Ankit

this is not AI: scientists recently simulated an entire fly brain more than 125,000 neurons and 50m connections then they connected it to a virtual fly body the result is one of the first demonstrations of a biological brain model producing real behaviors inside a simulation. again: this is not AI! here's what you'll learn today: → the idea behind whole-brain emulation → the first digital fly brain model → giving the brain a body → why this is different from previous simulations → what behaviors the system produces → what the researchers want to build next ▫️ the idea behind whole-brain emulation for years, the idea of the technological singularity focused almost entirely on artificial intelligence. the concept was simple. machines becoming smarter than humans. but there has always been a parallel idea in neuroscience. instead of building intelligence from scratch, scientists try to copy a biological brain exactly as it exists. this concept is called whole-brain emulation. the approach works like this: • map every neuron • map every connection between neurons • recreate that structure inside a computer if the structure is correct, the brain should theoretically behave like the original... just wow. until recently, this idea stayed mostly theoretical. but a new experiment suggests researchers may have crossed an early threshold. ▫️ the first digital fly brain model in 2024, researchers built a computational model of an entire fruit fly brain. the model contained: • more than 125,000 neurons • roughly 50 million synaptic connections the structure of the brain was reconstructed from biological data. scientists mapped the fly's neural wiring and used machine learning to predict which neurotransmitters each neuron used. when researchers tested the model, it produced surprisingly accurate results. the simulated brain could predict motor behavior w/ about 95% accuracy. but there was one major limitation. the brain existed in isolation. it could generate signals, but it had no body to control. so the model was essentially: a brain producing outputs that went nowhere. ▫️ giving the brain a body the new step solved that problem. researchers connected the digital brain to a physics-simulated fly body. the system works as a closed loop: • sensory input enters the brain model • neural signals propagate through the network • motor commands are generated • the simulated body moves movement then produces new sensory input. this feedback loop allows the brain simulation to interact w/ its environment. for the first time, a complete brain model derived from biological wiring is controlling a body in simulation. the result is *NOT* an animation. xitter.cf/i/status/2030217301929… it is the brain's own neural circuits producing behavior. ▫️ why this is different from previous simulations earlier attempts in this field usually followed one of two paths. 1️⃣ first category: brains w/ bodies. scientists simulated neural activity, but the brain had no physical system to control. 2️⃣ second category: bodies w/o real brains. robotics and AI research often used reinforcement learning to train virtual animals to move. these systems learned behaviors through optimization. they did not replicate biological neural wiring. this new experiment combines both elements. a biological brain model derived from real connectome data drives a simulated body and generates multiple behaviors. ▫️ what behaviors the system produces once the sensorimotor loop was connected, the simulated fly could generate several distinct behaviors. sensory signals enter the brain simulation. neural activity spreads across the connectome. motor neurons then trigger actions in the body simulation. the important part is that these behaviors are not scripted. they emerge from the brain's own circuit dynamics. this demonstrates that a fully reconstructed neural network can produce coordinated movement and responses when connected to a body. for neuroscience, this represents a major milestone. it means researchers can start testing how brain wiring leads to real behavior. ▫️ what the researchers want to build next the fly brain is only the starting point. the research team behind the project plans to scale the approach. their next target is a mouse brain. a mouse brain contains roughly 70 million neurons. that is about 560 times larger than a fly brain. to attempt this, scientists are collecting massive datasets. these include: • detailed maps of neural connections • large-scale recordings of brain activity • microscopy techniques capable of mapping entire neural networks the long-term ambition is much larger. eventually researchers hope to reach human-scale brain emulation. that goal remains extremely far away. but the current demonstration suggests something important. a biological brain copied neuron-by-neuron can run inside a computer and drive a body. for the first time, the gap between biological brains and digital simulations is beginning to close... crazy, right?? i thought only AI can disrupt things rn, but quite excited for this advancement as well. original post here: xitter.cf/alexwg/status/20302173…