Google Research Accelerates Neuroscience with Complete Fruit Fly Brain Connectome

Google just mapped every connection in a fruit fly's brain. Not a simulation. The actual wiring of a complete male fruit fly brain, cell to cell, released today by Google Research and HHMI, with AI helping trace the connections. Why a fly? Because you cannot fix what you cannot map. This is how scientists start to learn how a brain turns what it senses into what it does, and one day how damaged neural pathways might be repaired. My take: this matters more than any model launch this week. A faster chatbot saves us minutes. Maps like this could change medicine. If you had to pick, should AI labs spend the next five years on better assistants or on science like this?

The line that gets me is repairing damaged neural pathways one day. It's easy to get numb to AI announcements, but mapping a brain cell by cell in service of that is genuinely moving. This is the kind of AI story worth paying attention to.

This is the kind of AI progress I find genuinely exciting. Helping scientists map something as complex as the brain shows how AI can contribute far beyond productivity tools. The long-term possibilities for neuroscience and healthcare are fascinating.

🚀 Thrilled to share my latest exploration in neuroscience! I've been working with a fruit fly's brain, leveraging a real connectome to learn the cardiac arrest algorithm in a browser tab. 🧠✨ With 5,177 Kenyon cells from FlyWire, I'm watching real synapses at play, operating on the fly's own dopamine rule. The learning loop is just three cell types: Kenyon cells encode the situation, MBONs vote on what to do, and dopaminergic neurons deliver the single bit of feedback that rewires them — reward or punishment, nothing else. Run that loop on its own and it reaches 18.6%, barely above the 12.3% you get from guessing. What it was missing was memory. Give the fly a central complex to hold its own history — 2,875 neurons that nothing ever trains — and the very same learning loop climbs to 47.3%. And I can see exactly where that memory runs out. Chest compressions: 98% correct. IV access: 100%. Epinephrine — the one action that depends on remembering when you last gave it — 24%. It learned what to do, not when to do it. While it mastered those reflexes, it should be clear that a fly could not learn ACLS independently. Yet the efficiency is striking: 5,177 neurons, one bit of feedback per decision, no gradients and no GPU, achieving four times chance, all in a browser tab. 🌍💻 Have a play with it, I would love to hear what you make of it. Check it out here: 🔗 https://lnkd.in/gucbTGKQ Not a clinical tool — it's a neuroscience experiment. #Neuroscience #Connectomics #Drosophila #FlyWire #ComputationalNeuroscience

#2MinuteRead1331 NEUROSCIENCE Mapping communication between distinct brain areas Study published in Neuron outlines a computational method using data-constrained recurrent neural networks (RNNs) to map communication between distinct brain areas. Separating internal local neural activity from incoming signals sent by other brain regions presents a major challenge in neuroscience. This new framework addresses the issue by training a large neural network on multi-area recording data, mapping each artificial unit to a specific physical neuron. Methodology: The full network is partitioned into regional sub-networks. Internal currents driving network activity are measured to infer direction and strength of inter-region communication dynamics. Data Sources: Method tested across multi-species datasets, combining calcium imaging and electrophysiology recordings from zebrafish, mice, non-human primates, and humans. Validation: Framework successfully recovered established pathways, such as stress-related inputs to the zebrafish habenula. New Findings: Analysis revealed previously unknown, asymmetric reward-related interactions between the monkey striatum and cingulate cortex. Framework provides an analytical tool for interrogating across-area neural dynamics in complex population datasets. #Neuroscience #ComputationalNeuroscience #NeuralDynamics #BrainMapping #MachineLearning

What if one day we could map the human brain with the same level of detail that we can now map much smaller brains? 🧠 I recently put together a case study, “Mapping the Human Brain,” exploring how advances in connectomics, AI and computational modelling are bringing us closer to that possibility. The recent complete mapping of a fruit fly’s brain and nerve cord—with ~166,000 neurons and ~125 million connections—offers an interesting glimpse into what happens when biological wiring can be converted into a computational model. The bigger question is not simply “Can we map the human brain?” but “What could we do with that map?” Disease-specific circuit maps could potentially help us understand conditions such as Alzheimer’s, Parkinson’s, epilepsy and stroke in a much more targeted way. But the challenges go far beyond computation: scale, data, individual variability, privacy, consent and the fact that wiring alone cannot explain everything about the brain. I’ve tried to explore this journey—from the fruit fly connectome to the long-term possibility of human brain mapping—and where AI could become the bridge between massive biological datasets and actionable understanding. Which brain disorder do you think could benefit first from AI-driven brain mapping? #ArtificialIntelligence #AI #Neuroscience #BrainMapping #Connectomics #Neurotechnology #HealthcareInnovation #ComputationalBiology #DigitalHealth #FutureOfMedicine #Biotechnology #HealthTech

🧠 Neurotechnology & Brain–Computer Interfaces! A small change to this series: rather than highlighting papers from a single week, we’ll now draw from the past 4 weeks and cycle through each topic roughly once a month, so you don’t miss out on any cool new neuroscience! This month’s selection includes work from researchers at Cornell University, University of Tübingen, The University of North Carolina at Chapel Hill, and Rice University, asking some fascinating questions: - How can we study the brain during natural behaviour without tethering animals to lab equipment? - Can a brain-stimulation system learn when an individual brain is most responsive to stimulation? - Can a non-invasive brain–computer interface give users access to hundreds of different commands? - Could fMRI become almost silent — and make it easier to image the brain during natural behaviour? - Does the brain learn how to respond to an artificial electrical input? Each slide below includes a short NeuroBriefer summary covering the main question, approach, findings, and why they matter. Links to the five papers: A wireless modular platform for neuro-behavioral recording and closed-loop manipulation in small animals: https://lnkd.in/gY4n2N-u A first realization of reinforcement learning-based closed-loop EEG-TMS: https://lnkd.in/gianZnKX A high-rate 480-target hybrid BCI system based on SSVEP and sEMG: https://lnkd.in/gvQtRs4f SORDINO: silent, sensitive, specific and artifact-resistant fMRI in awake, behaving mice: https://lnkd.in/gB39wAMS Learning induces activation-mechanism-dependent neural plasticity in an intracortical microstimulation task: https://lnkd.in/gp3hDJyW #Neurotechnology #BrainComputerInterface #Neuroscience #NeuroBriefer

🧐 The specialist-neuron view of the cortex has a data problem. ⌛ For decades, neuroscience has operated on the assumption that each neuron is a specialist, devoted to a single function, one component in the brain's division of labour. 📜 But a new Nature study, analysing more than 14,000 neurons across 43 cortical areas of the mouse brain, found the opposite: Categorical structure only re-emerges when you zoom out to the whole cortex. ❓ Why it matters: if functional information lives at the population and circuit level rather than in specialised cells, the hunt for "the neuron that does X" may be chasing a statistical artefact, with real implications for how we interpret cell-type studies. 📕 Read the full story: https://lnkd.in/d8XbNukH

Brain-computer interface company Precision Neuroscience closed an oversubscribed $250M Series D co-led by Pershing Square and the Ackman Oxman Institute, bringing total funding to $430M. https://lnkd.in/gVjtRPWN

Organic computing—sometimes referred to as “wetware”—is fascinating. In my humble opinion, biological computing could become a serious contender for at least some parts of the future AI infrastructure landscape. The potential to perform complex computation while using a fraction of the power required by conventional hardware is especially interesting as AI data centers continue to face enormous energy demands. What intrigues me most, though, is the possibility that biological systems may process information and adapt in ways that more closely resemble human cognition than today’s traditional AI architectures. It’s still an emerging field with plenty of unanswered questions, but the intersection of biology, computing, and artificial intelligence may end up being one of the most important areas to watch over the next decade. #ArtificialIntelligence #AI #Wetware #Biocomputing #DataCenters #EmergingTechnology