For three decades the question of whether brain science and machine intelligence truly nourish one another has hung over a field that still borrows the brain's vocabulary while owing much of its engine to other sources. A large map of scholarly citations, drawn from hundreds of millions of works, suggests the traffic runs in both directions but along a narrow road, carried mostly by a handful of old classics and by general-purpose analytical tools. The bridge is strengthening as a movement called NeuroAI gathers prizes, institutes and manifestos, yet its foundations rest on a few aging ideas a
Neuroscience and AI Share a Real but Narrow Bridge, Built on Decades-Old Work
Something newer to cite than the works we produced 30 years ago
Mark
What's the headline finding here?
Mimi
The bridge between neuroscience and AI exists and is strengthening, but it's narrow. Much of what crosses is software tools one way and pre-2005 concepts the other.
Mark
Surprising that neuroscience influences AI more than the reverse?
Mimi
After correcting for field size, yes. A small fixed canon of classic papers, like Bi and Poo from 1998, drives it.
Luke
Careful with the recent surge, though. The author says 2024 and 2025 data are unreliable because OpenAlex is being rebuilt.
Mimi
Right, and the upturn predates those years, so it isn't an artifact. Its size just can't be measured yet.
Luke
There's also a dilution problem. Part of the post-2020 rise may be neuroscience shrinking as a share of a booming AI literature.
Mark
So the revival might be partly arithmetic?
Mimi
Partly. The raw share rose more modestly than the odds ratio shows.
Luke
And the raw numbers say AI overtook neuroscience as the bigger influence in 2018. Which view you pick matters.
Mimi
The author argues field-size correction is the honest one. Whether this is a sustained renaissance, nobody knows yet.
The Pulse
- Citation traffic between neuroscience and AI is growing fast, but it is pressed into a thin channel of old papers and shared software.
- The NeuroAI coalition's momentum since 2020 is colliding with doubts about whether the surge is real revival or just a larger AI literature diluting the numbers.
- Data from 2024 and 2025 is unreliable while OpenAlex is being rebuilt, so the newest chapter of the story is still being written in uncertain ink.
- Influence is concentrated in a few tech giants, elite universities and a small circle of researchers, raising the risk of narrow consensus.
- The path forward demands that neuroscience deliver fresh results or methods, such as tools for interpreting AI systems or more efficient architectures, before the bridge lands on a canon that is thirty years stale.
For three decades the question of whether brain science and machine intelligence truly nourish one another has hung over a field that still borrows the brain's vocabulary while owing much of its engine to other sources. A large map of scholarly citations, drawn from hundreds of millions of works, suggests the traffic runs in both directions but along a narrow road, carried mostly by a handful of old classics and by general-purpose analytical tools. The bridge is strengthening as a movement called NeuroAI gathers prizes, institutes and manifestos, yet its foundations rest on a few aging ideas and a few well-connected places. Whether the next generation of brain research can hand artificial intelligence something it has not already taken is the open question the data leaves behind.
The analysis started with an enormous body of scholarship: about 458 million works and nearly 2.9 billion citation links drawn from OpenAlex, an open database. Its author, writing in The Transmitter, set out to test a claim the NeuroAI movement had largely taken for granted, that neuroscience and AI should feed each other. The question was whether they actually do.
The history gives the question weight. Early AI borrowed freely from the brain, then grew into a discipline of its own. Neural networks, rebranded as deep learning in the mid-2000s, still dominate, yet their link to biology is now mostly a matter of inspiration rather than mechanism. Meanwhile the brain remains the only known intelligence that is general in the full sense, and over the past decade a coalition has pushed NeuroAI into institutes, endowments and events, with Nobel laureates among its voices.
The data suggest a real but narrow bridge. Adjusted for field size, neuroscience has shaped AI more than the reverse across three decades, and that influence rests on a small, fixed canon of classic papers. Its pace has quickened since 2016 and surged after 2020, matching the timing of the NeuroAI manifestos, and most of the neuroscience AI cites comes from cognitive work rather than cellular or molecular research.
The author adds two cautions. The 2024 and 2025 figures are shaky while the database is being rebuilt, and part of the recent rise may simply reflect AI's booming literature crowding neuroscience into a smaller share. Read as raw shares rather than field-adjusted odds, the picture looks different, though the author considers the adjusted view the more honest one.
The strongest connections are split evenly between general-purpose tools, such as statistical and machine-learning software, and a classical neuroscience canon whose newest landmark dates to 2003. Influence is also concentrated: a few large technology firms and elite universities hold much of the weight, and a small group of researchers carry much of the rest.
The conclusion is that the bridge is tool-heavy and concentrated. The brain runs on roughly 20 watts and learns from few examples, qualities that remain real engineering goals. For neuroscience to stay relevant, it must offer new results or methods, including ways to open AI's black boxes and ideas for efficient architectures. The challenge now is to give AI something newer to cite than work from thirty years ago.
The analysis began with a very large pile of paper: roughly 458 million works from OpenAlex, an open database of publications, and the 2.9 billion citation links among them. The author, writing in The Transmitter, used that web to ask a question the NeuroAI movement had mostly skipped. Neuroscience and AI should fuel each other, supporters say. Do they actually?
The backdrop is a long history. Early AI borrowed from the brain and tried to reverse-engineer its tricks. Over the following decades, building intelligent machines became its own discipline. The neural network, renamed deep learning in the mid-2000s, still dominates, but its resemblance to the brain is now mostly conceptual. The author says the recent achievements in AI owe neuroscience surprisingly little in their machinery. Yet the human brain remains the only known intelligent system in the full, general sense. Over the past decade, a coalition has pushed the idea under the name NeuroAI. It includes Geoffrey Hinton, John Hopfield and Demis Hassabis, who won Nobel Prizes in 2024 for AI-related work. It has brought dedicated institutes, endowments, manifestos and events.
The short answer from the data is that the bridge is real and gaining momentum, but narrow. Corrected for field size, neuroscience has influenced AI more over the past three decades than the reverse. From 1995 to 2025, AI papers were more likely to cite neuroscience than neuroscience papers were to cite AI. The cause is a small, fixed canon of classic neuroscience papers that underpin artificial neural networks.
What has changed is the pace. Neuroscience's share of citations to AI has risen steadily since 2006, accelerating from 2016 to 2019 as AI and machine-learning tools spread through neural data analysis. AI's share of citations to neuroscience stayed fairly flat and then declined from 2006 to 2019. It then recovered, with 2020 to 2025 showing the steepest slope of the series, which fits the timing of the NeuroAI manifestos. Of the neuroscience that AI papers cite, about 90 percent comes from cognitive neuroscience and under 5 percent from cellular and molecular work.
The author attaches two caveats. Data for 2024 and 2025 are shaky because OpenAlex is being rebuilt and recent records are still being relinked, so the magnitudes can't be trusted, though the upturn predates those years. And the post-2020 rise may partly reflect neuroscience shrinking as a share of a booming AI literature. Between 2020 and 2023 the share of AI citations going to neuroscience did rise, but more modestly than the odds ratio suggests. The raw numbers tell a different story: as a share of what each field cites, AI's influence on neuroscience took the lead in 2018, simply because there is more AI work to cite. The author calls correcting for field size the more honest view.
About half of the strongest bridge works are tools: general-purpose statistical and machine-learning software such as the BraTS tumor-segmentation challenge, the lme4 package, SciPy, ROC analysis and XGBoost. The other half are a classical neuroscience canon: Bi and Poo on spike-timing-dependent plasticity (1998), Izhikevich's spiking neuron models (2003), work on visual attention, object recognition, receptive fields and sparse coding, and working memory and short-term plasticity. The newest is from 2003. That does not mean neuroscience's influence is fading, since AI cites more neuroscience than ever, but in two decades no new landmark paper has become essential to both fields.
The authors read across the border include Yoshua Bengio, Hinton and Antonio Torralba on the AI side, and Karl Friston, Peter Dayan, Wulfram Gerstner, Henry Markram, Christof Koch and Aude Oliva among computational neuroscientists. The skew is institutional. Alphabet (Google and DeepMind) and Microsoft top the ranking, followed by MIT, UC Berkeley, Harvard, University College London and Stanford. The top 100 institutions hold nearly 40 percent of bridge weight; the top 100 authors hold under 7 percent.
The author's conclusion is that the bridge is tool-heavy and concentrated. The brain runs on roughly 20 watts and learns from a handful of examples, and energy and sample efficiency, robustness and out-of-distribution generalization are real engineering targets. To stay relevant, neuroscience must supply new results or methods, including tools for opening AI's black boxes and ideas for efficient architectures. Institutional concentration risks narrow consensus. The task now is to give AI something newer to cite than work produced 30 years ago.
Notable Quotes
The bridge is real and gaining momentum, but it is also narrow, tool-heavy and concentrated.— Essay author, The Transmitter
Want the full story?
Read the original at The Transmitter ↗
Technology