Fernando Rosas

Machines all the way up and cognition all the way down: Updating the machine metaphor in biology Cell and developmental biology (CDB) offer numerous remarkable examples of collective adaptive plasticity, as cells coordinate to implement large-scal…

Final version of this paper with Richard Watson is out!
www.sciencedirect.com/science/arti...
"Machines all the way up and cognition all the way down: Updating the machine metaphor in biology"

(quite a bit different than the original preprint at osf.io/preprints/os...).

Ok I should read the paper then. Thanks for the clarification 🙏

I thought that this new paper was saying that not only pairwise mechanisms can generate some high order behaviour, but that they can generate all of them…

(… and hence reverse inference from behaviour to mechanisms is kind of cursed)

But maybe I’m misunderstanding the paper?

One of the points our paper touches is that one doesn’t need high order mechanisms to generate high order behaviour. For example, one can have statistical synergy with pairwise (but frustrated) Ising models

So the expressivity is only in terms of mechanisms and not related to behaviours?

Disentangling high-order mechanisms and high-order behaviours in complex systems - Nature Physics Nature Physics - Disentangling high-order mechanisms and high-order behaviours in complex systems

Just read the abstract and intro and looks extremely interesting, congrats! One question: how are the arguments raised here related to the distinction between mechanism and behaviours discussed here? arxiv.org/abs/2203.12041

Happy to announce this new preprint!

In it, we use info decomp (ΦID) on fMRI in Alzheimer's and MCI to explore how info-dynamic representations change.

AD saw big decreases in synergy ('deductive' information) and increases in redundancy.

Check it out here:
👇
www.biorxiv.org/content/10.6...

The key idea is mesa-optimisation: just as humans were shaped by a process of fitness maximisation but are not fitness maximisers themselves (we do other stuff, like listen to music), LLMs are shaped by next token prediction but that is not what they do

Hot take 🔥

Why math classes work well in uni? Not because that specific math will be useful later, but because solving hard problems makes you build instrumental skills, which enable you to solve other problems later in life

… and that is exactly why next token prediction works so well

Indeed, rather than fight against it, it seems life surfs on entropy 🏄

Could it be that the issue is not so much computability in principle, but that waves may be a way to increase the signal-to-noise ratio?

Tomorrow in Amsterdam I’ll be giving a talk about my most recent work on abstractions and world modelling
👇🏽👇🏽

Mirroring the world: How symmetry shapes hierarchical beliefs - Institute for Advanced Study IAS In this edition of the DIEP seminar series, Dr. Fernando Rosas, lecturer in Computer Science and AI at the University of Sussex and Honorary Research Fellow Imperial College London will be talking ab...

DIEP seminar tomorrow 11am! 🪞
Join us to attend a talk by @frosas.bsky.social (lecturer at the University of Sussex) titled

“Mirroring the world: How symmetry shapes hierarchical beliefs” 🤼

Abstract & registration: ias.uva.nl/content/even...
See you there!

Why don’t neural networks learn all at once, but instead progress from simple to complex solutions? And what does “simple” even mean across different neural network architectures?

Sharing our new paper @iclr_conf led by Yedi Zhang with Peter Latham

arxiv.org/abs/2512.20607

Description Please note that job descriptions are not exhaustive, and you may be asked to take on additional duties that align with the key responsibilities ment...

There is a new Postdoctoral Research Associate position available in my group, to work on computational analysis of human neuroimaging workflows to develop robust, interpretable and scalable biomarkers of cognitive brain health.

Closing date 10th Feb 2026.

www.imperial.ac.uk/jobs/search-...

We also provide a roadmap for future directions organising research priorities into sequential stages.

Kudos to the awesome team brilliantly lead by Constanza Baquedano and Mar Estarellas! ✨

These benefits are not just fleeting states: longitudinal MRI studies suggest that cumulative greenspace exposure is linked to better white matter integrity and brain structure over time. Nature isn't just a luxury, but a scaffold for neurocognitive resilience

Across modalities, we found that nature shifts the brain into a distinct state via a “restorative cascade”: enhances sensory restoration and coherence, down-regulates stress circuitry, and fosters functional integration in self-related networks

Your brain on nature: A scoping review of the neuroscience of nature exposure The relationship between natural environments and human cognition has gathered increasing attention across disciplines, including neuroscience, enviro…

Finally published:
“Your brain on nature: A scoping review of the neuroscience of
nature exposure”
www.sciencedirect.com/science/arti...

A synthesis of multiple EEG, fMRI, and fNIRS studies to map the mechanisms behind the restoring effects of nature on the brain 🧠🌱

We're running a 5th edition of the always-exciting UCL Summer School on Consciousness and Metacognition this year, 8th-10th July 2026 in London. Accommodation and travel expenses are covered.

For more information and how to apply, check out metacoglab.org/summer-schoo...

cool work about information decomposition in artificial systems!

Synergistic information processing emerges through learning in LLMs and ablating these synergistic components causes large performance drops

#NeuroAI

👇👇👇

A Brain-like Synergistic Core in LLMs Drives Behaviour and Learning The independent evolution of intelligence in biological and artificial systems offers a unique opportunity to identify its fundamental computational principles. Here we show that large language models...

Preprint time:
A Brain-like Synergistic Core in LLMs Drives Behaviour and Learning
arxiv.org/abs/2601.06851

LLMs are clearly different from actual brains… but could they share a similar functional architecture related to how they process information?

Looking forward to participate in this workshop on high-order interactions and multivariate information theory:
mithic2026.github.io

The Entangled Brain Popular neuroscience accounts often focus on specific mind-brain aspects like addiction, cognition, or memory, but The Entangled Brain tackles a much bigger ...

𝗧𝗵𝗲 𝗘𝗻𝘁𝗮𝗻𝗴𝗹𝗲𝗱 𝗕𝗿𝗮𝗶𝗻
Annual reminder that the book is open access.
How do we think of the brain as a deeply interconnected system with highly distributed, non hierarchical processing.
Want to learn about the brain from a fresh perspective?
#neuroskyence
mitpress.mit.edu/978026254460...

Hi RL Enthusiasts!

RLC is coming to Montreal, Quebec, in the summer: Aug 16–19, 2026!

Call for Papers is up now:
Abstract: Mar 1 (AOE)
Submission: Mar 5 (AOE)

Excited to see what you’ve been up to - Submit your best work!
rl-conference.cc/callforpaper...

Please share widely!

Evolution by natural induction Abstract. It is conventionally assumed that all evolutionary adaptation is produced, and could only possibly be produced, by natural selection. Natural ind

A new, long paper on evolution - natural induction - split into 2:

royalsocietypublishing.org/rsfs/article...

royalsocietypublishing.org/rsfs/article...

@RichardWatson90 and Tim Lewens

Cooperation is a universal feature of complex systems, from the origins of life and microbiomes to societies. What universal patterns can be found in these systems? Here's our new @pnas.org paper. @jordipinero.bsky.social @artemyte.bsky.social @sfiscience.bsky.social www.pnas.org/doi/epdf/10....

Efficient event-based delay learning in spiking neural networks - Nature Communications Inspired by how neurons in the brain communicate, Spiking Neural Networks are gaining attention as efficient models for solving spatiotemporal AI tasks. The authors introduce a training method for syn...

Our paper on event-based delay learning is now published! @neworderofjamie.bsky.social @drtnowotny.bsky.social
TL;DR: It’s now possible to train synaptic delays in large-scale spiking neural networks with high temporal precision—even in recurrent connections.
www.nature.com/articles/s41...

Thanks Blake!

Are there other papers providing complementary evidence for this that you could recommend?