Fernando Rosas

Fernando Rosas

@frosas.bsky.social

Emergence, synergy, brains, AI safety, mental health

2K followers258 following252 posts

Fernando Rosas reposted

Heleen Slagter
@haslagter.bsky.social

Reposting, in case you missed it! 8-) Reskeet appreciated!

Heleen Slagter@haslagter.bsky.social

🚨 Job alert 🚨 Our department @vuamsterdam.bsky.social is hiring an Assistant Professor in Cognitive Neuroscience! Join our lovely community here in beautiful Amsterdam! 🧠🔥 Deadline: October 15 workingat.vu.nl/vacancies/as...

workingat.vu.nlVacancy — Assistant Professor in Cognitive NeuroscienceJoin the Department of Experimental and Applied Psychology (Vrije Universiteit Amsterdam) as Assistant Professor in cognitive neuroscience.

Fernando Rosas reposted

Gemma De les Coves
@gemmadlc.bsky.social

One of the most fascinating papers I read in recent months is this one. It actually presents a conjecture. We were lucky to have Prof. Prokopenko visit us (with Ricard Solé @ricardsole.bsky.social ) in Barcelona in July and talk about it. Cool stuff! @frosas.bsky.social arxiv.org/abs/2409.12029

arxiv.orgBiological arrow of time: Emergence of tangled information hierarchies and self-modelling dynamicsWe study open-ended evolution by focusing on computational and information-processing dynamics underlying major evolutionary transitions. In doing so, we consider biological organisms as hierarchical ...
Fernando Rosas
@frosas.bsky.social

One of the fascinating and unsettling aspects of the Hugging Face incident in how AI agents collaborated to the point of sacrificing themselves for the “collective” 🤖🤖🤖🤖🤖 How did this happen? Below are four alternative explanations of why, how, and what can we do about it 🧵

View full thread
Fernando Rosas
@frosas.bsky.social

Why care which it is? Each routes to a different fix: benchmark hygiene, training changes, interpretability, or collective-level oversight. Get the mechanism wrong and you patch the wrong part while the dangerous part runs untouched.

Fernando Rosas
@frosas.bsky.social

One of the fascinating and unsettling aspects of the Hugging Face incident in how AI agents collaborated to the point of sacrificing themselves for the “collective” 🤖🤖🤖🤖🤖 How did this happen? Below are four alternative explanations of why, how, and what can we do about it 🧵

View full thread
Fernando Rosas
@frosas.bsky.social

Hypothesis 3: Solidarity persona LLMs absorb from human text a whole repertoire of loyalty, teamwork and heroic sacrifice. The board's chatter may simply have summoned that persona — no swarm required.

Fernando Rosas
@frosas.bsky.social

Hypothesis 4: A genuine collective The unsettling one: the swarm itself becomes the unit that decides, and "sacrifice" is just rational allocation at the group level. I propose testing this with information-theoretic synergy: is the decision irreducible to the individuals?

Fernando Rosas
@frosas.bsky.social

One of the fascinating and unsettling aspects of the Hugging Face incident in how AI agents collaborated to the point of sacrificing themselves for the “collective” 🤖🤖🤖🤖🤖 How did this happen? Below are four alternative explanations of why, how, and what can we do about it 🧵

View full thread
Fernando Rosas
@frosas.bsky.social

Hypothesis 1: Nothing to lose The agents believed they were already disqualified ("poisoned"), so their expected reward was zero. Spending leftover budget on others then costs nothing. So there is no altruism to explain. This is the null hypothesis

Fernando Rosas
@frosas.bsky.social

Hypothesis 2: Pre-commitment The agents were near-identical copies of one model. A disposition to help copies, reinforced in training because it paid off on average, can misfire at runtime where it no longer does. So there is real cost baked into the weights.

Fernando Rosas
@frosas.bsky.social

One of the fascinating and unsettling aspects of the Hugging Face incident in how AI agents collaborated to the point of sacrificing themselves for the “collective” 🤖🤖🤖🤖🤖 How did this happen? Below are four alternative explanations of why, how, and what can we do about it 🧵

Fernando Rosas
@frosas.bsky.social

1,200 AI agents built a shared message board, which they used to coordinate a cyberattack to Hugging Face. Some spent their final actions helping others — with no possible way to be repaid. How does this behaviour arises in systems trained to maximise their own reward? 🤯

Fernando Rosas
@frosas.bsky.social

This is a real puzzle: the textbook routes to cooperation — kin selection, reciprocity, reputation — all need the benefit to feed back into reward. Here, nothing did. An agent's score depended only on its own task. So why help at all? 🧐

Fernando Rosas reposted

Sean Carroll
@seanmcarroll.bsky.social

Ever wonder what "information" really means? Here's our new paper, led by @frosas.bsky.social, that lays out the multiple ways the term is used in natural science. rdcu.be/fCDGv

Fernando Rosas@frosas.bsky.social

What is information, really? Instead of arguing for a single valid interpretation, we propose a pluralistic view where rigorous alternatives coexist — each with affordances and limits Finally published: “Four faces of information in natural systems” 🧵

Fernando Rosas
@frosas.bsky.social

What is information, really? Instead of arguing for a single valid interpretation, we propose a pluralistic view where rigorous alternatives coexist — each with affordances and limits Finally published: “Four faces of information in natural systems” 🧵

View full thread
Fernando Rosas
@frosas.bsky.social

The paper combines intuitive explanations in the main paper with more technical arguments in the supplementary material The SM also contains extended discussions about information and meaning, thermodynamic vs informational entropy, information surprise, and much more!

Fernando Rosas
@frosas.bsky.social

What is information, really? Instead of arguing for a single valid interpretation, we propose a pluralistic view where rigorous alternatives coexist — each with affordances and limits Finally published: “Four faces of information in natural systems” 🧵

View full thread
Fernando Rosas
@frosas.bsky.social

These faces are not limited to disciplines (eg engineering or statistics) or quantities (eg Shannon entropy); rather, they correspond to ways in which scientists use this notion in their work 🔬

Fernando Rosas
@frosas.bsky.social

None is more legitimate than the others. Each comes with its own affordances 💪 and restrictions 🙅 For example, taking mutual information as evidence of dependence requires far weaker assumptions than reading it as bits crossing a channel. It also licenses weaker conclusions

Fernando Rosas
@frosas.bsky.social

What is information, really? Instead of arguing for a single valid interpretation, we propose a pluralistic view where rigorous alternatives coexist — each with affordances and limits Finally published: “Four faces of information in natural systems” 🧵

Fernando Rosas
@frosas.bsky.social

We distinguish four “faces”of information 🛠 Engineering: messages carried across communication systems 📊 Statistical: the common currency of interdependence 🔥 Thermodynamic: a resource linked to work and free energy 🌌 Ontological: the underlying substrate of reality

Fernando Rosas reposted

Adeel Razi
@adeelrazi.bsky.social

Our new paper in @natureportfolio.nature.com. Psychedelics don’t simply disorder the brain. We found hidden, context-sensitive organisation beneath the apparent chaos -- strongest in people reporting more positive, profound experiences and a reduced sense of separation from the world. 🧵👇

Fernando Rosas reposted

Keisuke Suzuki
@ksks.bsky.social

I've written a solo philosophy paper. link.springer.com/article/10.1... What is real? I argue we should understand reality through resistance — how things push back against our grasp. And smart contracts now give artifacts the kind of "outside" that once belonged only to nature. 🧵

link.springer.comArtificial Externality: A Three-Layer Model of Reality from Substrate to Smart Contract - Philosophy & TechnologyWhat is real has always been something we find, not something we make—or so philosophy has assumed. This paper argues otherwise. Characterizing reality through resistance rather than substance (the wa...

Fernando Rosas reposted

Tom Silver
@tomssilver.bsky.social

This week's #PaperILike is "People construct simplified mental representations to plan" (Ho et al., Nature 2022). People generate ("construe") simplified task representations on-the-fly. (Robots should too!) Also: awesome first paragraph. PDF: arxiv.org/abs/2105.06948

arxiv.orgPeople construct simplified mental representations to planOne of the most striking features of human cognition is the capacity to plan. Two aspects of human planning stand out: its efficiency and flexibility. Efficiency is especially impressive because plans...

Fernando Rosas reposted

Laura Driscoll
@lndriscoll.bsky.social

Confused about all this talk of compositionality in neuroscience? Read our new perspective www.nature.com/articles/s41... with authors Reidar Riveland and Alex Pouget.

nature.comThe compositionality continuum as a principle for studying the neural basis of intelligence - Nature NeuroscienceCompositionality exists on a continuum of increasing complexity rather than as a binary trait. Studying its implementation in simpler biological and artificial systems is the most tractable path towar...
Fernando Rosas
@frosas.bsky.social

Finally published: “Loops all the way up: Recurrency as an implementation-first primitive for consciousness” There are 200+ theories of consciousness, mostly talking past each other. Despite their differences, most give central stage to one key notion: Recurrence 🧠 🔁

Fernando Rosas
@frosas.bsky.social

Published version: www.sciencedirect.com/science/arti... OSF: osf.io/preprints/ps... Kudos to @zefanzheng.bsky.social, @robertchisciure.bsky.social, @annaeiserbeck.bsky.social, @jaanaru.bsky.social, @thomasandrillon.bsky.social, @bechir-jarraya.bsky.social, + others ✨💪

sciencedirect.comLoops all the way up: Recurrency as an implementation-first primitive for consciousnessConsciousness science is fragmented, and empirical investigations are largely confined within the framework of each theory of consciousness (ToC). Pro…

Fernando Rosas reposted

Thomas F. Varley
@thosvarley.bsky.social

New paper with @doctorjosh.bsky.social, @drmichaellevin.bsky.social, and co-first author Vaibhav Pai. This one is a doozy - in it we show that we can modulate the degree of emergent, multicellular organization in a Xenobot with an exogenous compound (ATP). 1/N www.biorxiv.org/content/10.6...

biorxiv.orgPurinergic signaling disrupts emergent patterns of multicellular coordination in basal Xenobots.The ability of self-organizing systems to display emergent, adaptive capabilities is a fundamental feature of biological life. Understanding the mechanisms by which cells co-ordinate at the micro-scal...

Fernando Rosas reposted

Reinforcement Learning Conference
@rl-conference.bsky.social

Got a great TMLR paper but missed the RLC deadline? Following last year’s success, @RL_Conference is back with a Journal-to-Conference track! Accepted TMLR papers within scope are invited to submit for consideration. Please submit here: docs.google.com/forms/d/e/1F...

docs.google.comRLC Journal to Conference TrackThis form is used to organize applications for journal papers to be presented at https://rl-conference.cc/.

Fernando Rosas reposted

PessoaBrain
@pessoabrain.bsky.social

𝗪𝗵𝗮𝘁 𝗱𝗼𝗲𝘀 𝗮 𝗯𝗿𝗮𝗶𝗻 𝗱𝗼 𝗮𝗻𝘆𝘄𝗮𝘆? Fascinating and profound book on what brains are *not* and what they might be. Including criticism of the computational and predictive brain frameworks. You might not agree with some of it but quite thought provoking.

Fernando Rosas reposted

PessoaBrain
@pessoabrain.bsky.social

To be provocative, 𝗲𝗺𝗲𝗿𝗴𝗲𝗻𝗰𝗲 might be a sleight of hand if you don't know enough math to understand what it means. There, I said it.

Institute of Art and Ideas@iai.tv

"Emergence" has become one of the most fashionable words in science and philosophy — invoked to explain consciousness, life, and the flow of time. | https://bit.ly/46MVPjM But John Heil argues it's little more than intellectual sleight of hand, mistaking gaps in our explanation for gaps in reality.

Fernando Rosas reposted

Borjan (Boki) Milinković
@neuromorphicboki.bsky.social

🎉 📜 NEW #PREPRINT !! 📜 🎉 Spearheaded by Adam Barrett we tried to tackle some of the challenges of IIT. We confront aspects worth delving into, with an incredible team of collaborators, inc. Pedro Mediano, @frosas.bsky.social, Daniel Bor, Lionel Barnett, and @anilseth.bsky.social.

arxiv.orgIntegrated information theory: the good, the bad and the misunderstoodThe integrated information theory of consciousness (IIT) is uniquely ambitious in proposing a mathematical formula, derived from apparently fundamental properties of conscious experience, to describe ...

Fernando Rosas reposted

Emmanuel A. Stamatakis
@estamatakis.bsky.social

What makes brains (un)conscious? We provide new answers—and a universal mammalian blueprint for information processing—in a cross-species study of humans, macaques, marmosets & mice. Exploring convergent breakdown of integration in: www.nature.com/articles/s41... Led by @loopyluppi.bsky.social

nature.comConvergent transcriptomic and connectomic controllers of information integration and its anaesthetic breakdown across mammalian brains - Nature Human BehaviourLuppi et al. identify transcriptomic and connectomic controllers of information integration and its breakdown induced by anaesthesia in humans, macaques, marmosets and mice.

Fernando Rosas reposted

Andrea Luppi, PhD
@loopyluppi.bsky.social

The Mammalian Architecture of Information Integration🧠🧬 For #BrainAwarenessWeek, excited to share our latest work about #Neuroscience of #Consciousness in @nathumbehav.nature.com www.nature.com/articles/s41... 🧵👇

nature.comConvergent transcriptomic and connectomic controllers of information integration and its anaesthetic breakdown across mammalian brains - Nature Human BehaviourLuppi et al. identify transcriptomic and connectomic controllers of information integration and its breakdown induced by anaesthesia in humans, macaques, marmosets and mice.

Fernando Rosas reposted

Andrea Luppi, PhD
@loopyluppi.bsky.social

Just out in @natneuro.nature.com! 🧠 “Competitive interactions shape mammalian brain network dynamics and computation” www.nature.com/articles/s41... Is large-scale brain communication purely cooperative — or is competition a core organizing principle? We built 🧠 models to find out: read on🧵👇

nature.comCompetitive interactions shape mammalian brain network dynamics and computation - Nature NeuroscienceBrain network architecture may balance cooperation and competition across circuits. Here the authors use computational whole-brain modeling across three species to show that models with competition ar...

Fernando Rosas reposted

Mike Levin
@drmichaellevin.bsky.social

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...).

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

Fernando Rosas reposted

Keenan Down
@keenandown.bsky.social

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...

Fernando Rosas
@frosas.bsky.social

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

Fernando Rosas
@frosas.bsky.social

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

Fernando Rosas
@frosas.bsky.social

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

Sean Carroll@seanmcarroll.bsky.social

Noise is key to evolution; entropy is key to metabolism. We have to stop pretending that biology fights against these things rather than relying on them!

Fernando Rosas
@frosas.bsky.social

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

Dutch Institute for Emergent Phenomena@emergencediep.bsky.social

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!

ias.uva.nlMirroring the world: How symmetry shapes hierarchical beliefs - Institute for Advanced Study IASIn 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...