The Future of Artificial Intelligence Should Be Less Brain-Inspired and More Brain-Trained

Brain-inspired artificial intelligence and brain-trained artificial intelligence are two legitimate research directions that should ideally be pursued simultaneously. Nevertheless, we argue that in the foreseeable future, scientific research at the frontier of neuroscience and artificial intelligence should consider rebalancing the overall effort, and compared to the current situation, allocating a greater share of attention and resources to the second strategy. Indeed, leveraging neural signals to improve the performance of artificial intelligence models could be more naturally relevant for the requirements of general-purpose foundation models, and provide a more direct path for cumulative progress. In this paper, we establish a theoretical foundation for brain-trained artificial intelligence, with the objective of guiding future experimental research and applications. In particular, we review existing experimental methods, and explore two strategic uses of neural signals that could be prioritized for brain-trained foundation models: neural alignment, defined as the use of human brain data for a better alignment with human values, and neural reasoning, defined as the use of human brain data for a better emulation of human executive function. In both cases, we propose several algorithms that could be considered, building upon the neural signals of human valuation and decision-making. We also discuss how the neural scaling laws and the unreasonable effectiveness of human brain data for artificial intelligence models should be interpreted, reflect on the opportunities for neuroscience research, and formulate two theoretical upper bounds on the use of neuroimaging data for model training. Finally, we provide a synthesis in the form of a general strategy for future advances, which we call the Neuropoiesis strategy.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-17
DOI
https://doi.org/10.5281/zenodo.24317312
Primary Topic
Psychiatry, Mental Health, Neuroscience
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

The Future of Artificial Intelligence Should Be Less Brain-Inspired and More Brain-Trained

Maël Donoso
4 links
Zenodo (CERN European Organization for Nuclear Research)
Psychiatry, Mental Health, Neuroscience
preprint

The Future of Artificial Intelligence Should Be Less Brain-Inspired and More Brain-Trained

Maël Donoso
preprint en

Abstract

Brain-inspired artificial intelligence and brain-trained artificial intelligence are two legitimate research directions that should ideally be pursued simultaneously. Nevertheless, we argue that in the foreseeable future, scientific research at the frontier of neuroscience and artificial intelligence should consider rebalancing the overall effort, and compared to the current situation, allocating a greater share of attention and resources to the second strategy. Indeed, leveraging neural signals to improve the performance of artificial intelligence models could be more naturally relevant for the requirements of general-purpose foundation models, and provide a more direct path for cumulative progress. In this paper, we establish a theoretical foundation for brain-trained artificial intelligence, with the objective of guiding future experimental research and applications. In particular, we review existing experimental methods, and explore two strategic uses of neural signals that could be prioritized for brain-trained foundation models: neural alignment, defined as the use of human brain data for a better alignment with human values, and neural reasoning, defined as the use of human brain data for a better emulation of human executive function. In both cases, we propose several algorithms that could be considered, building upon the neural signals of human valuation and decision-making. We also discuss how the neural scaling laws and the unreasonable effectiveness of human brain data for artificial intelligence models should be interpreted, reflect on the opportunities for neuroscience research, and formulate two theoretical upper bounds on the use of neuroimaging data for model training. Finally, we provide a synthesis in the form of a general strategy for future advances, which we call the Neuropoiesis strategy.

Zenodo (CERN European Organization for Nuclear Research)
Psychiatry, Mental Health, Neuroscience
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon