Wordpress site is up - thoughtforms.life/. Register to be notified of book progress, specific events, news, and new posts with photography, essays, interviews, & more. Unlike at drmichaellevin.org, here I will post ideas not fully baked yet, & academic-adjacent content.

New preprint: "On Growth and Form, and Function" with @miltonllera, @MarcelloBarylli, @risi1979, @drmichaellevin Inspired by D'Arcy Thompson's grid transformations, we find that "reusable regulatory handles control phenotypic variation" in NCAs. 🧵 arxiv.org/abs/2609.29755

New #preprint - @HananelHazan "Computational Framework for Identifying Ion Channel Mutation-Compensating Interventions" biorxiv.org/content/10.64898… How to fix hardware problems (an ion channel mutation), in software (without fixing the mutation)? We took a computational approach to this question in neural tissues - a system that helps discover what other channels could be opened/closed via pharmacological agents (electroceutical drugs) to compensate for the mutation's effects on neural firing patterns. Abstract: "We present an automated pipeline for finding therapeutic interventions in channelopathies. It starts from patch clamp recordings of the variant channel, searches over the pharmacologically accessible conductances, and says which currents must change and by how much. The intervention never touches the mutated channel: it compensates by modulating others. The feasibility of this approach is demonstrated through two computational strategies, each addressing a specific capability gap. First, we employ high-fidelity NEURON simulations combined with exploratory search algorithms to model two clinically described variants, and identify "stability windows" in which channel modulations restore the reference firing pattern. The compensating configurations are degenerate: 37 conductance triples fire the same number of action potentials at every one of the 35 injected-current levels, and the 10,780 triples reaching the best intervention value produce only six distinct firing patterns. Second, to address the computational cost of such detailed modeling, we implement a differentiable Hodgkin-Huxley model in PyTorch. Applied here to theoretical mutation screening, this approach trades granular detail for speed, treating the search as a continuous optimization problem that makes very large, non-local searches of the conductance parameter space affordable: the differentiable forward model evaluates 34,000-64,000 candidate parameter sets per second on a single consumer graphics card, against 11.9-16.9 per second for the non-differentiable pipeline on a 72-core node. The gain arises from forward-model throughput feeding a search rather than from the gradient itself, and it is what makes the 32,000,000-point survey of the solution topology reported here affordable. These computational predictions can inform practical drug development and high-throughput screening by naming the currents that must change and by how much."

Meet The Scientist Redefining Intelligence With Michael Levin "There's a FALSE assumption that Intelligence and cognition is rare in the tree of life and late in the evolutionary story. This is very wrong". "Many of the characteristics, morphology, behaviour, phenotypes of organisms are not determined by their genes." "Learning only needs a dynamical system made out of interactions that give way under stress. No genes necessary" The remarkable @RichardWatson90, a coauthor on many papers with @drmichaellevin is live now! ⬇️

Lots of interesting stuff here!

Out today !!!!

oh and if anyone wants to see the main steps leading up to this paper, a (needs to be updated) map is here: drmichaellevin.org/research/…

Can self-interested, self-improving, self-replicating agents learn to cooperate? Our new paper, Tapes Together Strong, shows they can: when social behavior, computation, and reproduction share one energy budget, cooperation evolves from scratch. arxiv.org/abs/2609.10817 🧵

Credit to @dz_gong for putting together this great resource. The GEVI jungle can be hard to navigate!

Btw, if anyone is interested in the background to all of this, here is where my stuff lives: drmichaellevin.org/ - official lab website with all the preprints, papers, software, resources, talks, interviews, etc. Browse around - there are a lot of deep pages. piped.video/channel/UC3pVafx… - our Center youtube channel, with not only my talks but also amazing people I've invited to give presentations and participate in discussions. No fancy editing or optimization for views, just ideas (I appreciate the offers to replace my 1999 webcam; sorry - it's got all the pixels I plan to show). mlevin77.substack.com/ - shorter, less rigorous things - haven't figured out exactly what will go here but I've got some AI-relevant things planned. thoughtforms.life/ - more personal, public-level musings, 1 click beyond "what we can say strongly" (which is what my lab website sticks with). Sign up there to be notified of new content, book updates, etc. Also nature photos.

We wrote a tutorial guide on voltage imaging, now out in Neuron: authors.elsevier.com/a/1nlgh…. Indicators, microscopes, and software have all advanced to the point that it's not so hard -- you can now directly watch the electrical activity of neurons in live animals or in culture.

🧵 New paper: RL for discovering and controlling self-organizing phenomena. We introduce CARL, a closed-loop agent that learns to create self-organizing patterns in Lenia, steer their behavior, and let humans guide them in real time. Project website + interactive demos below ↓

New #preprint in computational modeling of #bioelectricity! @cellbioelectric and Joel Grodstein: "Membrane voltage and connexin expression work together to enhance tumor growth and metastasis in cancer" biorxiv.org/content/10.64898… Abstract: "There is strong evidence of tumors manipulating their resting membrane potential (Vmem). While most fully-differentiated cells have a Vmem of roughly -70mV, tumor cells are generally depolarized, with Vmem ≈-30mV, which more closely resembles the Vmem of stem cells. This is often believed to serve the purpose of accelerating the cell cycle and hence advantaging tumor proliferation. But when the tumor becomes invasive, its cells sometimes revert to a hyperpolarized Vmem with no obvious reason why. Separately, it is well accepted that solid tumors that are not yet invasive greatly underexpress connexins relative to healthy tissue; connexins, for our purpose, form gap junctions (GJs), small connecting tubes between nearby cells. Tumors that are invasive, by contrast, overexpress connexins. There is very little explanation for the paradox that connexins are first underexpressed and then overexpressed. However, it has long been known that Vmem electrically gates GJs; specifically, that homotypic GJs conduct best when the two cells they connect have a similar Vmem. Our in-silico model results explain this phenomenon, showing that when considered together, tumors' electrical and connexin-expression behaviors form a unified and effective strategy to control communication between the tumor and its healthy neighbor cells. This has implications for the emerging field of cancer bioelectrics, potentially leading to more precisely-targeted therapies."

New #preprint! Part of our efforts to develop tools to communicate with various unconventional agents, and help them communicate with each other. Which they're already pretty good at! biorxiv.org/content/10.64898… Led by @DrElenaSergeeva. @PigozziFederico, @thosvarley, Colin J. Comerci, Zhe Zhao, Robert Brucker Gürol M. Süel, @DoctorJosh "Living multicellular systems induce decodable spatial patterns in bacterial collectives"

How to train your cells: check out our new platform! In @drmichaellevin's lab, we aim to extend the growing field of single-cell learning to non-neural human cells, enabling scientists to study training techniques as a new approach to medicine and bioengineering