Local processors: Active dendrites and artificial intelligence
- 1Florey Institute of Neuroscience and Mental Health, Florey Department of Neuroscience and Mental Health, University of Melbourne, Parkville, VIC, Australia.
Abstract
Neuroscience and artificial intelligence have been closely connected for over a century, with each field influencing the other. Modern artificial intelligence systems rely almost exclusively on the point-neuron model, where inputs are linearly summed and passed through a global activation function. Although highly successful, this architecture differs substantially from neural computation that occurs in the brain, and current leading models require massive infrastructure consuming megawatts of power, which contrasts to the brain's remarkable efficiency. One underexplored source of biologically inspired computational principles lies in the dendrites of pyramidal neurons. This review examines the electrical properties of active dendrites and how they could inform the next generation of artificial intelligence architectures. We describe how dendrites can function as semi-independent computational subunits, generating local nonlinear events that implement coincidence detection, signal amplification, and logical operations within a single neuron. We discuss how the structural segregation of different dendritic compartments provides a biologically grounded alternative to global error backpropagation through local credit assignment. We further examine how activity-dependent spine growth and pruning support continual learning without catastrophic forgetting, and how the spatiotemporal selectivity of dendritic integration gives rise to sparse, energy-efficient representations. Together, dendritic computation represents a potential blueprint for artificial intelligence architectures that are adaptive, context-aware, and energetically frugal.
Related
- Online synaptic credit assignment in active dendrites
- Local translation couples synaptic activity to mitochondrial adaptation in dendrites
- Opposing Roles for the Spectraplakin Short Stop in Stable and Dynamic Dendrites Reveal Divergent DLK Signaling and a Role in Dendrite Regeneration
- Voltage dynamics of cortical dendrites in vivo
- Toward Enactive Artificial Intelligence
BibTeX
@article{hu2026local,
title = {{Local processors: Active dendrites and artificial intelligence}},
author = {Yi Hu and Nicholas S Glatz and Lucy M Palmer},
journal = {Current research in neurobiology},
year = {2026},
month = oct,
doi = {10.1016/j.crneur.2026.100169},
url = {https://europepmc.org/article/PMC/PMC13717092},
note = {CC BY},
}
RIS
TY - JOUR
TI - Local processors: Active dendrites and artificial intelligence
AU - Yi Hu
AU - Nicholas S Glatz
AU - Lucy M Palmer
PY - 2026
DA - 2026/10/15
JO - Current research in neurobiology
DO - 10.1016/j.crneur.2026.100169
UR - https://europepmc.org/article/PMC/PMC13717092
AB - Neuroscience and artificial intelligence have been closely connected for over a century, with each field influencing the other. Modern artificial intelligence systems rely almost exclusively on the point-neuron model, where inputs are linearly summed and passed through a global activation function. Although highly successful, this architecture differs substantially from neural computation that occurs in the brain, and current leading models require massive infrastructure consuming megawatts of power, which contrasts to the brain's remarkable efficiency. One underexplored source of biologically inspired computational principles lies in the dendrites of pyramidal neurons. This review examines the electrical properties of active dendrites and how they could inform the next generation of artificial intelligence architectures. We describe how dendrites can function as semi-independent computational subunits, generating local nonlinear events that implement coincidence detection, signal amplification, and logical operations within a single neuron. We discuss how the structural segregation of different dendritic compartments provides a biologically grounded alternative to global error backpropagation through local credit assignment. We further examine how activity-dependent spine growth and pruning support continual learning without catastrophic forgetting, and how the spatiotemporal selectivity of dendritic integration gives rise to sparse, energy-efficient representations. Together, dendritic computation represents a potential blueprint for artificial intelligence architectures that are adaptive, context-aware, and energetically frugal.
SN - 2665-945X
ER -