Summary: Modern AI models generally imitate the brain’s cortex—the outer, high-level layer—while largely overlooking the older, deeper subcortical structures. A research team in the Netherlands proposes a new computational architecture that integrates those subcortical pathways with cortical hierarchies, producing models that are more flexible, efficient, and biologically realistic.
The study finds that pairing a fast, shallow subcortical route with the traditional deep, hierarchical cortical route enables quicker responses for simple tasks and preserves deep processing for complex problems. This parallel strategy mirrors known brain anatomy and offers practical benefits for artificial intelligence systems.
Key Findings
- The Shallow Brain Hypothesis: Extending their 2023 proposal, the authors show the brain relies not only on stepwise cortical hierarchies but also on parallel interactions between cortex and subcortex.
- Functional Complementarity: In decision-making simulations, the shallow subcortical pathway handles rapid stimulus–response reactions while the cortical hierarchy supports more demanding, context-dependent decisions.
- Efficiency Gains: The dual-route architecture processes information more flexibly and avoids wasting computation on simple tasks that do not require deep hierarchical analysis.
- Biological Realism: The model restores feedback loops and short-range pathways typical of the nervous system, bringing artificial networks closer to actual brain organization.
Source: EBRAINS
Overview
Most contemporary deep learning systems process information through many sequential layers, inspired by cortico-cortical connectivity in the cortex. However, biological brains also depend on subcortical regions—structures that support movement, emotion, quick learned responses and basic learning—that interact continuously with cortical areas. These subcortical channels are largely absent from standard artificial neural network designs.
Published in Current Research in Neurobiology and supported by the Human Brain Project, the new study proposes a computational architecture that explicitly integrates cortical hierarchies with faster, shallower subcortical routes. The resulting design is parallel rather than purely hierarchical, reflecting anatomical and functional features observed in the brain.
The researchers implemented two variants of the cortical component—a convolutional feedforward network and a hierarchical predictive coding model—while adding a shallow subcortical pathway in both cases. They tested these models on perceptual, context-dependent decision-making tasks and found consistent patterns: for easy trials, the subcortical route reliably guides rapid decisions; for harder trials, the cortical hierarchy engages and refines choices.
These results support the idea that parallel cortico-subcortical processing is a fundamental computational principle in biological brains and that incorporating it into AI can yield systems that better balance speed and complexity.
Key Questions Answered
A: Deep networks excel at complex recognition tasks but are computationally expensive and slow for simple reflexive actions. A shallow route provides a biologically inspired shortcut, enabling instant responses to straightforward stimuli without routing every signal through many layers.
A: Yes. Most AI models focus on cortical-style hierarchies and largely ignore subcortical regions that govern emotions, survival behaviors, and rapid learned responses. Omitting these elements can make systems less adaptable and less efficient.
A: By combining a fast, instinctive pathway with a slower, deliberative one, the architecture produces behavior that mirrors human balance between quick “gut” reactions and deeper, reflective thinking—prioritizing speed for simple tasks and depth for complex ones.
Editorial Notes
- This article was edited by a Neuroscience News editor.
- The journal article was reviewed in full by staff.
- Additional context and explanations were added by the editorial team.
About this neuroscience and AI research news
Author: Helen Mendes Lima
Source: EBRAINS
Contact: Helen Mendes Lima – EBRAINS
Image: The image is credited to Neuroscience News
Original Research: Open access. “A computational architecture incorporating shallow brain networks: integrating parallel cortical and subcortical processing” by Kwangjun Lee, Lorenzo Gabriele Baracco, Cyriel M.A. Pennartz, Mototaka Suzuki, and Jorge F. Mejias. DOI: 10.1016/j.crneur.2026.100155
Abstract
A computational architecture incorporating shallow brain networks: integrating parallel cortical and subcortical processing
Artificial neural networks are typically built as deep hierarchical systems inspired by cortico-cortical connections. These models under-represent non-hierarchical brain features—specifically subcortical pathways and cortico-subcortical interactions that operate regardless of hierarchical position.
Motivated by neuroanatomy, the authors present a model that combines cortical hierarchical processing with anatomically informed subcortical pathways. They demonstrate the model’s versatility by instantiating the cortical hierarchy as either a convolutional feedforward network or a predictive coding network.
Both variants reproduce behavioral patterns observed in humans and monkeys performing a perceptual, context-dependent decision-making task. The simulations show subcortical structures dominate decisions on easy trials, while the hierarchical cortex is required for harder ones. These findings suggest parallel cortico-subcortical processing is a core computational principle that should be considered in both neuroscience and AI design.