NeuroAI and the brain: Integrating brain-inspired computational models and large-scale AI architectures with computational theories of neural function

The rapid ascent of large-scale artificial intelligence has provided neuroscience with a new set of powerful tools for modeling complex cognitive functions. Historically, computational neuroscience has balanced between biophysically detailed models of single neurons and abstract, small-scale circuit motifs. However, the ability of modern AI to perform human-like tasks in sensory processing, motor control, and linguistics offers a "top-down" framework to understand how the brain might organize high-dimensional data. This convergence, known as NeuroAI, suggests that the representational spaces of deep networks can serve as predictive scaffolds for neural activity. Despite this potential, a critical gap remains: most AI models lack the biological constraints, metabolic efficiency, and architectural diversity of the brain, while many neural models struggle to scale to the complexity of real-world behavior.The goal of this Research Topic is to bridge the divide between high-performing, large-scale AI models and mechanistic computational theories of neural function. We seek to move beyond simple correlation-based comparisons between AI and brain activity toward contributions that explain cognition through principled, brain-inspired computational models whose internal representations reflect the causal and relational structure of the world, grounded in the principles by which biological intelligence evolved. This means addressing the "black box" problem of modern AI by integrating established neurobiological principles, such as predictive coding, canonical microcircuitry, and homeostatic plasticity, into large-scale artificial architectures. We also aim to explore how biological constraints (e.g., spiking dynamics, energy efficiency, and modularity) can be implemented within AI to increase its biological plausibility without sacrificing performance. By aligning the objective functions of artificial systems with the evolutionary pressures of biological ones, we can develop a more robust, cross-disciplinary understanding of the fundamental principles of intelligence.This Research Topic welcomes contributions that leverage large-scale AI to gain insights into brain function, or conversely, use neural principles to innovate artificial architectures. We encourage submissions of original research, reviews, and perspective pieces that emphasize mechanistic links over purely descriptive comparisons. Specific themes of interest include, ordered from more general to more specific:- Scaling laws in biological brains versus artificial neural networks: Comparative analysis of how computational capability and representational structure scale with network size, connectivity, and learning dynamics across biological and artificial systems.- Representational similarity analysis between large-scale AI models (e.g., transformers, diffusion models) and high-density neural recordings, with emphasis on moving beyond descriptive comparisons toward causal and mechanistic accounts.- The implementation of biologically plausible learning rules (e.g., local plasticity, three-factor rules) in large-scale networks, including self-supervised mechanisms that reflect how the brain acquires structured representations from experience.- The role of feedback, recurrence, and lateral inhibition in enhancing the robustness and generalization of AI, informed by cortical architecture and top-down predictive processing.- Energy-efficient AI and neuromorphic computing inspired by the brain's metabolic constraints, including spiking neural networks and event-driven architectures.- Structural-functional alignment: Modeling specific brain systems (e.g., basal ganglia, cerebellum, hippocampus) using specialized AI motifs that respect known biological constraints and connectivity patterns.- Brain-inspired world models and episodic memory: How hippocampal and entorhinal mechanisms for relational memory, predictive inference, and structured world representations can be formalized and integrated into large-scale AI architectures.- The integration of "system 1" (intuitive) and "system 2" (rational) processing within neuro-symbolic AI frameworks, including hybrid architectures that bridge fast pattern recognition with deliberate, model-based reasoning.- Using large-scale models to simulate neurophysiological changes in aging, psychiatric disorders, or neurological conditions, connecting computational mechanisms to clinical observations. Topic editors: M. Andrea Pisauro, Siamak Sorooshyari, Ivan Georgiev Raikov, Mario Negrello, Masanori Shimono Source: https://www.frontiersin.org/research-topics/81076/neuroai-and-the-brain-integrating-brain-inspired-computational-models-and-large-scale-ai-architectures-with-computational-theories-of-neural-function