arxiv: 2604.18637 · v2 · submitted 2026-04-19 · 🧬 q-bio.NC · cs.AI· cs.CY

Recognition: unknown

NeuroAI and Beyond: Bridging Between Advances in Neuroscience and ArtificialIntelligence

Anthony Zador , Jean-Marc Fellous , Terrence Sejnowski , Gina Adam , James B Aimone , Akwasi Akwaboah , Yiannis Aloimonos , Carmen Amo Alonso

show 23 more authors

Chiara Bartolozzi Michael J. Bennington Michael Berry Bing W. Brunton Gert Cauwenberghs Hillel J. Chiel Tobi Delbruck John Doyle Jason Eshraghian Ralph Etienne-Cummings Cornelia Fermuller Matthew Jacobsen Ali A. Minai Barbara Oakley Alexander G. Ororbia II Joe Paton Blake Richards Yulia Sandamirskaya Abhronil Sengupta Shihab Shamma Michael P. Stryker Seong Jong Yoo Steven W. Zucker

Authors on Pith no claims yet

Pith reviewed 2026-05-10 06:04 UTC · model grok-4.3

classification 🧬 q-bio.NC cs.AIcs.CY

keywords NeuroAIneuroscience-inspired AIAI capability gapsresearch roadmapinterdisciplinary trainingembodied AIsparse computationneuromodulation

0 comments

The pith

Neuroscience principles can address AI's gaps in physical interaction, robust learning, and energy efficiency.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Current artificial intelligence systems cannot interact effectively with the physical world, learn in ways that avoid brittleness, or operate without enormous energy and data costs. The paper identifies five neuroscience principles that map directly onto these gaps: co-design of body and controller, prediction through interaction, multi-scale learning with neuromodulatory control, hierarchical distributed architectures, and sparse event-driven computation. It organizes these into a staged research roadmap spanning near-term, mid-term, and long-term horizons. The work further argues that progress requires a new cohort of researchers trained at the neuroscience-engineering boundary together with supporting institutional structures. If the mapping holds, NeuroAI would produce more capable and sustainable artificial systems while also clarifying how biological brains compute.

Core claim

Current AI systems fall short in three areas: interacting with the physical world, learning in ways that avoid brittleness, and operating with reasonable energy and data costs. Neuroscience offers corresponding solutions including the co-design of body and controller, learning through prediction and interaction, multi-scale learning controlled by neuromodulation, hierarchical distributed processing, and sparse event-driven signaling. The paper maps these onto a research program spanning near-term, medium-term, and long-term goals, while emphasizing the need for researchers cross-trained in neuroscience and engineering along with supporting institutional structures.

What carries the argument

The NeuroAI framework that translates five neuroscience principles—co-design of body and controller, prediction through interaction, multi-scale learning with neuromodulatory control, hierarchical distributed architectures, and sparse event-driven computation—into solutions for AI capability gaps.

If this is right

Co-design of body and controller will enable AI to interact effectively with the physical world.
Prediction through interaction and multi-scale learning will reduce brittleness in AI systems.
Hierarchical distributed architectures and sparse event-driven computation will lower energy and data requirements.
Institutional support for interdisciplinary training and hardware access will be needed to advance the program.
Progress in NeuroAI will simultaneously improve artificial systems and deepen knowledge of biological neural computation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The roadmap implies that hardware development should prioritize neuromorphic or event-driven designs to realize efficiency gains.
Empirical validation could come from head-to-head comparisons of NeuroAI agents versus standard reinforcement-learning agents on robotic tasks.
The same principles might generate new hypotheses about how biological circuits achieve robustness and efficiency that could be tested in neuroscience experiments.
Institutional recommendations point to the value of shared community benchmarks and ethics guidelines for the emerging intersection field.

Load-bearing premise

The listed neuroscience principles are the right ones and can be translated into practical engineering solutions that close the three identified AI gaps.

What would settle it

Build and test prototype AI systems that incorporate the five neuroscience principles and measure whether they outperform current AI in physical-world interaction, robustness to distributional shift, and energy or data consumption.

read the original abstract

Neuroscience and Artificial Intelligence (AI) have made impressive progress in recent years but remain only loosely interconnected. Based on a workshop convened by the National Science Foundation in August 2025, we identify three fundamental capability gaps in current AI: the inability to interact with the physical world, inadequate learning that produces brittle systems, and unsustainable energy and data inefficiency. We describe the neuroscience principles that address each: co-design of body and controller, prediction through interaction, multi-scale learning with neuromodulatory control, hierarchical distributed architectures, and sparse event-driven computation. We present a research roadmap organized around these principles at near, mid, and long-term horizons. We argue that realizing this program requires a new generation of researchers trained across the boundary between neuroscience and engineering, and describe the institutional conditions: interdisciplinary training, hardware access, community standards, and ethics, needed to support them. We conclude that NeuroAI, neuroscience-informed artificial intelligence, has the potential to overcome limitations of current AI while deepening our understanding of biological neural computation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This is a workshop-derived position paper that organizes known neuroscience ideas into an AI roadmap but delivers no new technical results or tests.

read the letter

The main point is that this paper comes out of an NSF workshop and lays out a research agenda for NeuroAI. It names three AI shortcomings—poor physical interaction, brittle learning, and high energy and data costs—then ties them to five brain-inspired principles such as body-controller co-design, prediction via interaction, and sparse event-driven computation. A timeline of near-, mid-, and long-term steps follows, along with a call for cross-trained researchers and better institutional support like shared hardware and ethics guidelines.

Referee Report

2 major / 2 minor

Summary. This position paper, based on an NSF workshop, identifies three core capability gaps in current AI: inability to interact effectively with the physical world, learning processes that yield brittle systems, and unsustainable demands on energy and data. It nominates five neuroscience-derived principles—co-design of body and controller, prediction through interaction, multi-scale learning with neuromodulatory control, hierarchical distributed architectures, and sparse event-driven computation—as targeted remedies. The manuscript sketches a three-horizon research roadmap, argues for a new cohort of boundary-spanning researchers, and specifies institutional requirements (interdisciplinary training, hardware access, community standards, ethics) needed to realize the program. It concludes that NeuroAI can overcome present AI limitations while advancing insight into biological neural computation.

Significance. If the proposed linkages and roadmap are pursued, the work could usefully orient future research toward more embodied, robust, and efficient AI systems informed by biological principles. Its value lies in the explicit synthesis of workshop consensus into actionable near-, mid-, and long-term directions plus the structural recommendations for training and infrastructure; these elements are constructive even in the absence of new data or theorems.

major comments (2)

[Principles addressing gaps] The claim that the five listed neuroscience principles 'address' the three AI gaps is central yet remains at the level of enumeration; the text does not supply concrete mechanisms, worked examples, or citations showing, for instance, how sparse event-driven computation would measurably alleviate energy inefficiency relative to existing neuromorphic or sparse-training baselines.
[Research roadmap, long-term horizon] The long-term research horizon for hierarchical distributed architectures is presented without explicit contrast to current deep-learning hierarchies or specification of the additional biological constraints (e.g., neuromodulatory control) that would differentiate the proposed approach and justify the claim of overcoming brittleness.

minor comments (2)

The abstract and opening paragraphs would be clearer if the three capability gaps and the five principles were explicitly paired in a single table or enumerated list.
[Institutional conditions] The institutional-conditions section mentions ethics only briefly; adding one or two concrete examples of ethical issues arising at the neuroscience-AI boundary would strengthen the discussion without lengthening the paper substantially.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive review and recommendation of minor revision. Their comments have identified opportunities to strengthen the connections between principles and gaps as well as the roadmap details. We respond to each major comment below and indicate the revisions we will make.

read point-by-point responses

Referee: [Principles addressing gaps] The claim that the five listed neuroscience principles 'address' the three AI gaps is central yet remains at the level of enumeration; the text does not supply concrete mechanisms, worked examples, or citations showing, for instance, how sparse event-driven computation would measurably alleviate energy inefficiency relative to existing neuromorphic or sparse-training baselines.

Authors: We appreciate this observation. As a workshop-derived position paper, the manuscript prioritizes a high-level synthesis and forward-looking roadmap over exhaustive technical detail. We agree, however, that the linkages would benefit from greater specificity. In the revised manuscript we will add targeted citations and short illustrative examples for each principle-gap pairing, including references to neuromorphic hardware literature demonstrating energy reductions achievable via sparse event-driven computation relative to conventional approaches. revision: partial
Referee: [Research roadmap, long-term horizon] The long-term research horizon for hierarchical distributed architectures is presented without explicit contrast to current deep-learning hierarchies or specification of the additional biological constraints (e.g., neuromodulatory control) that would differentiate the proposed approach and justify the claim of overcoming brittleness.

Authors: We thank the referee for this suggestion. The long-term horizon is framed prospectively, yet we concur that explicit differentiation would strengthen the argument. We will revise the relevant section to include a direct comparison with existing deep-learning hierarchies and to specify how additional biological constraints such as neuromodulatory control can help address brittleness, drawing on established neuroscience findings. revision: yes

Circularity Check

0 steps flagged

No significant circularity: position paper synthesis without derivations or self-referential reductions

full rationale

The manuscript is a workshop-derived position paper that enumerates three AI capability gaps and nominates five neuroscience principles as remedies, then sketches research directions and training needs. No equations, fitted parameters, predictions, or closed-form derivations appear anywhere in the text. The central claims are prospective assertions about potential rather than deductive steps that could reduce to inputs by construction. Arguments draw on external workshop input and established domain principles without load-bearing self-citations or ansatzes smuggled via prior work. The derivation chain is therefore self-contained and non-circular.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on the domain assumption that the workshop correctly identified the fundamental AI gaps and that the listed neuroscience principles are translatable and sufficient to address them. No free parameters, new entities, or mathematical axioms are introduced.

axioms (2)

domain assumption The three capability gaps (physical interaction, brittle learning, energy/data inefficiency) are the fundamental limitations of current AI.
Stated directly in the abstract as the starting point for the roadmap.
domain assumption Neuroscience principles can be applied to engineer AI systems that overcome these gaps.
The paper maps specific neuro principles to each gap without further justification in the abstract.

pith-pipeline@v0.9.0 · 5634 in / 1601 out tokens · 51696 ms · 2026-05-10T06:04:29.185080+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

28 extracted references · 3 canonical work pages · 1 internal anchor

[1]

Neuroscience- inspired artificial intelligence.Neuron, 95(2):245–258, 2017

Demis Hassabis, Dharshan Kumaran, Christopher Summerfield, and Matthew Botvinick. Neuroscience- inspired artificial intelligence.Neuron, 95(2):245–258, 2017

2017
[2]

Exploring the bidirectional relationship between artificial intelligence and neuroscience.National Academies of Sciences, Engineering, and Medicine, 2024

Celia Ford, Eva Childers, and Sheena M Posey Nor- ris. Exploring the bidirectional relationship between artificial intelligence and neuroscience.National Academies of Sciences, Engineering, and Medicine, 2024

2024
[3]

Catalyzing next- generation artificial intelligence through neuroai.Na- ture communications, 14(1):1597, 2023

Anthony Zador, Sean Escola, Blake Richards, Bence ¨Olveczky, Yoshua Bengio, Kwabena Boahen, Matthew Botvinick, Dmitri Chklovskii, Anne Church- land, Claudia Clopath, et al. Catalyzing next- generation artificial intelligence through neuroai.Na- ture communications, 14(1):1597, 2023

2023
[4]

Marblestone, Greg Wayne, and Konrad P

Adam H. Marblestone, Greg Wayne, and Konrad P. Kording. Toward an integration of deep learning and neuroscience.Frontiers in Computational Neu- roscience, 10:94, 2016

2016
[5]

The language network as a natural kind within the broader landscape of the human brain.Nature Reviews Neuroscience, 25(5):289–312, 2024

Evelina Fedorenko, Anna A Ivanova, and Tamar I Regev. The language network as a natural kind within the broader landscape of the human brain.Nature Reviews Neuroscience, 25(5):289–312, 2024

2024
[6]

Se- jnowski

Jean-Marc Fellous, Gert Cauwenberghs, Cornelia Ferm¨ uller, Yulia Sandamirskaya, and Terrence J. Se- jnowski. NeuroAI and beyond. Preprint at https: //arxiv.org/abs/2601.19955, 2026

work page arXiv 2026
[7]

Sejnowski.ChatGPT and the Future of AI

Terrence J. Sejnowski.ChatGPT and the Future of AI. MIT Press, 2024

2024
[8]

Moravec.Mind Children: The Future of Robot and Human Intelligence

Hans P. Moravec.Mind Children: The Future of Robot and Human Intelligence. Harvard University Press, 1988

1988
[9]

Chiel and Randall D

Hillel J. Chiel and Randall D. Beer. The brain has a body: adaptive behavior emerges from interactions of nervous system, body and environment.Trends in Neurosciences, 20(12):553–557, 1997

1997
[10]

Pedersen, Taimoor Has- san, et al

Giulia D’Angelo, Jens E. Pedersen, Taimoor Has- san, et al. A benchmarking framework for embodied neuromorphic agents.Nature Machine Intelligence, 8:300–312, 2026

2026
[11]

Phase shift between joint rotation and actuation reflects dominant forces and predicts muscle activation patterns.PNAS nexus, 2(10):pgad298, 2023

Gregory P Sutton, Nicholas S Szczecinski, Roger D Quinn, and Hillel J Chiel. Phase shift between joint rotation and actuation reflects dominant forces and predicts muscle activation patterns.PNAS nexus, 2(10):pgad298, 2023

2023
[12]

Mali, Christopher L

Tommaso Salvatori, A. Mali, Christopher L. Buck- ley, Thomas Lukasiewicz, Rajesh P. N. Rao, Karl J. Friston, and Alexander Ororbia. A survey on neuro- mimetic deep learning via predictive coding.Neural networks, 195:108161, 2025

2025
[13]

A predictive approach to enhance time- series forecasting.Nature Communications, 16:8645, 2025

Skye Gunasekaran, Assel Kembay, Hugo Ladret, Rui- Jie Zhu, Laurent Perrinet, Omid Kavehei, and Jason Eshraghian. A predictive approach to enhance time- series forecasting.Nature Communications, 16:8645, 2025

2025
[14]

Self-supervised learning from images with a joint-embedding predictive archi- tecture

Mahmoud Assran, Quentin Duval, Ishan Misra, Piotr Bojanowski, Pascal Vincent, Michael Rabbat, Yann LeCun, and Nicolas Ballas. Self-supervised learning from images with a joint-embedding predictive archi- tecture. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 15619–15629, 2023

2023
[15]

Emotions: from brain to robot.Trends in cognitive sciences, 8(12):554–561, 2004

Michael A Arbib and Jean-Marc Fellous. Emotions: from brain to robot.Trends in cognitive sciences, 8(12):554–561, 2004

2004
[16]

MacKenzie W. Mathis. Leveraging insights from neuroscience to build adaptive artificial intelligence. Nature Neuroscience, 29:13–24, 2025

2025
[17]

Dynamical Mechanisms for Coordinating Long-term Working Memory Based on the Precision of Spike-timing in Cortical Neurons

Terrence J. Sejnowski. Dynamical mechanisms for coordinating long-term working memory based on the precision of spike-timing in cortical neurons. Preprint athttps://arxiv.org/abs/2512.15891, 2025

work page internal anchor Pith review Pith/arXiv arXiv 2025
[18]

Acetyl- choline demixes heterogeneous dopamine signals for learning and moving.Nature Neuroscience, 29:840– 850, 2026

Hee Jae Jang, Royall McMahon Ward, Carla EM Golden, and Christine M Constantinople. Acetyl- choline demixes heterogeneous dopamine signals for learning and moving.Nature Neuroscience, 29:840– 850, 2026

2026
[19]

Pardis Hajiseyedrazi, and Jason Eshraghian

Samuel Schmidgall, Rojin Ziaei, Jascha Achterberg, Thomas Miconi, Louis Kirsch, S. Pardis Hajiseyedrazi, and Jason Eshraghian. Brain-inspired learning in artificial neural networks: A review.APL Machine Learning, 2(2):021501, 2024

2024
[20]

Ames, and John C

Nikolai Matni, Aaron D. Ames, and John C. Doyle. A quantitative framework for layered multirate con- trol: Toward a theory of control architecture.IEEE Control Systems, 44:52–94, 2024

2024
[21]

Anthony M. Zador. A critique of pure learning and what artificial neural networks can learn from animal brains.Nature Communications, 10(1):3770, 2019

2019
[22]

Ali A. Minai. Deep intelligence: what AI should learn from nature’s imagination.Cognitive Computation, 16:2389–2404, 2024

2024
[23]

Brain-inspired energy efficient 10 technologies for next-generation artificial intelligence

Hillel J Chiel, Jay S Coggan, Gourav Datta, Jean- Marc Fellous, William RP Nourse, Roger D Quinn, and Peter J Thomas. Brain-inspired energy efficient 10 technologies for next-generation artificial intelligence. Biological cybernetics, 120(2):5, 2026

2026
[24]

I ’m born in 1 8 7 1

Rui-Jie Zhu, Zixuan Wang, Kai Hua, et al. Scaling latent reasoning via looped language models.arXiv preprint arXiv:2510.25741, 2025

work page arXiv 2025
[25]

Eshraghian, Max Ward, Emre O

Jason K. Eshraghian, Max Ward, Emre O. Neftci, Xinxin Wang, Gregor Lenz, Girish Dwivedi, Mo- hammed Bennamoun, Doo Seok Jeong, and Wei D. Lu. Training spiking neural networks using lessons from deep learning.Proceedings of the IEEE, 111(9):1016–1054, 2023

2023
[26]

Event- based backpropagation can compute exact gradi- ents for spiking neural networks.Scientific Reports, 11(1):12829, 2021

Timo C Wunderlich and Christian Pehle. Event- based backpropagation can compute exact gradi- ents for spiking neural networks.Scientific Reports, 11(1):12829, 2021

2021
[27]

Exploiting neuro-inspired dynamic sparsity for energy-efficient intelligent perception.Nature Communications, 16:9928, 2025

Shuo Zhou, Chang Gao, Tobi Delbruck, and Shih- Chii Liu. Exploiting neuro-inspired dynamic sparsity for energy-efficient intelligent perception.Nature Communications, 16:9928, 2025

2025
[28]

Schuman, Craig M

Dhireesha Kudithipudi, Catherine D. Schuman, Craig M. Vineyard, et al. Neuromorphic comput- ing at scale.Nature, 637:801–812, 2025. 11

2025