Sakana AI Builds a Brain-Like Neural Network That Learns Without Backpropagation

Every neuron in your brain is either an excitator or an inhibitor , it never switches sides. This is Dale's principle, a fundamental rule of biological neural circuits. Standard artificial neural networks ignore it entirely, freely mixing positive and negative weights on the same unit. A new paper accepted at ALIFE 2026, authored by Yutaro Yamada, Luca Grillotti, Rujikorn Charakorn, Sebastian Risi, David Ha, and Robert Tjarko Lange (all affiliated with Sakana AI), asks a pointed question: can a network that strictly obeys Dale's principle still learn well? Their answer, detailed in Diffusing Blame: Task-Dependent Credit Assignment in Biologically Plausible Dual-Stream Networks, is yes , with some important caveats.

Two problems backprop ignores

To understand why this matters, you need to know about two biological violations that standard deep learning commits simultaneously.

• Dale's principle: Biological neural circuits obey Dale's principle , each neuron releases the same neurotransmitter at all of its synapses, making it uniformly excitatory or inhibitory. Artificial neural networks freely assign positive and negative weights to any connection, a simplification that enables efficient credit assignment via backpropagation but bears little resemblance to biological learning.
• The weight transport problem: Training deep networks with backpropagation requires the backward pass to use exact transposes of forward weight matrices , the weight transport problem , which lacks biological support. The brain has no known mechanism for copying weight matrices backward through layers.

Several prior methods (Feedback Alignment, Direct Feedback Alignment) solve the weight transport problem by replacing backward weights with fixed random matrices. But none of these methods enforce Dale's principle , all allow arbitrary sign weights. The authors set out to fix both problems at once.

The dual-stream architecture

The core idea is to split every layer into two parallel streams: one excitatory (p) and one inhibitory (n). Error Diffusion (ED) was originally proposed in a dual-stream excitatory/inhibitory architecture, where learning is driven by routing global error signals to all layers without transporting transposed forward weights or relying on random feedback matrices.