This Breakthrough Could Change the Path to AGI

This new AI breakthrough could actually change the path we take towards AGI. Let's talk about it. You see, something happened that is pretty crazy because it could be one of the most important things to happen in the entire field ever. And that's not an exaggeration. A company called Eon Systems just took a real flies brain, an actual biological brain from a real fruitfly, scanned every single cell, every single connection between those brain cells, copied all of that into a computer, put that digital brain inside a virtual fly, and the fly just started walking. Nobody taught it how to walk. Nobody programmed it to walk. Nobody trained it on walking data. It just woke up and it knew exactly what to do. So, let's actually get into what happened here and what precedent this sets for the future because it's absolutely insane. So, here's the thing in science called a conneto and it's a big fancy word and it basically just means that it's a map of a brain. Every brain cell scientists call them neurons and every connection between them. Think of this as like a blueprint for someone's entire brain drawn out wire by wire. Now, in 2024, a team of scientists finished mapping the complete brain of an adult fruitfly. This was published in Nature, which is basically the Super Bowl of science journals. The fruitfly brain has around 125,000 neurons and 50 million connections between them. Now, 125,000 neurons might not sound like a lot when you compare it to the human brain, which has about 86 billion neurons. But mapping every single neuron and every single connection between them, that is an insane amount of work. We're talking about electron microscopes scanning tiny slices of brain tissue, then using computers to stitch it all together. This thing took years. It was not simple. So, this team at Aon Systems, co-founded by a guy called Dr. Alex Whistner, who's a physicist from Harvard and MIT, they took that brain map that was discovered and they did something that nobody had ever done before. They didn't just look at the map and study it. They actually ran it. They took that entire brain blueprint and put it into a computer simulation, connected it to a virtual fly body using a system called Neurommechfly and a physics engine called Mujo Co, which is basically the same kind of software used to simulate robots. And then they let it rip. And here's the wild part out of all of this, okay? The virtual fly started doing real fly stuff. Walking, grooming, even egg laying behavior, all from nothing but the brain's own wiring. No instructions, no training, just a brain doing what brains do. The senior scientist on this, a guy named Philip Shu, had already know had already shown in 2024 that this brain model could predict fly behavior with 95% accuracy. But that was just numbers on a screen. Now, for the first time, that brain was actually connected to a body and making it move. Now, you might be asking yourselves, well, we've already seen this before with animals and computers. What is the big deal here? And that is a fair question, but this is completely different from anything that's come before. And I'm going to explain to you exactly why that is the case and why this is a much bigger deal than most people even realize. You see, the current AI that we use on a day-to-day basis, Chat, GBT, Gemini, Claude, all of those AI systems, they work by learning from data. We feed them billions of words from the internet and then they learn patterns. They learned that after the cat sat on thee, the next word is probably going to be the mat. And that's a massive oversimplification, but you get the idea. They learn by looking at vast amounts of data and examples. You can see here you've got the data set sources, pattern recognition, and then next word prediction. Now, when companies like Google, DeepMind make virtual animals that can walk in simulations, they use something called reinforcement learning. That's basically where you let a virtual creature try and fail millions of times and you reward it when something goes right and then after enough tries, it figures out how to walk. But it's not walking the real way a real animal walks. It's walking the way that a computer figured out it's the most efficient way to walk. Sometimes it looks really weird and alien. But what Eon did is the complete opposite of all of that. They didn't train anything. They didn't give data to learn from. They didn't let it try and fail a million times. Whaton did is the complete opposite of that. They literally just copied the brain. And that is the key difference here. They just looked at exactly how a real fly's brain was wired, which brain cell connects to which other brain cell and how strong those connections are. And then they recreated that wiring inside a computer. And then they gave that a body and said go. And you can see here it went off just like an exact fly actually would. Now, the wild part is that this is the difference between teaching a robot dog to act like a dog versus actually copying a dog's brain into a robot. One is imitation. The other is the real thing just running on different hardware. Well, to put this another way, AI is like building a plane by studying

birds. But what did putting a bird's brain inside a plane and letting it fly itself. So, let me explain to you guys how this actually works in a simplified manner. So step one, what you would have to do is you'd have to take the real fruitly's brain. You'd slice it into incredibly thin layers. We are talking thinner than a human hair and you scan each layer with an electron microscope. And this gives you an unbelievably detailed 3D map of every single brain cell. Then you have step two. This is where you use machine learning to figure out what kind of brain cell each one is. Some brain cells excite other brain cells, making them fire. Others inhibit them, making them calm down. You need to know which is which. The team figured out this with about 91% accuracy using four things. The connection map, the strength of each connection, which cells are excitatory versus inhibitory, and a simple model of how brain cells fire. Then step three is you build the virtual fly body. This is a physics simulation. The virtual fly that has legs that obey gravity, joints that bend, and a body with realistic weight. It is basically a very accurate digital puppet but with nobody pulling the strings. And then of course step four is where you connect the brain to the body. Sensory signals go into the brain like your left leg just touched the ground and the brain processes that signal through all 125,000 neurons and 50 million connections. Motor signals come out like move your right leg forward. The body executes the command and then the cycle repeats. That's the entire loop. Sense, act, think, sense, think, act over and over, thousands of times per second, just like a real brain in a real body. And the stunning thing is that nobody designed what the brain should do. The behaviors just naturally emerge from the brain's own wiring. The same way your brain doesn't need a manual to tell your heart to beat or your lungs to breathe. Now, of course, I'm not going to sit here and pretend this is some perfect discovery, because it isn't. Everything has limitations and there you know are some smart people that are raising real concerns. The first criticism is about what's missing from the simulation. So one thing is that you know that most people don't understand is that a real brain doesn't just run on electrical signals between neurons. It also uses chemicals and these chemicals change how the brain works in real time. They're responsible for things like mood, motivation, learning, pleasure and pain. And the Eon simulation doesn't include any of that just yet. It's just copying. It's basically like copying someone's computer files, but not their operating system settings. The real information and stuff is going to happen when you actually have the neurochemicals and neuro signals. Right now, it's pretty stateless. Now, the second criticism here is about the gile cells. Now, about half of the cells in your brain aren't actually neurons. They support cells called gile cells. They help neurons work properly. They clean up waste and they help form memories. And the simulation actually ignores those completely. I mean, when you think about it, having a brain simulation that ignores 50%. That is pretty significant. And then, of course, comes the most significant. Remember, this is a snapshot of a dead brain. The fly that was scanned is obviously no longer alive. So, this is a copy of how the brain was wired at one specific moment. It can't learn new things. It can't form new memories. It can't adapt. It is frozen in a time. And fourth, right here, some scientists are pointing out that the 91% accuracy sounds impressive. But what does that actually mean? 91% compared to what? How was it measured? The team validated their results using real cameras and tracking flies, a technique called optogenetics, where you use light to activate specific brain cells, and then you see what happens. But still, some researchers want to see more rigorous testing. And one of the craziest things of all is that the actual co-founder of Von and the guy who announced all of this, you can see that he's got a financial interest in the company. This guy isn't some random person on the interest. He's not just a neutral natural observer. He's actually selling us something. And that doesn't mean that the science is wrong, but it does mean that you have to think critically about the situation. If someone that has a financial incentive is telling you a story, they're incentivized to make it sound better than it actually is. But that said, even then, the skeptics are still admitting that this is genuinely impressive. The fly is doing real recognizable fly behaviors. It's not random garbage. It's not glitching out. It actually looks like a fly being a fly. And later on in the video, I'll talk to you guys about how potentially this could lead to, of course, super intelligence simulating human brains and where all of that is going to go. Now, if you're wondering how we actually got to this point, you have to understand what came before all of this. The idea of brain emulation. And you can see on screen from this squiggly line which is not squiggly line. This was the first attempt of something and it was called open worm. Scientists essentially mapped the brain of a tiny worm called sea elegance. And

this worm only had 302 neurons. 302 is absolutely tiny. And even with just 302 neurons, it took years to map all of them and create a simulation. And honestly, the results were pretty basic. The simulated worm could kind of wiggle, but it was never that great. Then in 2024, Flyw Wire posted the completed project for the full connections of a fruitfly. This was a massive international effort published in nature. 125,000 neurons, 50 million connections. This was the first time we had a complete brain map of anything this complex. You can see all of the different, you know, parts of the brain here, the sensory, the other sections, visual projection. It honestly is very impressive how complex some of these tiny organisms are. I honestly didn't realize just how much engineering may go into these tiny creatures, but it truly is interesting when you look at it at this level just how much there is. Then of course there were other teams doing cool stuff with virtual flies. You had Google Deep Mind building a simulated fly in Mujo Co that could walk and even do aerial maneuvers. But this was a key difference. Deep Mind's fly was trained using reinforcement learning. It wasn't a copy of a real brain. It was an AI that learned to move like a fly through trial and error. What EON actually did was take the biological data, the real conneto, and move it to a new body. And that's never been done at this scale. So, we went from 302 neurons in a worm to 125,000 in applied. That's a 400x jump. And what they're saying is the next jump is from 70 million neurons into a mouse brain. That's 560 times neuron scaling jump. That is pretty insane. That's a 560 time jump from a fly. Now, that is massive, but they say it's a scaling challenge and it's not a fundamental science problem. The basic approach should be the same. It just needs to be bigger. And remember guys, what do you think is going to happen after they go from fly to mouse to human and they're able to actually simulate a human brain? Can you not start to see where this goes? and of course the implications for Agi super intelligence because the obvious question is what happens when you can do this with a human. Now first computer power keeps growing. The amount of computing power available has been doubling every couple of years for decades. We're now building chips specifically designed for this kind of neural stimulation. Second brain scanning technology is currently getting better fast. On says they're using something called expansion microscopy where you literally make brain tissue swell up so you can see the fine details more easily combined with calcium and voltage imaging to watch live brain activity in real time. And third, and this is the philosophical bombshell, if a fly's brain can produce realistic movement just from its wiring, does that mean that consciousness is just wiring? Think about that for a second. If you could scan a human brain at the same level of detail, copy every neuron and every connection and run it in a simulation, would that simulation be you? Would it be conscious? Would it think it's you? Would it have your memories, your personality, your sense of self? Ion's stated mission is to build the world's largest conneto and the most accurate brain emulation ever created. Think about it. That's what these current AI companies are all doing. They're all trying to build AGI, which is, I guess you could say, human level intelligence. But this company is taking a fundamentally different approach. They want to do the mouse next and then eventually they want to try a human. And as one commenter on Reddit put it, we stole the weights of an adult fruitfly and are successfully running inference on it. In AI language that's very provocative way to describe what's happened. I mean this is where things get really wild because the obvious question is what happens next. So look, I want to be pretty clear about something. We are a pretty long way away from being close to uploading a human brain. The gap between a fruitly and a person is enormous. The missing biology, the chemicals, the support cells, the learning, adaptation, everything matters. But remember, the reason this story matters is because it proves the concept. It proves that you can copy your brain accurately enough. Like if you do it accurately enough, you actually get real behavior out the other side. And that's not a theory anymore. That's a demonstrated fact for at least a fly. Remember, a fruitfly has 125,000 neurons. A mouse has 70 million neurons. And a human has 86 billion neurons. Okay? I don't even want to do the math on just how much more of a jump that is, but if you did the math, you'd understand that we are a long way away. Now, think about this, okay? This could be a completely different path to something that could be even bigger. This isn't AI. This is copied intelligence. Think about it. If right about the scaling, then we might look back at this little fruitfly the same way we look back at the Wright brother's first flight. It was a small and it was short but it did change the path. It changed everything. So for me I think this could be you know an interesting step towards AGI albeit a slow one. I still think that this has entered a new frontier and may actually get increased funding from this kind of research. And I have to ask you guys the question.