sizes, stay robust under morphological perturbations, and enable cargo transport without retraining, promising for microrobotics and drug delivery.

The same framework offers a new lens on aging: as a loss of anatomical goal-directedness. In evolved NCA-based systems, we observe long-term anatomical degradation, and show that rejuvenation can be induced in silico by reactivating dormant developmental pathways.

Across these systems a common motif recurs: iterative local error correction that progressively stabilizes global structure. We show that development and evolution can be understood as iterative denoising processes akin to diffusion models. More broadly, biological organization relies on generative, bowtie-like architectures that compress information into latent representations, flexibly decoded into adaptive behavior.

His work explores how treating biology as multi-scale collective intelligence may inform new approaches to regeneration, longevity, and synthetic, bio-hybrid, or living systems.