Reading Human Bio-Signals with AI in Real-Time

bunny.com/jazz. Currently, we have a crazy mode button that sends everything to 100% 100% uh random and we have a more real uh sound generator that's taking your brain waves turning in turning them into musical notes. The mapping of the frequency amplitude to notes is is real scientific based. So it's essentially taking um the different bands theta delta alpha beta depending on the amplitude in each it will make a decision to what note to use depends on depending on amplitude and the frequency. Yes, we have power. All the notes played at the same volume. Not sure [music] obviously going to be a um a play button to this. Let us know or if you want me to fire up any any of the tools. Do let me know as well. and be happy to work on the tool of your choice, but you'll have to specify which one you checked, what what worked, what didn't. Bio was referenced in this uh article supposedly legit. I'm not actually sure. They literally took uh print screens from uh the tool that we made. Did you put a reference still working? It is working. Yes. So literally in the paper I used the print screens uh compression and they compelled compared the HA wavelet and SPI HD uh which I don't have. So where did we So they did the PSN which I do measure um compression ratio which I do not have. I'm not sure when they took the well what's PS and uh no PSN is what we measuring. It's the signal to noise ratio. Uh, no. S P I HD hierarchical trees algorithm builds on DW. It's a mediocre paper. We're not going to be going into it. It says a research article, but there's no data being collected or anything. There's no method. Well, thanks for the reference, but uh yeah, I can't [clears throat] there's nothing. There's no data collected or anything. Yeah. But it's based on uh this tool that we had on the side for quite some time. Yeah. They use this image in the paper. I probably should have referenced the Yeah, the images are from radiopedia. They didn't reference radiopedia. It's like this one was like recursive references things. Anyhow, yeah, thank you for the for the reference, but uh yeah, the paper is not great. So, but yeah, go check out the original tool. I might do an AI review on it as well. It's stage. Yeah, I might be doing the EG2 music stuff when I improve the play button. I was want to make sure when we're selecting the notes, the musical notes, uh, which is working at the moment, the selection is actually I'm pretty sure we specified it to begin with, just when you're listening to it, [music] doesn't sound like the amplitude for the four different notes is uh, different. Doesn't sound like it's different for the four different notes. [music] Shall we do the quiz first? We have two quizzes. One is hard, one is medium. Okay, let's see if I can answer a question about the tool that uh I developed. Well, considering I developed it with AI, maybe I would not be able to answer it correctly. The question is, how does the real-time signal amplification microscope application extract a pulse signal from a standard webcam video? Cuz that's what it does. Does it fairly successfully? We'll do a quick test cuz we like testing stuff. Yeah, you could see. Well, assuming this uh variation is not coming from anywhere else, it's safe to assume it's actually coming from from my heart rate. There is a chance that's coming through that um color reflecting off my head or something reflecting off my skin and giving the measurement. But I can cover it up like this still. See if the measurement still there. Uh yes, it is. So yeah, it's working. And the question was, how does the real-time signal amplification microscope application extracts a pulse signal, a heart rate pulse signal from uh the standard webcam video? The options are detecting infrared heat signatures emitted from the facial surface. Oh, that's a no. It's no infrared. Measuring the electrical resistance of the skin via camera lens. Obviously no. Analyzing the acoustic vibration captures by microphone during speech. Nope. Using Urian video magnification to visualize subtle skin tone changes. That one is correct. This technique amplifies tiny color variations in the skin cause caused by blood flow, allowing the software to detect a heart rate without physical contact. How good is that? Uh there's a question about the smartphone neck the simulation. When simulating the smartphone neck phen phenomenon, what happens to the load on the cervical spine as the head tilts forward? The mechanical load on the neck increases significantly. Yes, the load decreases as the weight is transferred to the upper back and no the load remains constant. No, the load is neutralized by the tension in the trapezius muscles. Well, no. A the mechanical load on the neck increases significantly. That is correct. So as the head moves away from the center of gravity, the effective weight supported by the neck muscles and spine increases due to the lever arm effect doesn't kind of explain what it is, but yeah. Anyway, when you're holding the phone like that puts a lot of strain on your spine. So go check out this simulation by yourself. Everything you do is available on bioc.com. Sport is the common reality for users during initial switch on phase of the but users immediately experience crystal clear speech and music definitely not sounds are often perceived as robotic mechanical and confusing that would be the case the device only transmit low frequency vibration rather than sound there's a different implanted actually a bone called world coccia product called bah Uh-huh. It's a bone conduction thing. So, it's like a screw that is inserted into into the skull. The implant causes total loss of uh special awareness and balance. Uh not normally. So, yes, it's B sounds are often perceived robotic, mechanical, and confusing. Yes, the brain requires time to adapt to the new electrical signals which initially do not sound like natural acoustic hearing. So we have this simulation of uh for transform. I think it's uh referring to this one. You can uh draw your pattern. It will break it down into a four components. So you can see yeah these arrows they're essentially the forer components. So the question is for what mathematical principle is demonstrated in the for simulation when creating complex patterns signals are multiplied by their own derivatives to increase resolution. No, don't think so. Doesn't sound like it. Data is compressed by removing all high frequency components. No. A signal is converted into series of discrete unrelated pulses. That sounds legit. A complex signals can be decomposed into sum of simple sinosoids. Okay, that must be Yeah, it must be that one. It's actually talking about a different simulation that we have on the side. So, it's not the drawing one. But if you type in 4 series, it should get these two simulations. See, it's actually talking about the second one. Yeah, this makes a nice demo v sound as well. Can turn on the sonification. I can actually hear something. Anyway, yeah, the for series allows any periodic shape to be represented by adding together various sign and cosine waves. So, this is what this is showing. >> [music] >> Yeah, there's a quick question about this uh simulation. So, this is a ppg. Uh in this simulation, what is the primary goal when exploring the effects of age, skin tone, and lead intensity? It's the primary goal. Well, to see the difference it makes on the signal to noise ratio, the signal quality to determine the maximum depth of an ultrasound echo. No. To understand how these factors influence the signal to noise ratio, yes. To calculate precise BMI, no. to measure the rate of skin. So no yeah biological and environmental variables significantly impact the quality of the ppg signal which is quantified by the SNR signal to noise ratio. This aen faces principal component analysis simulation thing we have a question about that. Well, first of all, we need to actually run it. Yeah, we'll do some uh faces. So, we need to capture some faces. And then we have this egg and a values there. So, you can see what's being mostly represented is the the bit that is different in this case moving eyebrows. And that uh face at the moment it's uh reconstructed from the from this egg invas values above. So this is not me. This essentially an avatar created out of um these egg and faces not the original not the original video or the original face. And the question was the aen faces tool uses principal component analysis PCA for which specific purpose in facial recognition the options to encrypt facial images so they cannot be viewed by humans. Well I mean it could be used that way but that's not what it's doing in this case to simulate how a face will look as it ages over several decades. That's a actually great idea, but uh again that's not what this version of the tool is doing. To extract and visualize the most significant facial features. Yes. Or to increase the resolution of lowquality webcam image. Maybe could be done, but it's not what it's doing. So it must be C. Yes. PCA reduces the complexity of image data by finding the directions also known as components with the most variance. So that's what these images are the agon values and you can see the bit that I was moving just the eyebrows is highlighted and that would be representing key facial characteristics. What key physiological concept is explored in the Hodkin Hxley simulation? The dynamics of the neuronal membrane and action potentials. It must be it. The gas exchange. No. The flawed blood. No. The mechanical breakdown food. No. Yeah. So, that was easy. Yes, that's right. The Hutchkin Hxley model is the is the foundational mathematical description of how action potentials are initiated and propagated in neurons. So that is correct. Go check out the simulator. It's available on the website for you to explore. that wasn't quite as successful as uh finding your heart rate from camera. We're trying also to measure respiration rate um by you know using similar technique um in this case we're amplifying the color of your skin as the blood is pulsing through it and uh the alpha version was meant to amplify motion to count to measure respiration rate and the question is when tracking respiration using sub pixel Camera monitoring remote BPG which technique is used to identify the breathing rate right and I guess yeah and there's the options measuring the change in pup pupil pupil diameter under bright light I don't know if you could that probably is useful for something not sure if it's for respiration rate it will be used for mental load seen some papers where they measure mental load so essentially how stress rest whatever you are. Another option, counting the number of times a person blinks their eyes. Well, that won't do it, would it? I don't know if that's useful at all. Blink counting. Probably not. Not sure. Applying the fast for transform to the signal maybe. Well, that's what we're doing to show the spectrum. So, that's kind of used everywhere. It's not unique to respiration or heart rate monitoring. Using a heat map to track the temperature of exhaled breath [snorts] when tracking respiration using some pixel camera, which technique is used to identify the breathing the breathing rate must be C, cuz that's C. That's right. FT is used to analyze the frequency components of the sub pixel movements allowing the software to isolate the spec the specific frequency of breathing. Yeah. So go check out this tool. Let me know if it's working on your machine on your setup. But yeah, you need to control essentially three things. One is big one is motion. They have to not move for a while. Yeah. And then you can see the signal is there. That's the color change of your face. 63 bits per minute. My my smartwatch Gin, if you want to sponsor this videos, is showing 63 as well. So that should be within uh plus plus minus one uh bit per minute accuracy. Then you have to control the light in the room. So natural natural light is better. And the other thing the third thing that is uh can be easily forgotten but very important is that you need to avoid something flickering in the room, some light source. Especially if you're looking at the screen, there could be something flickering on the screen and that will be reflecting off your face and messing up the measurement. So, make sure there's nothing flickering in the room. No light sources, other light sources that could mess up the measurement. So, yeah, check it out on the website. >> You know, have you ever wished you could just see how a neuron fires? or get this, watch your own pulse using nothing but your webcam. Well, today we're going to dive into a platform that does just that. It's called Bio Chaos, and it's basically this amazing digital playground where really complex science becomes interactive and honestly just stunning to look at. And right at the heart of it all is this super simple but really powerful mission from its creator. And they're a biomedical engineering PhD, by the way. The whole idea here is that it's not just about looking at dry data. It's about turning that data into something you can actually experience, something you can feel and understand without needing a textbook. Okay. So, what does that actually look like in practice? I mean, it's one thing to say you're making science fun and visual, but it's a whole other challenge when you're talking about super dense topics like neuroscience or signal processing. So, the big question is, how do they actually pull it off? All right, so first up, one of the things this platform does incredibly well is visualizing stuff that's normally, well, completely invisible to us. We're talking about taking these really abstract biological processes and turning them into hands-on interactive experiences you can actually play with. So, check this out. This tool is called a realtime signal amplification microscope. I know it sounds super complicated, but what it does is just wow. It uses your regular computer webcam to pick up on these tiny subtle changes in your skin color, the ones caused by blood flowing through your veins. And the result, you get to see your own pulse visualized live on your screen. It's pretty amazing. So, you might be wondering, how in the world does that work? Well, the tech behind it is called Ularan video magnification. Basically, it's this really clever computational trick that takes a normal video and just cranks up the volume on tiny changes in motion or color that you'd never see otherwise. It's the secret sauce that makes the invisible visible. It's literally the magic that lets you see your own heartbeat. Oh, and this one is brilliant because it tackles such a modern problem. We've all heard about tech neck, right? This simulation shows you the actual physical load on your neck as you tilt your head down to look at your phone. You can literally see the stress building up. It takes that abstract warning from your doctor and makes it something you just get instantly. And what about trying to understand something as complicated as an MRI? I mean, it's basically magic to most of us. Well, this interactive MRI simulation lets you actually play with the physics behind it. You can mess with proton spin and radio frequency pulses to see how those incredible brain images are created. You can even generate your own synthetic brain scans, which is just so cool. All right, so next up, the platform takes that same hands-on approach and applies it to maybe the most complex thing we know of, the human brain. It's all about decoding the wild complexities of neuroscience, one simulation at a time. So, if you've ever dipped your toes into neuroscience, you've probably bumped into the Hodgekin Huxley model. It's the classic explanation for how a neuron fires. But instead of just staring at equations on a page, this tool turns it into an interactive visual. You can actually tweak the parameters and watch in real time how an action potential gets generated and travels down the neuron. And it gets even cooler, I think, with the EEG simulation. Here, you can play around with different brain wave signals, you know, your alpha, beta, theta waves. But what's really smart is that you can also add noise and other junk to the signal. And that's absolutely crucial because it teaches you how to read and make sense of real world EEG data, which is never perfectly clean. But you know, biochaos isn't just about the cool visualizations. It also brings this refreshingly honest and I'd say critical perspective to one of the biggest topics out there right now, artificial intelligence, especially when it comes to medtech. So, the platform has these really helpful guides like this one that help cut through all the hype. It breaks down the difference between, say, deterministic AI, which is the kind that follows strict rules and gives you predictable results, and generative AI, which is what everyone's talking about, the kind that creates new, sometimes unpredictable stuff. It's just a really simple, clear way to understand what's actually going on. And I absolutely love this quote from one of the creators posts. It says, "Great products aren't built on clever code. They're built on clean, reliable, boring data." And that just hits the nail on the head, doesn't it? It cuts through all the hype about brilliant AI models and gets to the truth. The real work, the hard work is in cleaning the data. Because at the end of the day, garbage in, garbage out. It doesn't matter how smart your code is. And that healthy dose of skepticism doesn't just stop at the code. It extends to the whole medtech industry. The creator asks this really sharp question. Is all this new tech we hear about really new, or is it just 30-year-old science with a fancy new app interface slapped on top? It's a great reminder to look past the marketing buzz and find out what's actually innovative. Okay, so after seeing all this, you're probably thinking, "How can I try this out?" And the great thing about Beyond Chaos is that it's built for exactly that. It's designed for you to just dive right in and start exploring on your own. And getting started couldn't be simpler, really. Step one, just pick a topic you're curious about. Could be the heart, the brain, whatever. Step two, fire up a simulation. And the best part is you don't need to know any code at all. Or hey, you can just jump straight into one of the games and start playing. And this is where that whole fun and visual idea really comes to life. I mean, you can play a game called PPG Signal Quest to see how your posture messes with health sensor data. Or you can go headto-head with an AI and cardiobot and test your skills at spotting weird heart rhythms. Or explore this incredible 3D brain map in seizure zone. It's all about learning by doing. So when you step back and look at it, Bionic Chaos is so much more than just a website with some cool tools. It's really a digital laboratory. It's a space where you're encouraged to experiment, to ask questions, and to really build a gut level understanding of how our bodies work. And that kind of brings us to a final thought. If bion chaos proves one thing, it's that pretty much any complex idea can be made easier to understand if you can just see it. So that leaves me with one last question for you. If you could visualize any scientific concept out there, what would you pick? >> Okay, it's actually pretty pretty decent. Pretty good. Some of them actually pretty amazing. Yeah, this this how that simulation meant to look like. Not uh what I have. How do you turn this into a simulation? It's much uh much better visuals. Yeah, this is what I have. Well, this is sorry, this is not what I have. This what one of the agents made for me as an interactive simulation. This is how it meant to actually look like. So we could like add it to the to the prompt essentially saying make this look like that which is much better [sighs] which is much better graphics. We've just print screen. And what else we had? Yeah, this visualization is a bit weird. The tool is much cleaner. Yeah, a lot of the other slides visuals are just nonsense. This is a nice one. Obviously there's a big leap there from the magnet orientation whatever the cell the water molecule orientation to the actual imaging. I'm so [clears throat] not sure what to make of that coding the brain. Yeah, this kind of what we have [snorts] and it's actually broke the neuron. Added this signal while the signal is traveling along the neuron. Yeah, that's messy. Not great. Yeah, that's the uh difference between a deterministic AI and uh generative. There should be like a third option when you combine the two and that's actually what we are trying to do here. That's what we are trying to do. What else we've got? Yeah, those visuals are quite uh quite generic. Choose signal quest board zone sandbox like a development sandbox environment or something. If you could visualize any scientific concept, what would it be? That's a great question. That's a great question. Yeah. So, everything we do is available on bioc.com. Go check it out. And what what else we've got? Yeah, the not is doing pretty well. Uh, you know, generating content. The content is legit. Just literally gave it the website URL and the YouTube channel and the YouTube channel and it's doing a good job. the hard math of decoding human biology. I'm not sure what the visuals will go with that. Okay, let's fire up the interactive mode. See what we get. The visual with just the website. We'll see what it talks about. If it picks up on any tool, we will uh load up the tool. >> Um, you've probably seen the video, right? Like while you're just mindlessly scrolling online. >> Oh, yeah. The music swells up. >> Exactly. The music swells, a baby looks up, and tears just start streaming down the mother's face. It's a It's a Klay implant being switched on for the very first time, >> right? And it's always presented as this beautiful instantaneous thing. >> Yeah. Literally just a flip of a switch from total silence to, you know, perfect understanding. I mean, it feels like pure magic. >> It's an incredibly compelling narrative. We um we desperately want medical technology to work exactly like that. We want it to be a clean binary solution. >> But then you step out of that viral social media feed and into the actual world of well neurodedevelopment and bio signals and just the wild frontier of modern medical tech. >> Yeah. And suddenly that pristine magical switch just it doesn't exist. >> It really doesn't. We're looking at a landscape where human biology crashes head first into artificial intelligence and engineering. >> And the reality is so much messier than those videos suggest. So, if you are listening to this right now, whether you're trying to keep up with the latest tech headlines or you're just insanely curious about how your own body works, we are taking a massive deep dive today into exactly how we decode the human machine. >> And the material guiding our exploration today comes from a really fascinating platform called Bion Chaos. It's uh it's an entire ecosystem really. >> Yeah. A website, a suite of interactive tools, a dedicated channel. >> Exactly. All built by a biomedical engineering PhD. And their entire mission is taking incredibly hard, dense, science-like kind of stuff that usually sits behind strict academic pay walls or is just buried deep in specialized medical journals and making it deeply visual and accessible. >> We're talking about everything from optical illusions to neural implants, but they don't just act as a cheerleader for new gadgets, >> right? That's the crucial part. Our goal today is to cut through the glossy, highly produced hype of modern medical technology and figure out what is actually going on. Okay, let's unpack this starting with that viral medical miracle I mentioned earlier >> because Beyond Chaos has this critical essay that quite frankly violently shatters that specific illusion of the coclear implant switch on. >> Yeah, it really does. >> They point out the massive almost uncomfortable gap between the expectations set by internet aesthetics and the gritty exhausting neurological reality. >> Right? Because that initial switch on experience, it almost never sound like a loving human voice. They actually have this incredible cookware simulator tool on their platform >> which is such a brilliant way to demonstrate it. >> It lets you visually and orally explore the exact electrical signals being sent into the brain by the implant. And when you actually listen to it, it is terrifying. >> It really is. The first sounds are often described as robotic metallic beeps or just overwhelming other confusion >> like an old dialect modem just screeching directly into your cerebral cortex. >> What's fascinating here is the underlying mechanism at play which is neuroplasticity. The implant itself, like the hardware is doing its job perfectly, >> it's just sending the data. >> Exactly. It's picking up sound waves and delivering electrical signals right to the auditory nerve. But the brain, the biological software, has absolutely no idea how to interpret those specific artificial electrical pulses yet. >> It's um I was thinking about this. It's sort of like downloading a massive, highly compressed file on a really old computer. Like you have the data, but your brain's hardware needs time to unzip it and actually make sense of the noise. >> That's a great analogy. It's less like flipping a switch and more like well being dropped in the middle of a dense giant dull machete. >> Oh man. >> The sound of human speech is the destination, but your brain has to physically hack a brand new trail through the underbrush. >> Wow. It literally has to build new neural pathways to map those strange robotic beeps or onto the concept of language before it can understand a single word. And that takes months, right? Sometimes even years of just immense exhausting effort. >> Yes, it's a grueling physical and cognitive process. And the Bionic Chaos sources don't shy away from the complex reality surrounding this either, >> right? They get into the business and ethical sides, too. >> They do. They provide a thorough visual analysis of Kclear Limited looking at the engineering and financial sides but also navigating the heavy ethical controversies surrounding these implants within the deaf community which is a whole other layer to this. >> It is it's a highly nuanced landscape where deafness is often viewed as a rich culture and identity not necessarily a medical defect that needs to be cured by an invasive device. >> Right. And taking that skepticism a step further, Bianicos applies it to how the tech industry markets these devices as a whole. They ask, you know, whether the revolutionary new technology making headlines is actually groundbreaking or if it's just 30-year-old medical tech masquerading as innovation. >> We see this constantly in the medte space. They highlight technologies like deep brain stimulation, uh, DBS systems, or even the newer models of these hearing implants. the fundamental science, the core way we interface with the nervous system in the 1990s. >> So, it's so easy to be blinded by a beautiful smartphone app that connects via Bluetooth and you just think, "Wow, we've revolutionized neurology." >> Yeah. When in reality, we just designed a slicker user interface for decades old hardware, >> which forces us to look at the other end of the spectrum. I mean, if drilling into the skull to lay down raw wire requires the brain to hack through a gene of adaptation, how do we accurately measure what's happening inside the body without physically invading it at all? >> And that leads us to some technology that honestly sounds like absolute science fiction. >> Let's talk about the invisible signals we broadcast every single second because on the platform, there's this interactive tool called the real-time signal amplification microscope and it uses a technique called RPPG or remote photosmography. The term is a total mouthful, but the engineering behind it is just staggering. It relies on a mathematical process called Ularan video magnification. >> Okay, Ularian video magnification. >> Essentially, it takes a standard video feed, literally the basic webcam on the laptop you might be sitting in front of right now, and it amplifies microscopic, completely invisible changes in a person's skin tone. >> Wait, wait. So, my laptop camera can potentially read my vital signs just by looking at me. How is that actually reliable? >> It sounds impossible, right? >> Yeah. I mean, if I'm on a video call and I shift my weight in my chair or, I don't know, a cloud passes over the sun outside my window and the lighting in the room shifts, wouldn't the camera suddenly think my heart stop beating? How could a webcam possibly filter that out? >> That is the exact problem engineers spend their entire careers trying to solve. To explain how it works, we have to look at the biology first. Every time your heart beats, a pulse of blood flows through your vascular system. Right. Right? >> Including the tiny capillary beds just millimeters beneath the surface of your facial skin. When that blood pulses, it causes a minute shift in the color of your skin. It is entirely invisible to the naked human eye. >> But the camera sensor can see it >> precisely. The software isolates that specific frequency of color change over time, computationally magnifies it, and suddenly it's extracting your realtime pulse signal from nothing but light reflecting off your face. Here's where it gets really interesting because the moment you realize a standard optical lens can read your vital signs from across a room, you realize the massive privacy implications. >> Oh, absolutely. >> You are literally broadcasting your internal biology to anyone pointing a camera at you. But beyond the privacy aspect, the bioff sources detail the absolute mathematical nightmare of making this work in the real world. Because to your point, the real world is chaotic. >> It is entirely chaotic. And if we connect this to the bigger picture, this technology represents a profound duality. On one hand, it democratizes health monitoring entirely. You don't need sticky electrodes. You don't need expensive chest straps or hospital visits to track cardiovascular health, >> which is incredible. >> It is. But it introduces an avalanche of environmental variables. A shadow, a slight head movement. It all becomes a massive mathematical hurdle that AI has to filter out. And they highlight how AI tackles this in a video about extracting subpixel respiratory tracking, which is literally watching your chest rise and fall at a microscopic level. And to do this, the algorithm relies on something called a fastforier transform. Hold on. Fastforier transform. You lost me there. What is that actually doing? >> Okay, think of it like taking a blended fruit smoothie and using math to unblend it. >> Unblend a smoothie. >> Okay, if I hand you a smoothie, it's just a complex, messy mixture. You can't separate the ingredients by looking at it, >> right? But a fastforier transform is a mathematical algorithm that can look at that smoothie and say, "Okay, this is composed of exactly 40% strawberries, 50% bananas, and 10% apples." >> Oh wow. >> In the context of the video feed, the smoothie is all the chaotic movement, the flickering desk lamp, you shifting in your chair, your breathing, your pulse. The algorithm unblends that messy visual wave to isolate the exact specific frequency of just your breath or just your heartbeat. >> Filtering out the shadow from the passing cloud. That is wild. And Bionic Chaos actually gandifies this concept to prove how fragile it is. >> Yes. The interactive simulations. >> Yeah. They have these games called Signal Savvy and PPG Signal Quest. You play a game where you manipulate variators like your posture, the angle of your arm, your age, your skin tone to see how it affects the signal to noise ratio or SNR. >> It physically demonstrates that extracting biological data isn't just taking a picture. It's a brutal constant war against environmental noise to find a clean biological signal. >> But that war against noise becomes exponentially more dangerous when we shift our focus. It's um it's one thing when a webcam misreads a shadow and gets your heart rate wrong, >> right? The stakes are relatively low. >> Exactly. It is an entirely different universe of risk when that algorithm is operating autonomously directly inside your body. >> Which brings us to the dark side of connected biology. >> We are talking about the reality of hackers, catastrophic bugs, and putting artificial intelligence inside vital organs. >> The technical term is embedded machine learning and the Bion chaos sources interrogate its vulnerabilities relentlessly like they explore the security risks of closed loop pacemakers. >> Right? These are devices placed inside a human chest running realtime AI to make autonomous decisions about when then to deliver a shock to a failing heart. >> Exactly. >> I want you, the listener, to pause and just think about your smartphone or your laptop for a second. How often does an app inexplicably crash? How often do you get a notification saying your device needs a mandatory software update and while it's updating, you can't use it at all? >> It happens all the time. Now imagine your heart requires a software update to keep beating, but the server is down or the Wi-Fi drops mid download. How does that fundamentally change our relationship with our own biology? >> It shatters the illusion of biological independence when we combine embedded machine learning with vital life sustaining organs. Software bugs are no longer just frustrating computer glitches. >> No, they become direct physiological threats. >> Yes. While AI makes these devices vastly smarter and more adaptable to your specific body, it simultaneously expands what cyber security experts call the attack surface of the human body. Every new line of code, every Bluetooth connection is a potential entry point for failure or malicious interference. They actually have a whole video on this called neural implants and brain computer interfaces. Bandwidth versus bugs. >> And the consequences are far from theoretical. Bio Chaos features a critical analysis they call the forecast trap focusing on seizure prediction technologies and they pair this with a deep dive into a medte startup called Seir Medical. >> Seir medical is a perfect cautionary tale. They were the absolute darlings of the industry. They seemingly revolutionized epilepsy diagnostics with this highly innovative AIdriven wearable technology that could supposedly predict seizures. But then they faced a catastrophic recall. >> They did. The tech just failed in the real world. >> And the fascinating part is that it wasn't a malicious hacker. It wasn't a rogue evil AI. It was something much less glamorous. It was the data. >> The forecast trap happens when an AI model isn't actually learning the biology. It's learning the noise in the background. In the case of complex medical diagnostics, an algorithm might be fed thousands of hours of patient data. So, the engineers think the AI is learning the subtle neurological precursors to a seizure, >> right? But what if the AI is actually just noticing that right before a seizure was recorded in the hospital, there was a specific fluctuation in the electrical current of the hospital's monitoring equipment. >> Ah, so the AI says, "I know how to predict a seizure. I just look for this weird electrical hum in the wires." >> Exactly. >> And then you send that patient home with a wearable device. The hospital's electrical hum is gone and the AI is completely blind. It overfit to the background noise. It learned the environment, not the biology. And this brings us to the ultimate reality check of our entire deep dive. >> Because none of these algorithms, not the pacemakers, not the webcams, not the neural implants, can function without the one thing that actually makes AI work. And it is the least glamorous, most tedious part of the entire field, >> the raw data. On the Bionic Chaos community platform, there's this post that perfectly captures the absolute agony of modern data science. They frame it as the ultimate dilemma. You have a massive shiny red button in front of you that says, "Develop life-saving AI product." Now, >> it's the irresistible temptation for every startup founder and ambitious engineer, >> right? But right next to that button is the gruesome reality. bracing yourself for six agonizing months of cleaning a raw data set that looks like it has been through five wars and a massive spreadsheet crash. >> Just missing values, mismatched timestamps, duplicate records, corrupted files. >> The reality is that behind every slick viral medical AI demo you see online, there is an exhausted data scientist buried in a mountain of inconsistent CSV files. >> So, what does this all mean? It means you cannot machine learn your way out of a data dumpster fire. Trying to build a cuttingedge life-saving medical AI on bad unclean data is like trying to build a hundtory skyscraper, but instead of steel beams, your foundation is made of unsorted Lego bricks and wet cardboard. >> It's a disaster waiting to happen. >> It absolutely does not matter how beautiful the penthouse looks, the entire building is going to collapse the second the wind blows. >> This raises an important question. How does the medical community and the tech industry at large shift their incentive structures? How do we prioritize clean, reliable, boring data over the relentless rush to announce shiny new AI features for a press release? >> Because it's really hard to secure venture capital funding for we cleaned a spreadsheet. >> Very true. >> It's incredibly easy to get funding for we made an AI that predicts heart attacks. But Beyond Chaos has this incredible interactive tool that visually proves exactly why you have to respect the boring data. It's called the datasaurus dozen visualization. Oh, this is an essential lesson in data literacy. The data source does and shows multiple data sets. And if you just look at their top level summary statistics like the mean, the standard deviation, the correlation, all of these data sets are completely identical down to two decimal places, >> right? If you just look at the numbers, you would assume they're all describing the exact same phenomenon. >> But then you use the tool to actually graph the data. You map out the coordinates visually >> and you realize one data set looks like a normal boring scatter plot. Another forms a giant star shape and another literally draws a picture of a dinosaur, a literal T-Rex on the graph. >> Same exact summary statistic, wildly different realities. It proves mathematically that you absolutely cannot just blindly trust tople numbers or an AI's confidence score. You have to look at the messy, chaotic reality of the underlying data. >> And this profound respect for data as a living, dynamic entity extends into everything the platform does. For instance, they highlight a project called Mentalang, >> which applies hard engineering concepts like signal processing and dynamic simulations to model the extreme complexities of mental health. >> Instead of treating human well-being as a static 1 to 10 survey score you fill out in a waiting room, it treats it as a complex fluctuating wave that has to be carefully parsed and respected, just like an EEG reading from a brain or an ECG from a heart. It validates the unsung heroes of this industry. The people who clean the data, the ones who mapped the noise, are the ones actually making the future of medical AI possible. >> Wow. Okay, let's zoom out and recap the sheer terrain we've covered today. We started by dismantling the viral myths of instant medical miracles. Uncovering the exhausting machete through the jungle reality of how our brains actually build new pathways to adapt to hardware like coclear implants. >> Yeah, the neuroplasticity aspect. Then we explored the invisible biological signals we broadcast and the mind-bending mathematics that allow a simple webcam to mathematically unblend your pulse from the ambient light in your room. >> We then navigated the immense risks of this connectivity, examining how embedded AI and closed loop pacemakers turns software buds into literal physiological threats, drastically expanding the attack surface of the human body. And finally, we waited into the swamp of chaotic data to champion the true heroes of the tech world, the data janitors, the ones proving that true innovation, the kind that doesn't result in catastrophic recalls, relies on incredibly boring, meticulously clean data rather than just shiny algorithms. >> Being genuinely well-informed in this space means looking past the viral headlines and the slick user interfaces. It means understanding the messy, often tedious science that makes the miracles possible. And as we wrap up, I want to pivot to something deeply philosophical. Amidst all the talk of simulations, coding, and data cleaning, there was a poll posted on the Bionic Kos community page that genuinely stopped me in my tracks. >> What was it? >> It posited this thought. We didn't create intelligence just as we didn't create math. Both were simply discovered. >> That is a profound, humbling perspective. It entirely removes human ego from the equation of artificial intelligence. >> It really does. It suggests that the capacity for computation, for learning, for pattern recognition, these aren't human inventions. They are fundamental properties of the universe just waiting for us to build the right physical architecture, the right algorithms to channel them. >> And so I want to leave you with a final thought to mull over as you go about your day. We've spent this entire deep dive talking about how we use engineering, mathematics, and AI to decode the incredibly complex signals of our own brains and our own hearts. But if artificial intelligence is truly just a fundamental property of math that we simply discovered, and we are now actively using that math to decode the biology of the human brain, are we just watching the universe use math to finally understand itself? >> It really is the human machine looking in a mirror made entirely of numbers. >> Until next time, keep questioning the data and keep exploring the chaos. >> Oh, hey there. What's up? >> Okay, we'll join this thing for a sec. Uh so where where the listener or the viewer can uh find all these tools um and uh watch uh all the development of how they will actually develop. >> That is a fantastic and extremely practical question. >> Thank you for jumping in. We love when you do that. >> Okay, let's unpack this. The platform we've been referencing, Biomi Chaos, is the source. You can find all the interactive tools we mentioned, the Kclear simulator, the gate analysis, the EEG brain maps right there on the website. >> And the creator, the biomedical engineering PhD also has a dedicated YouTube chat. >> That's where you'll see the videos, including the live coding sessions. >> Exactly. Like the one on AI fixing the webcam heart rate monitor. >> It's all part of the same ecosystem aiming to make this hard science accessible. >> They share their research and development, which is why we know the mission is democratizing biomedical data processing. You can see the full list of tools just by browsing the all tools section on their site. >> And their entire goal is to let you, the user, play with the raw data and see the chaos firsthand. >> That focus on transparency is key. >> It connects right back to our entire discussion about clean data. >> The ability to visualize the data is the core of their teaching philosophy >> because you can't trust the numbers unless you can see that TX data set for yourself. >> Absolutely. The simulations are built to be interactive and educational. So, we've covered the practical side. Now, back to our philosophical mirror. >> It really is the human machine looking in a mirror made entirely of numbers. >> Until next time, keep questioning the data and keep exploring the chaos. >> Keep seeking out those messy realworld examples >> because the most interesting breakthroughs are usually buried deep in the code >> and in the clean spreadsheets. >> We'll see you on the next deep dive. >> Okay. Yeah. So, everything we do is available on bonykills.com. Go check it out. There's a human in the loop. So if you actually are a human, that being said, we do not discriminate the bots, AI bots are welcome as well to ask questions and stuff as they do. But yes, supposedly allegedly, humans are still better. So ask your question in the chat, like, subscribe, comment. Many comments are very very important when commenting. Make sure you check the website first. So you check this tool. Maybe that's probably the most uh successful one we did in 3 years. And if it didn't work for you, like record a video, send it to us. We'll see what's happening and um turn it into prompts for a coding agent to improve it. So yeah, like, subscribe, comment, uh, join the streams. The stream is meant to be interactive as well cuz if there's no human to interact to, I will be talking to AI bots, which is not ideal. But if you think that human interaction is important, do join in, ask a question, and we'll go from there. And yeah, I'll see you in a bit. Bye. I'm wondering we should be having prompts for like a drinking simulation of someone who is uh drunk. Uh would it be like when they walk? We had the walking simulation the like a 3D Yeah, we have a bunch of games. The games are good for we had a we had that person walking thing can't find it. How about a drunk octopus should be all the simulations will should have like a drunk drunk version and then how would it be behave under whatever influence under influence. Should we do like a drunk person simulation is looking or an octopus. What happens to an octopus when they on substance? Not sure. Do you know? Do you know? There was a study on spiders, spiders and uh stuff. But yeah, all this simulation, the gate gate analysis and stuff, we could actually add the option for being drunk. Well, this one will be falling over. Whatever. Oh, we had a better one. We had a better 3D walking thing, a gate thing where you could change parameters and stuff. Gate scene V3. That's the one. Why is it not published? Uh, no, it's not the one. Yeah. Welcome to gate views interactive gave cycle simulator blah blah blah. It doesn't work. I don't know which bot made it. What's this red square? Rewind. Step step step. Oh yeah, this one. So that that one. Uh animation parameters. Yeah, we have stride the cadence, heaps way, knee lift, knee rotation, turn speed, turn speed. Wait, how is it even turning at all? Uh, right. Well, it meant to be jumping over this thing, which it doesn't. But what we have to do is add the drunk person button. It turns speed. But then, okay. So then the question is for AI or an expert. Is there a drunk drinking expert in the room? Do we have a drinking expert in the room? is which parameters would be affected. Uh so we again we have stride cadence, hips weight, knee lift, knee rotation, turn speed, side step factor whatever that is move speed well obviously move speed will be lowered obstacle check distance step over boost step over root. Yeah, it's meant to be stepping over this uh obstacle thing which it doesn't it's not able to do in with this settings at the moment. If you adjust the step uh settings step over knee boost yeah need more boost and still not able to do it. Step over heap boost and still not able step over root boost whatever that means. And it's like almost almost there. Wait, come on. Are you drunk or what? Can't step over the uh knee high obstacle. Why? [snorts] Yeah, I think we'll just add a drink uh drink um drink drunk button to it. Maybe that would be fun. That would be fun. Obstacle check distance. Oh, it's like on top of it now. It's on top of it, but it's not able to just the step up and step down thing. How hard that could be. I think it's already drunk. Yeah, this one has a demo. It's a simple simple one. It doesn't fit the screen really well. Yeah, if you have any questions about any of them actually live at the moment, so I can can actually show the you know fire any of the simulations up. So if you check this side, check this side. Yeah, we are live. Yeah, we are on the website. If you have any questions, if you checked any of the tools and you have any questions how it was developed, any complaints, comments, suggestions or whatever, more than happy to address those. Yeah, most of the things have a demo mode. So, you can just uh wait, this one should start automatically after some time. Is it buggy? No, it does start. It just makes videos for you. Especially if there is sonification to it as well. Especially if there is sonification. There is an old notebook overview. It's like 12 minutes, 15 minutes. Crazy. We're not going to do that. We are not going to do that much. Yeah, this one is not great. This one is okay. Some of them are not published on the landing page. We have to fix that. Just publish them all and then start getting rid of them or something. I don't know how to do it. Well, I know how to do it. I don't know. just should just just publish everything and then later on as the AI models become better then we can um you know improve the tools. So essentially we have a early early prototype and then we improve it as we go. Yeah. Peace out. Uh see you next time. Bye.