Pinned Tweet
It is the deepest honor to have been joined by Michael Levin (@drmichaellevin), Victoria Klimaj, Zahra Sheikhbahaee (@zah_bah), Dalton Sakthivadivel (@DaltonSakthi), Adeel Razi (@adeelrazi), David Ha (@hardmaru), Nick Hay, Kevin Schmidt, Irina Rish (@irinarish), David Krakauer (@sfiscience), Melanie Mitchell (@MelMitchell1), Samuel Gershman (@gershbrain), and Joshua Tenenbaum in organizing this special issue of the Royal Society’s (@RSocPublishing) Philosophical Transactions A:
“World models, A(G)I, and the Hard problems of life-mind continuity: Toward a unified understanding of natural and artificial intelligence”
royalsocietypublishing.org/r…
This collection was motivated by a question with far reaching implications, ranging from the fundamental nature(s) of mind to choices that may determine the future of our civilization/species: what kinds of world modeling capabilities are likely to be realized by which kinds of minds and what world might we be in with respect to increasingly advanced artificial intelligences?
Will the scaling and refinement of present approaches result in AI with human-like (and beyond) cognitive abilities, or do we need radically different paradigms that more closely follow the principles of natural intelligence? Learning “world models” to predict/compress information may be how biological learners so efficiently learn (to learn) to achieve goals and generalize that knowledge across a broad range of task environments. World models may also be useful for reverse-engineering forms of “System 2” cognition, or the self-reflexive, deliberate, multi-step reasoning associated with cognitive capabilities that may be unique to humans. Predictive models that reflect how the world may be causally modified by actions allow agents to adaptively control their behavior with flexibility and context-sensitivity. Spatiotemporally and causally coherent models of the physical world may not only be the key for creating AIs that we can rely on for real-world deployment, but may even be the (dynamic) core of conscious cognition.
The contributions to this special issue consider the varieties of world models worth modeling from diverse points of view:
Douglas Hofstadter explores whether sufficiently coherent self-referential world modeling could ground meaning, consciousness, and a genuine “I” in future AI systems.
David Krakauer (@sfiscience), Melanie Mitchell (@MelMitchell1), and John Krakauer (@blamlab) examine the principles of emergent intelligence from a complex systems perspective.
Alexander Ku (@alex_y_ku), Declan Campbell, Xuechunzi Bai (@baixuechunzi), Jiayi Geng (@JiayiiGeng), Ryan Liu (@theryanliu), Raja Marjieh (@RajaMarjieh), R. Thomas McCoy (@RTomMcCoy), Andrew Nam, Ilia Sucholutsky (@sucholutsky), Liyi Zhang (@LiyiZhang_Leo), Jian-Qiao Zhu (@JQ_Zhu), and Thomas Griffiths (@cocosci_lab) argue for using the tools of cognitive science to understand and evaluate LLMs across multiple levels of analysis.
Evelina Leivada (@EvelinaLeivada), Gary Marcus (@GaryMarcus), Fritz Günther, and Elliot Murphy (@ElliotMurphy91) test whether LLMs deeply understand language and the “world behind words,” or primarily learn surface statistical regularities.
Pedro Tsividis (@ptsividis), João Loula, Jake Burga, Juan Pablo Rodriguez, Sergio Arnaud, Nate Foss (@_npfoss), Andres Campero, Ajay Subramanian (@ajaysub110), Thomas Pouncy, Samuel Gershman (@gershbrain), and Joshua Tenenbaum introduce a theory-based meta-learning architecture inspired by the remarkable flexibility and efficiency of human cognition.
Eunice Yiu (@eunice_yiu_), Kelsey Allen, Shiry Ginosar (@shiryginosar), and Alison Gopnik (@AlisonGopnik) explore empowerment, controllability, and causal reasoning as means of understanding the remarkable learning abilities of both child and adult minds.
Nadav Amir, Stas Tiomkin, and Angela Langdon investigate how goals shape the structure of experience and how the world modeling abilities of natural intelligences may be inseparable from values.
Vickram Premakumar, Michael Vaiana, Florin Pop (@FlorinPop17), Judd Rosenblatt (@juddrosenblatt), Diogo Schwerz de Lucena, Kirsten Ziman, and Michael Graziano show unexpected benefits of self-modeling as an inductive bias and regularizer for training artificial agents.
Hanlin Zhu, Baihe Huang, and Stuart Russell analyze why model-based reinforcement learning may fundamentally outperform model-free approaches in representational efficiency.
Bradly Alicea (@balicea1), Morgan Hough (@mhough), Amanda Nelson, and Jesse Parent (@JesParent) revisit fundamental cybernetic principles of regulation, adaptation, and world modeling across a wide assortment of complex adaptive systems.
Francesco Sacco (@FrancescoSacco1), Dalton Sakthivadivel (@DaltonSakthi), and Michael Levin explore topological constraints on self-organization and suggest that biological systems maintain long-range coherence in ways that are fundamentally different from current transformer architectures.
Georg Northoff (@NorthoffL), Yasir Catal, and Samira Abbasi examine how biological intelligence may depend on capabilities for flexible “inner time” to ensure adaptive alignment between the dynamics of system and world.
Nicolas Rouleau (@DrNRouleau) and Michael Levin explore whether theories of consciousness generalize beyond brains to unconventional embodiments and living systems more broadly.
Benjamin Lyons and Michael Levin investigate economies and collective intelligence as systems coordinated by “cognitive glues” in the form of shared models of scarcity and value.
Katherine Collins (@katie_m_collins), Umang Bhatt (@umangsbhatt), and Ilia Sucholutsky (@sucholutsky) consider “Rogers’ paradox” to demonstrate ways in which collective learning is impacted by different kinds of human-AI interactions.
Ruairidh Battleday (@RMBattleday) and Samuel Gershman (@gershbrain) distinguish between the “easy” and “hard” problems of science, and describe how while current AI systems demonstrate powerful narrow forms of optimization with respect to well-defined inference-spaces, further developments are needed for achieving capabilities for novel scientific discovery.
Fritz Breithaupt (@FritzBreithaupt) explores narrative world models and the roles of uncertainty and transformative experiences in natural intelligences, suggesting that coherent agency may depend on better understanding human-like meaning-making.
Taken together, these diverse perspectives suggest that while LLMs can clearly learn powerful generative models of language, they likely do so without having world models of sufficient spatiotemporal and causal coherence to achieve human-like reasoning abilities, capacities for generating subjective conscious experiences, or pathways to realizing artificial general superintelligence. However, by further developing world modeling architectures, we may eventually be able to create forms of intelligence that recapitulate the remarkable flexibility and generality of human intelligence. Finally, enhanced (e.g. more coherent/integrated) world models may not only afford expanded capabilities, but could potentially help ensure that increasingly powerful AI systems achieve both inner and outer alignment with human(e) values.
I find Georg @NorthoffL's perspective(s) to be extremely compelling, and we explored a few points of intersection with IWMT (xitter.cf/adamsafron/status/2106…) in this semi-recent paper:
royalsocietypublishing.org/r…
Also extremely interested in @MillerLabMIT's theories of consciousness, and in particular his emphasis on traveling waves. His suggestions regarding analog computing seem consonant with this work on "complex" (or (quasi-)quantum) harmonics by Deco et al., as well as work from @wellingmax's group on "flow equivariant" networks for modeling continuous phenomena:
journals.aps.org/pre/abstrac…
arxiv.org/abs/2409.13669
Also compelled by the idea that ephaphtic coupling may be important, although recent evidence seems to suggest it is not significantly implicated with cortical traveling waves in general:
cell.com/neuron/abstract/S08…
Although perhaps electromagnetic fields evoked by population activity could be important for some parts of the brain if my speculations are correct regarding the importance of concentrated message passing in geometrically-structured cortical hubs as a source of workspace dynamics and phenomenal binding of body-world estimates (i.e., it seems that electromagnetic fields may not be able to effectively enslave neural dynamics if they involve spatially distributed functional networks).
@adamsafron Are you familiar with Northoff (TTC) and Earl K. Millers (Analog brain waves) theories? This experiment seems to support pieces of your construction as well as theirs.
"(i.e., it seems that electromagnetic fields may not be able to effectively enslave neural dynamics if they involve spatially distributed functional networks)."
Correction: "spatially distributed functional networks" should read "highly spatially distributed functional networks (of effective connectivity)"
Although something that's unclear to me is what constitutes too "highly spatially distributed" for electromagnetic fields to be ineffective at constraining the synchronization manifolds of the neuronal activity from which they emerge with circular causality?
Looking forward to reading this.
Theta rhythms may support social interactions.
(We are finding that theta provides a broader temporal scaffold, organizing local processing at higher frequencies, gamma vs alpha/beta).
nature.com/articles/s41467-0…
Very well said.
And that's precisely what I've been doing for the better part of a decade:
loc.closertotruth.com/theory…
Slide 8 of this presentation has an attempted fit-on-a-t-shirt encapsulation of the (synthetic) theory:
1drv.ms/p/c/bfa212aa225a3f04…
Sam Harris' confidently asserted category error 'argument' regarding free will makes no sense. It is literally nonsense in terms of being "not even wrong."
Causation doesn't cross ontological levels, and for causal analyses that do involve level-crossing, macro-level coarse grainings can have greater predictive power and channel capacity for communicating information about a system.
@yudapearl
@erikphoel
What if heat didn't simply spread out, but travelled in focused beams like light?
Researchers have now observed exactly that at room temperature.
Inside boron arsenide, long-lived phonons carry heat along preferred crystal directions, producing a ray-like patterns instead of ordinary circular diffusion.
Looking forward to reading this.
Can we build world models purely in the latent space, without pixel prediction?
We present Contrastive World Models - we train the latent states to maximize mutual information with future observations, without any decoders / pixel reconstruction.
Contrastive World Models enables
⚡ Substantially more robust representations
🤖 More efficient training by removing the pixel decoder entirely
🌍 A general approach with minimal assumptions [1/n]
😍
Researchers have directly observed Kelvin-wave turbulence along a single macroscopic vortex filament.
This breakthrough validates a long-standing prediction of wave-turbulence theory, bridging a crucial gap in our understanding of how energy dissipates along vortices in systems ranging from quantum superfluids to atmospheric tornados.
**Experimental Evidence of Kelvin-Wave Turbulence along a Vortex Core", by Barckicke et al., Physical Review Letters (2026)
Wish we were focusing more of our preciously limited conversational/attentional(/emotional?) bandwidth on these kinds of more reasonable/grounded concerns (and hopes) 👇
i put no special weight on the billion bit all these stuff like enzymes is bullshit, productivity has been modest, and cybercrime is ramping up fast and could easily mess up infrastructure; meanwhile we have already a close call on an accidental war from an llm hallucination,
we need to up our game if we are going to get the benefits without major costs
Why 1 billion? Why not 2 billion or 3 billion lives?
I find almost all of these numerical estimates (especially with respect to probabilities of events) from our thought leaders to be both highly questionable in their assumptions and intellectually/socially dishonest/irresponsible.
I wouldn't be surprised if marginal harm of that magnitude was possible with these technologies. I also wouldn't be surprised if advancing the frontier could lead to benefits to humanity of similar (if not greater) magnitudes.
i am not a doomer, but Bill Gates is unfortunately right about this.
the greed of a small number of people could in fact lead to the death of a billion.
ft.com/content/1ade33b1-b5eb…
Looking forward to reading this.
What is the nature of universal representations in AI models, and what determines whether they emerge?
Our paper accepted at #neurips2026 led by @florianmahner & @andropar addressed these questions comparing 162 vision models, with intriguing results.
arxiv.org/abs/2605.13675
👇
Looking forward to reading this.
JEPA-style world models are biased towards learning slow, simple features.
We're releasing MotionJEPA, built for learning a balanced representation of both static and dynamic features.
mkarmann.github.io/motion-je…
How accurate would it be to say that the present situation with AI is one in which we have jagged landscape/frontier superintelligences with impressive-but-not-human-level generality, but with unimpressive capacities for generalization in terms of not being capable of continual/lifelong learning?
@ShaneLegg
@bengoertzel
@SchmidhuberAI
@RichardSSutton
@MelMitchell1
@GaryMarcus
@dwarkesh_sp
@fchollet
Very much agree with most of this.
To what extent can LLMs reliably pass the test originally envisioned by Turing, in terms of an imitation game involving attempted deception with respect to a complex social category such as gender?
en.wikipedia.org/wiki/Turing…
All fields? Hardly. As I have often pointed out, the only AI that works well today is AI in the virtual world behind the screen, for example for automatically summarizing documents, playing chess, proving theorems, creating images, programs and PowerPoint slides. But there is no AI-controlled robot that can do what a plumber can do, or what a capuchin monkey can do, because the physical world is much more demanding than the world behind the screen. Passing the Turing Test is much easier than True AI or Real AI in the physical world. Hence the Turing Test is not a good way of measuring intelligence. To achieve True AI, AI software research per se is not enough, it has to be combined with the physical world of machines and robots. No AGI without mastery of the real world!
Taken from Sec. 20 of the "Annotated History of Modern AI and Deep Learning" xitter.cf/SchmidhuberAI/status/1…
Looking forward to listening to this.
A new podcast in the Early Childhood Matters series - my new thoughts about grandparents and kids with shout outs to Kristen Hawkes grandmother hypothesis and Michael Gurven on elders as teachers.
earlychildhoodmatters.online…
Looking forward to reading this.
Can diversity stabilize network dynamics?
Our Science paper, “Disorder-promoted stability,” is now available on arXiv:
arxiv.org/abs/2609.25226
Interactive animation:
montanariarthur.com/animatio…
Many thanks to Arthur Montanari and Pietro Zanin for invaluable contributions.
Looking forward to reading this.
Dr. David Yaden (@ExistWell) was interviewed for the @RollingStone article on the rise of psychedelic spirituality in America, weighing in on what the research actually shows about belief change after psychedelic experiences. rollingstone.com/culture/cul…
