#28 Brain Computers: The Future of Biocomputing | FinalSpark

Today's conversation is with Dr. Fred Jordan. Fred holds a PhD in signal processing and he's the co-founder of a company called Final Spark. Now, Final Spark is exploring this radically different approach to computing by using biological neurons and brain organoids as a computational substrate. That's right, we're talking about using brain cells as computer servers. Now, so Fred has spent years thinking about the limits of traditional computing, mostly around energy efficiency. Like if you've seen all the data centers popping up and been reading about how much electricity they all need and how that's totally unsustainable, it's really a big problem. And so he's looking at what it might look like to move beyond a siliconbased computing world and into a carbon-based computing world. And we talk about why biology kind of offers us that path forward. what it actually means to compute with a neuron by attaching electrodes onto the petri dish and how far away, you know, we are from us all being able to log into our cloud-based server and send these brain cells some math so they can do some things for us. But we like these kinds of conversations because it's a place where sci-fi meets reality and we have to dream of a better world. I think humanity is oftenimes very good at making incremental improvements, but we're terrible at inventing new things. So we have to keep our eyes on guys like Fred that are attempting these moonshots. For example, it took humanity thousands of years up until 1903 for the Wright brothers to achieve controlled powered flight. But then it only takes like 50 to 100 years and we have global air travel, supersonic flights, high earth orbit, you know, you name it. And look at how much that technology has improved our lives. So this might be one of those zero to one moments. And I'm so grateful we got to have this conversation with him. If you haven't subscribed yet, please do. We just found out from our Spotify wrapped for creators that we are in the top 2% of all brand new shows. And so it's all thanks to everybody like you who's tuning in and we're going to keep getting all of these great guests. But please subscribe if you haven't because it really helps us and we're still within our first year. Okay, onto the show. Welcome to the Austin and Matt podcast. Yeah, it sounds like something out of science fiction. And h but well, I have a question backing up. How how do you know how do we know as a species how a digital neuron works? And I and I you have a a PhD in applied mathematics. So I think that has something to do with it. But how if we if you can help me understand how do we even know what a digital neuron does or work? Like what does that even mean before we get into the brain the actual neuron like how do you then model that in math? Well when okay so when we say digital neuron we say a simulation of a neuron. Okay. Okay. So basic simulation of neuron things that we know since about one century is that basically neurons have dendrites and with the dendrites they collect some information and above some threshold they are going to have a spike that means that release is what is called an action potential which is a sudden uh change in the polarity of the neuron. This is known and this has been modeled with differential equation like 30 years ago. Okay. Now the mathematical models that are used today in our computers are way way simpler than this. But say that we have also there is already a very strong literature on this. That doesn't mean that we know exactly in all details how neurons work because there are many types of neurons in our brain and knowing even knowing how a neuron work is not enough because there is a network of neurons and the dynamic here is very much mysterious. in moreover in our brain we have different areas that interact together and at the end of the day what matters the most surprisingly is not the neurons uh it's more the synaptic connections that means how neurons connect together okay this is really where the magic happens and the magic is learning and uh so here we are lacking some mathematical model uh and uh so we know a part mathematically and most of the things we don't know. Okay. Okay. And this looks like a black box. Yeah, it looks very much like a black box. Uh so there are some very uh specific details that are very well known in neurons. Okay. But it's the problem in neurobiology. There are very some things that we know very very well but most of the things we don't know at all. That's incredible. So that so like for engineer you know you can say okay I'm going to either have to understand exactly in all details how it works okay or I can find a shorter way and use the real thing even if I don't understand how it works in all details and just use it. Yeah it's pragmatic very pragmatic you know. Yeah. So kind of just looking at inputs and outputs is that the idea like forget about how a neuron works on the inside just look at inputs and outputs. Okay. Absolutely. Absolutely. But you know this is uh how we do even with artificial neural networks. Artificial neural networks we know exactly actually how it works very much exactly. Okay. Um but the network uh is so complex that at the end of the day uh we only qualify the output of the network in relation to the input and we don't really understand what's going on even on artificial ones. So and it works at the end of the day. So the pragmatic approach works with artificial neurons. Therefore, it's a reasonable assumption to say that we could do the same thing with biological neur networks knowing something else which is very important is that you know I know that biological networks work because we are able to discuss together right now. That's right. They're firing. They Yeah, they seem to work sort of okay. They're kind of Yeah, they don't self-organize really well all the way up. So, whose neuron did you choose to start with first? It's like who got the first brain cell plucked out of their head to start testing it out? Well, well, it doesn't work this way. And uh there was um a discovery by a professor Yamanaka in Japan and he got the Nobel Prize for this. Oh yeah. Yeah. like 15 years ago and he found a way to take uh cells of your skin, okay, from human skin therefore and transform them into stem cells. And once you've got stem cells, human stem cells, you can you can do basically any kind of cells including neurons. So this is what we do here in the lab. We just receive stem cells. We keep them in liquid nitrogen and when we want to use them, we store them and then we transform them. There are specific protocols into neurons and gile cells which are another types of cells that you have in your brain. What's the difference? I'm really curious since you're looking at gile cells because I heard that whales have a higher percentage of glyio cells than humans do. What are the glow cells function inside of a neural network like that compared to a normal stem cell? Well, I I can say according to the literature um glad cells um support for neuron cells. Okay. Uh I would say this is um a quite traditional and old way to characterize gly cells. Okay. this kind of supporting uh the life of neurons and the activity of neurons. Um for instance, they're going to recycle glutamate. So glutamate on on the syninnapse glutamate are emitted and will be maybe recycled at some later point or the pamine also by astroytes for instance which are a specific type of gile cells. Okay. So there are number of support functions which are done by glyio cells. However, uh a number of researchers and I remind you that I'm not a neurobiologist in any way um have discovered that maybe glau cells have much more importance that we thought. So it maybe it's not only life sustaining for the neurons or life assistance. Maybe they have may way other roles including in training. Okay. So this is kind of still very mysterious all the functions of uh gly cells. Okay. So you're using skin cells that you somehow demarcate into stem cells and then there are also glyio cells involved or is it just like your computers are just a ball of of skin cells turn into stem cells? Uh no actually uh uh the bowl that we have okay is a mixture of glyial cells and neural neuron cells okay which so initially at at the very beginning you're right at the very beginning we have a spheroid people or neurosphere okay that is made basically of stem cells and this ball we are going to transform it and force there are protocols for this with specific molecules proteins that are going to force them into differentiate. Yeah. And at the end we get something like around 50/50 between gly cells and neural cells. And in a small bowl like 1 millm of diameter will have I don't know 10,000 neural cells and and gile cells. That's incredible. So you can take the skin cells use the Yakamoto factors to bring it back to a pur potent stem cell and then differentiate it back into neuron neurons and gile cells. Is that right? Correct. Yeah. Absolutely correct. Yes. Do they automatically differentiate or do you have to like do you have to keep the sort of the goo that they're they're sitting in? Do you have to keep that a certain way so that they'll some of them will become gly cells and some will become neurons kind of? Yes. Correct. So this is basically going to be a bit a random process. Okay. Which one are going to from our point of view? Yeah. Which are going to come gal cells and uh and neurons. uh but for instance we introduce uh um protein like BDNF brain derived neurotrophic factors or GDNF uh into the medium. So medium is basically water and amino acids and vitamins and everything and we introduce this molecule and this will force the differentiation and remember as soon as we force the differentiation we stop the multiplication because be aware that as long as you have stem cells you are going to expense you are they are going to multiply. Okay so um we are actually even to take care of this because um otherwise we would get too many neurons. Yeah, it'll keep growing. It keep it keep splitting. It'll keep going. Yeah, it keep Yeah. So, it's incredible process. It's really life. So, imagine when you have a microprocessor, you do a microp processor, you don't have this problem of a transistor multiplying by themselves. Yeah, that's right. But this is the situation that we have. Okay. So we have really to time this prec precisely week by week to force this and stop the process not to get too too uh too large too much large uh balls because uh if they get too big actually the center will die. Ah because it doesn't get enough nutrients or something or it doesn't you're because you you're basically artificially keeping this these cells alive. You have to keep putting in glucose or like keep feeding them to keep them alive essentially. Absolutely. Yes, we have a full microfilic system that will perfuse okay uh some new medium and uh get the the waste uh 24/7 uh to make sure it's uh you basically you've created a synthetic body, a synthetic body for the cells to kind of exist in by giving all of their inputs and outputs they need for life. Yes, absolutely. But it's not that different from if you have for instance fish small fishes at home. Sure. In in a fish tank, you got to put the fish in and clean clean the tank. Yeah, it's the same. That's incredible. How how then? So then what do you do? How do you do you plug it in plug wires into it? Like how how do you then use this ball of cells to get any information out of it? put first put information into it like a question or something or an input and then get an output and and derive what that even means. What do you do once you now have this this this ball of cells? So what there are many questions in your question. Yeah. One question. Let me ask you five different ways. Yeah. What's the process? I think he's trying to ask which operating system are you installing on this computer? Did you install Linux or or Yeah. So we by far not there yet. Okay. But so we will take so we we call this basically a brain organoid because it's not the a brain organ. It's not brain but it's a bit like a small part of the organ. So people call it organoid. Okay. So we will take this small ball and put it on electrodes. Number of electrodes. So just put it. Okay. And uh and then after sometimes uh that this will create uh connections actually in some cases in less than one hour we can already monitor some activity. Whoa. Is this similar to Neurolink how Elon is putting a bunch of uh electrodes electrodes into people's brains and then it's measuring the electric signals out of it. Is it similar to that? It's similar but I don't know the details of the kind of electro they are using. Okay. In our case, we are really measuring the individual actual potential so spikes of each of the neurons. Okay? So we are not measuring um like the average activity of the neurons. No no it's really we can say this neuron has spiked at this precise time and again and again. Now this is for the output uh from from the the organoid to us. But we can also to answer the other question the other way around. Okay. We can also input using the same electrodes some signals. Okay. So now just if I tell you about what we are measuring electrical signals. Okay. So just that you are the order of magnitude right we are measuring something like 50 microvolt. This is a spike. It's 50 microvolt. It's very very small. Okay. And this comes with a number of problems everywhere. Yeah. And the other way around when we stimulate, we stimulate with a very small current like one micro amper. Uh this is going to need a stimulation current and then we have a start of conversation. When you first thought, okay, we're going to create a ball of neurons, then we're going to put electrodes on. Was there any consumer hardware that you could use to measure those tiny voltages or did you have to basically create hardware? No, no, that's that's that's the good thing. It's a very important question that you say here. Okay. Uh so it's not really consumer hardware. But um this is called the field of electrophysiology. Okay. Electrphysiology is not new. It's been going on for the past 40 years. Therefore there are number of suppliers who already supply this kind of electrodes. So basically most of the hardware already exist but it was not used for this purpose. It was used to study uh neurons. It was not for creating a processor. I mean this is really a wild idea. When you talk with neurobiologist say I'm going to create a processor. Yeah. Not possible. Doesn't make sense. I mean we don't use neurons for this. We study them. We don't use them. And that's a very, you know, this is the big difference. We are engineers in this company. Okay. We are less interested by understanding all the details and studying things. We want to make things work. It's way different. This is the difference. Yep. Yeah. You're tinkerers. I I think tinkerers make all the biggest revelations in in human history and science. Normally, it's the tinkerers first and then the scientists later figure out how to describe it. That's my understanding of of life on earth. It talks a lot to me. Okay. Things uh about the the first people who did airplanes. Okay. There are equations like BUI equations that defined the differential equations that defined the fluid motion around a solid body. Okay. This were discovered way after the first airplanes were successfully able to fly. Exactly. Well, when you and when you're connecting these this hardware to measure the voltages, is it to is it connected specifically to one neuron or is it like a batch of neurons that all sort of act together at the attachment point? Yeah, that's also a good question. You have a lot of good questions and u yeah so I would say 70% of the case I would say it's one neuron per electrode. Okay. But there are cases when we have two or three neurons per electrode and this we can know because each neuron has its very own way to spike. So there are sort of a unique signature. Specific neurons will always spike the same way. Same duration, same amplitude. So if you have two neurons, you will see two shapes which are not overlapping correctly and you say, "Ah, I've got two." So neurons sort of have a fingerprint really. Yes, it's absolutely a fingerprint and some algorithms and people know this very well in electrophysiology field or so. Uh this is called spike sorting. So there are a lot of people who have been working this past 20 years about developing specific algorithms for spike sorting and people of course obviously are also using artificial neuronet networks for doing spike sorting of biological neuronet networks. Fantastic. Oh, well that's that's a little meta. We're using Yeah, we're using code now neural networks to study real neural networks. Yeah, that that was designed from real neural networks. Yeah. Right. Right. That's the point. Yeah, that's incredible. You know, I mean u there is something called epistemology. Okay. Which is the study the philosophy of science and this brings a very interesting story. When we came from a bali and we made a mathematical model then we realized actually we can do things like this a bit like which has some processing uh capabilities and we come back to biology again. Wow. It's incredible journey when you think about it. It well it is. And does it ever make you wonder if the neuron in the first place was actually engineered by someone? No. No. um you know it looks like um if you think for about uh the energy efficiency of neurons okay uh it's incredible okay like 1 million times better uh than we have with artificial neurons okay but it doesn't mean that you need intelligence to do this what you need actually is to only select the living bodies who are using the most efficient uh way to process information and we are talking here about I don't know hundreds of millions of years of selection of the most efficient way to process information and at the end of the day well you get the neurons basically the same in bees and human beings and so if game theory gets to be the conductor it seems like the prize always that the outcome is the same it doesn't matter the species necessarily and well that and that makes me wonder this ball of of neurons that you that you have it's not really like a full brain would you call it like more of the prefrontal cortex. Is it like part it's like you've taken a little part of a brain kind of? Well, actually you we decide okay in your brain you have about 400 types and subtypes of neurons. Okay, so we are going to say for instance I want to have cortical neurons. Okay, we have specific protocols for differentiating into cortical neurons. But if I want hypocmpol neurons from your upper compass or stratum neurons, we are going to use another protocols and then we are going to get some balls under the microscope. They look the same. Uh so you have we have to do um analysis like RNA sequencing to to know the difference. Okay. But uh you get these different balls and then you can do something even more interesting. You can do what is called assemblids. So assemblid is when you assemble organoids. So you take different organoids and you put them together. Say okay I use some cortical neurons here for processing information and I want to be sensitive to neomodulators for instance. So I add three atom organoid and they are going to connect. Remember they have the same DNA so they love each other. So they will connect very easily. When you were a kid, how many times did you read about Dr. Frankenstein? [Laughter] Just kidding. I'm sorry. Well, well, it's it's fascinating though, isn't it? Because you can start to imagine different configurations of a brain and really you'll be able to do it. You'll be able to configure a brain with different because I imagine like the amydala has certain things it cares about and the cortexes have different things they care about. And I always wonder, what if you configured it slightly differently? What would you get? and we only have the examples that exist in the animal kingdom. But that's not going to be true for long if you keep going. Yes, it's correct. So here I must say it's a bit trial and errors at at this point in time. Okay. To be really frank. Okay. Because we so the first thing that we are trying to reproduce is the learning process. Yeah. Uh you know learning is very important. um obviously but I tell you uh about think about artificial networks okay 40 years ago uh people invented artificial networks and we didn't know how to change the weight between the neurons in order to have an input output which is useful and after some years some people discovered one way to tune these connections and this is actually a learning algorithm. Now we have to do the same in vitro. Remember that in vivo it works. You are able to learn. I'm able to learn. So we know it's possible. That's for sure. Okay. We just find the the the minimal the simplest way to reproduce this in vitro and and actually this has never been done. And this time it's not like electrophysiology. There is no literature on this. We are lost. You are just this is new. This is the this is where your frontier and then then you have to tinkle then you're well and it almost speaks to that we don't even know how we learn in vivo like we know we learn but we don't even know how our brains do it actually like how do you learn and what synapses fire that means we don't even know how it works for us because even if we had that one if we could do it that one way and we knew then maybe we could do it in vitro uh but we don't know so it's still trial and error we're we're a blackbox input output like we don't even know how it works Yes, you're right. So the good thing about in vitro is do you can perform experiments. Okay, which you can do. You cannot really want to do in vivo with human beings. That's right. So and we have also some some clues. Okay, we know for instance that dopamine uh is implied in the learning process. Okay, this is why we have also implemented the release of neuromodulators in our system. So we can uh basically uh very schematically reimplement the reward process. Okay. And so be able for instance like when you give a treat to a dog to do something. Okay. You just give a little you give a little dopamine hit to all these neurons literally. Yes. Sprinkle sprinkle the dopamine on and they get it and it's like a chemical. Yeah. Great. Wow. Yeah. Yeah. We we do this. Okay. So we have a specific process which is called uncaging. So uncaging basically it comes like this. You have a dopamine molecule which is um linked to another molecules and the full molecule is totally inactive. Okay. There is no receptors in the neurons for this. It's we put this in the medium. So it's all around the neurons. But if we illuminate with ultraviolet light, we are able to cut these two molecules and then you have only this time the dopamine. That means when you give a treat to a dog, you have to be very timely. Okay, you give at the right time. It's the same here. Uh we can be very precise because we are going to use ultraviolet light at a very specific moment. say, "Okay, at this point it made exactly the output that we wanted and we reward it." Wow, that's incredible, dude. That is incredible. Yeah. What's like a real world example of how some of your customers have used that or like what would be an example that you can that that that would help me understand that pra practically? You you mean practical applications of what we have? How can we use this? Okay. So I'm sorry you're going to be horribly disappointed at this point that was expected. I've found very few practical applications of my own neural network. I wish. Yeah. No, no, but uh this is still a field of research, okay? Because I told you, okay, this this could work, okay? This reward process, I tell you, but so far sometimes it works, sometimes it doesn't. Okay. So, it's still a field of fundamental research and we don't even know. For instance, um you know that when you repeat things, you get better at them. Sure. Okay. So it's not only dopamine there also the repetition pattern. Okay. Therefore we have also algorithms that will repeat things and repeat and repeat plus some dopamine at sometimes. So these are actually algorithms and we run these algorithms in Python okay and C++. So then you can program this. Okay. And what is cool here is that at least for me I come back into my own world which is digital again and algorithms and things that this so we can try algorithm and see what are is success rate. So and do do we cross decrease the loss for instance we can define loss like in artificial neural networks mathematically. So um that's the way people are trained and not only us but also researchers of many universities around the world because people access to our neurons through the internet. So we give them access so they can run their own Python script and their own ideas. So, so you have the first biological brains connected to the internet really like hardwired in I mean as hardwired in as you can be until you have a meta headset that can reach through your eyeballs basically you know there is um normally uh we are approached by many people okay including artist okay normally I'm not that interested because this is not the field however we made an exception uh a few uh a few months ago there were artists who came and what they did is incredible about uh remote interactions with uh living biological neurons. They made a kind of like a a plastic skin, okay, that you can push with your hand, okay? And when you push this in your hand hundred of kilometers away, it it sends a signals to our lab and stimulate the neurons. And if this neurons response reply, then this plastic skin is going to um react and change shape. So basically you can actually push your hands and interact remotely with your bare hands with neurons. And you can find a video on on YouTube of this. I found this incredible. Well, that's incredible. Okay, so let's say you have a brand new ball of neurons. How do you how do you tell them what's going to be important to them? Like I saw there was a simulation where one of the the little organoids was controlling a butterfly on a screen and it was sort of making it flap. That was very cool. How do you get how because you're sort of instilling a value system into this organoid? Like it's a little brain that's never had a thought before and you're kind of telling it, okay, what's important to you is butterflies. How do you make how do you choose which value system to instill into an organoid? Well, first of the butterfly system, basically uh I'm going to talk about the science behind. Okay, it's way less impressive when you do the trick. So, uh what it does is that um you have a 3D butterfly flying on the screen and you can also see the video and uh each time uh you can move the lamp okay light. Okay. All right. Each time while randomly uh rotating the butterfly sees the lamp is going to send some stimulation signals across the internet from your internet browser to our neurons here in Switzerland. If the neurons reply, it stops moving randomly and says okay I've got the right direction. So and it starts moving in right in the right direction and at the end it gets to the light. Okay. So here you understand that there is no learning implied there is just the capability to react to successfully to a stimulation. So there so on the sand side there is little things okay on the engineering side you cannot imagine the problems that we have to had this running reliably so because you have to stimulate remotely across the internet five times per second and have make sure that the are going to reply. So there's a bunch of problems that come with this but uh we did it and we did it several times including live. How often do you have to swap out? Let's say you have a customer paying to use a a little brain. How often do you have to swap that that little brain out? Yes, this is also a very good question. So that depends. Okay. And that's the problem actually in this field. A lot of questions are going to to be answered by it depends. Um so it's going to vary between a few days and a few months. Okay. uh sometimes we have variability and variability is a problem in in vitro biology. This has been the case for decades. Okay. And particularly on organoids which is basically a new field. Okay. Remember that I'm talking about brain organoids but many researchers doing organoids of other kinds of organs of the body. Okay. But we all have the same problems variability. Therefore, I put some arganoids on my electrodes, but I don't know how they are going to react and how long they are going to live. Do you see any of the cells that the electrodes are touching directly? Do they have any like are they the first ones to burn out? I don't know. I don't know because um we are we don't have um the ability to to see under the microscope so precisely which the neurons which are interacting. We see it only indirectly through uh the electrical field. All right. I I'm so curious what your sales calls are like like when you when you first decided that you were going to go try to sell a a real brain organoid. What what was your what was the first approach with a customer like? Did like do people sometimes not believe you? Well, that depends on the people. Okay, there are many people that that we connect with. Uh but you know, just realize that this whole story of um giving access to uh neurons remotely started with the COVID. Oh, there you go. And you know why? Yeah. People couldn't go to the lab. Yeah. the the the the the reason is that we could not come to the lab. Yeah. So we had found a way to work from home on our neurons remotely. Okay. That's so crazy. So it's not about vision from the beginning of my story just about a success of incredible story of luck and incredible circumstances. So at the end of the covid actually we could operate the whole lab include the microfilics and everything from home and this is why at this point we started to connect with universities and said well guys we are able to do this from 50 50 kilometers away you are standing 2,000 kilometers away but are you still interested to make experiments and we have so many universities they wanted now yes we want to try this okay uh because it was the only way to do neurobiology unless you have all the infrastructure that you that we have and uh right now I think we have nine universities who are using our neurons and uh and at some point in time so to come to your to your questions uh so we were giving this access for free okay and still giving for free for university okay but some private companies also came remember that we didn't make any marketing around this okay it was not and I said yeah we want also to access but we don't want to work on your particular project. We want to work on our own thing. We are not going to tell you. Um so we say okay can you pay? I said okay yeah we can pay. All right. So we started like okay so pay $500 a month. They say okay I pay. Uh then we had too many. So we say okay let's put $1,000 then. Okay. It should be okay. And now too many. So we put $5,000 a month. So now it's okay. We can handle what we have. So that's incredible. Yeah, it is. And you guys are just the servers. You're like AWS. You just Yeah, a very small AWS. But this is a very very But Matt, this is our target. Okay. I want that we do the same thing as AWS in biomp computing. I want in 10 years that we have a computing computing center like is uh with a nervous tissue of 100 meters by 100 meters by 30 cm thick and this going to serve millions of people for like 100 times or 1,000 times electric power is the when it comes to the competitive advantage of that I think well I think that's incredible uh other than electrical savings because I know that that's huge I mean the the gig gawatt plants that are being built right now and all of that is you know that AI is going to drink more electricity than like our whole country produces right now. So but other than electrical savings is there do you see any other competitive advantages to this in terms of like ways it calculates or speed or or anything like that or do or will this fundamentally kind of do something similar but could be at a drastically cheaper cost because of the electrical because it's way le way less electricity. Uh so some people say um and suspect that um the learning capabilities of biological neuronet networks may be far superior. Indeed, if you have a child, you don't need to show one million dogs and 1 million cats to your child so that he or she knows what a cat and a dog is. Okay. For sure. Sure. So we have some some clues here that there may be also way more efficiency in the learning process itself. Okay. Um cool. Now uh I'm pretty sure that there are other major advantages or superiority. Uh but I don't know which ones. Okay. I have and and I mean I know it's so new. It's so we're so early days. That's why I was wondering if anything was emerging. But it does make sense that you know a kid learns what a dog is after seeing two of them and that's done and you're done. So what if there and what and there might even be more intuition there. It might actually be more intuitive in some ways as well that I could learn things that it's harder to model just out of silicon based uh processors. It's possible you know I'm trying just to keep scientific. Okay. So I just trying to say things that I more or less say that which are backed up by science. Okay. So, as soon as we start in this, I am I'm just pragmatic prosaic. I'm going to use this because it's the result of hundreds of millions of years, it it must be good. So, so very basic reasoning, you know. Yeah. It's like the ultimate engineering thought. It's like if I find a tool that's better than all my other tools, I'm going to use it. It doesn't really matter what's under the hood. That's incredible. Do you as far as scaling because big data centers they talk about when they're trying to scale it they're going to have issues with cooling and things like that. If you were to scale to as as big as you could imagine what would be the sorts of things that you would struggle with in scaling in this type of hardware. Um I I I I don't see well I see many problems okay but these are engineering problems okay uh which can be solved for instance contamination okay um okay but uh this has been solved in the biotech industries like already 20 years ago and these are this is working fine um uh I need billions literally billions of electrodes uh in order to to have the input output uh I need Um I I don't see major things but there there are more fundamental things like um if you're training neurons for bological networks for a given task okay how are you able to uh uh to transfer this learning to uh other neurons other brain organoids what so the scaling process in this direction is still open okay now If I think about the lifespan, okay, which is also an operational uh consideration. The good news is that our neurons live I don't know 80 100 years. The same neurons, okay, we live with the same, okay, basically all our life. Okay, so u this quite robust if we do theis uh correctly. Uh so this should not be a major issue. Um I see technical problems but not things that uh I find so for me major roadblocks. Yeah, it's all just engineering. You just keep engineering and all of them are solvable. Yeah, I think so. Yeah. How long does it take to grow one of the organoids to start from just like a few cells? So you talk about grow. Okay. If I answer strictly to your question, it's going to be like two weeks. Okay. Two, three weeks. Okay. But uh it's a lie because after three weeks, if you go under the microscope in our lab just below, you will see this ball that I have a promise to you. Okay? But what I didn't say is that this ball is bly mainly based on um stem cells which are totally inactive because these are not even neurons. So so give it at least two additional months. So we have basically a majority of gly cells and neurons. Okay. But this is not enough. Okay. Because uh we you still need to have additional maturation of one or two months to have a spontaneous electrical activity. So the total thing is going to be from start to finish between I would say two and four months or three and four months. Answer. That makes me wonder where you got the name from. Where did you come up with the name Final Spark and what exactly is it a reference to? Well, it's not that sophisticated. Again, sorry to disappoint. I'm a very simple guy, but you know, we we we are two two guy here, Martin and I. And uh we created the first company which is called Alp Vision. So, Al Vision that means there is three concept uh Alp V which are following very difficult to pronounce. Okay. Uh I find okay. Um so this time I say okay company I want something which is very easy to pronounce. Okay that sounds good like like a rap thing. You go final spark. Okay. Okay. You feel the rhythm here. Uh the.com is free. Um okay that's it. There it is. It's a great name and the com was available. That's incredible. Actually Final Spark's a great name. Yeah. So, so when you're talking about that it takes a few months for this the ball of neurons to have its like initial electrical impulse that that that wasn't the final spark. I guess that would be the first spark that you're looking for really. Yes. Okay. Correct. And what do you think causes that? Like why do the neurons just just do it all of a sudden after a few months? I don't know. I don't know. Um but this is really uh more about me uh than science uh because I'm I'm sure there are some studies about this uh that explain uh at which points in the maturation process why uh neurons become active. Uh we just had an experiment one month ago where I was getting mad because you know I told you something before. I told you that for a given neuron they have always the same shape always the same uh potentials and um one month before in the past we had actually a neuron which was not like this never saw that before like it was it the senor was able to spike with different amplitude okay and after discussion with a number of experts we concluded that this was probably immature neurons okay so in the maturation process maybe They start to spike very very small. We cannot even see it. And pro progressively they start more and more and then we see ah it's mature. It's spiking. But actually it was spiking before but you you couldn't see it. It's like a teenager. It's just like it's just random spikes in the brain. You can't understand it basically. So well take care about random. Okay. Random is a powerful word. Okay. Okay. Okay. My apologies to teenagers. Yeah. Good. you know uh it cannot be fully random okay otherwise we wouldn't be able to have a conversation right right okay it cannot be okay uh there is here there is some causality uh some so um for sure there is um some level of randomness but some level only okay uh our behavior are mostly predictive as we can see here and therefore the internal circuitry also is Interesting. Will you allow customers to have lab add-ons? Like if someone wants to play Mozart next to its baby brain that's growing, will you ever allow that to be part of the service you offer to to do to to Well, like I went to tequila recently and there is these people that play Mozart on the tequila as it's as it's going through the process of like becoming tequila, the bacteria, and they say that it changes the actual product. I feel like your product is especially one where if we play Mozart at it, it's definitely going to change the way that the brain grows. And even if that's not true, would you allow customers to install that in the lab so that you could have some Mozart brains? Uh, okay. Um, just first there is no science to back up this uh that's that's not what we believe in Mexico. We don't believe that. But but um um yeah uh actually um I mean for Spark is a small company. Okay. Uh if a private customer want me to put a smartphone playing Mozart in the incubator and is pay is ready to pay $5,000 to make this experiment. Actually he will be the first one in the history of mankind to do this uh experiment. I'm happy to rent it. Wow. Yeah. No problem. Okay. I'm glad we got to the bottom of that. You know, um I I I tell you something about not not exactly about about Mozart. I don't like Mozart, by the way. I love Shopen, but not Mozart. But anyway, so interesting. Yeah, Mozart's Mozart's 6,000 a month. Actually, Shopan's only 5,000 a month. Yeah, because Fred has to listen to it. But uh about noise, we've made some experiments on making noise with smartphones in the incubator. uh because there is a mysterious behavior of neurons that we have not fully understood uh so far which is um you know all the neurons are inside a so-called incubator which basically keep the neurons at 37° 80% humidity 5% CO2 right and uh we notice that when we open the door they react we see the activity and normally they have they are not they are not equipped with sensors uh the neurons to react. Okay, they are not unable to see light, unable to to hear things and and we still don't know how and why they react when we open the door. And one of the test that we did was actually to put some sounds in the smart with a smartphone inside the cuber to simulate the opening of the door to see if they were reacting to the sound and they were not. So just to tell you, I don't know if it make a difference with Mozart but with the sound of the noise. No. Wow, that's so interesting. Okay, I I just want to be the first person to ask you this question. There's there's a podcast that's come out called the telepathy tapes where these autistic children apparently have telepathy with their parents in particular. It's they've they've been doing studies. I don't know how voracious the science is. So don't hold me to that. But I'm curious if you will be able to detect, you think, if there is such thing as telepathy between these organoids, like if somehow they're sinking or somehow they're communicating in ways in some ether that we don't understand, you should be able to see this eventually in patterns coming out of different organoids because they're sinking up together. Is that right? Yes, it's correct. So um technically speaking, it's it's actually is very simple experiment to do. uh particularly because if you go on the on far website in the live section you will see the setup that we are using okay and often we are using a setup when we have basically four organoids sitting close to to to each other like 1 millm of distance okay therefore it's very they electrons are totally independent okay so uh if we would find a correlation between the activity uh temple correlation that would sort of prove telepathy okay which we never saw uh uh so far. So that's cool. No, that's a cool data point in that conversation because a lot of people are talking about this idea of telepathy right now. That's interesting. And that actually it makes me think of do do you think that in the in scaling? Do you want these uh all these cells to be the exact same DNA or is there any benefit to kind of having different DNA at scale? Uh short answer is I don't know. Okay. Okay. Uh yeah. Okay. Um you know so so we don't have uh immune system um here. Um so I don't know how it would react. So yeah some some people some researchers are mixing cells from rats with neurons from human and it works well actually. Um so basically uh it could be possible to use different types of DNA. Uh technically it's possible. Uh don't remind that I've read publications in this direction. Um I don't know if that would be useful or adverse actually but you know I want to be conservative. Okay. It's already enough science fiction. Okay. So and in our brain we have only one DNA it already works okay so let's be an engineer and be conservative works this way if it works don't fix it so okay let's keep the same DNA but even at scale in a huge building would it all be one DNA or does it little battery packs of like maybe it could be other cells that kind of thing or I mean it almost because I I start thinking well if it's all the same DNA and all of a sudden you have biological mass That's huge. I mean, that's it's going to be bigger than a whale and it's all one piece of DNA. Like that that sounds so science fictiony to me. I don't know. I don't know if that even makes sense, but uh is there is there a method to it? You know what I mean? Well, it seems surprising. Okay. It is actually you. This would be the largest uh nervous tissue uh ever to live on Earth. Okay. Hickey. Okay. You're right. Um but technically there is um there is no reason why why we cannot do this. Um yeah as I said at the beginning we have to stop uh the process of otherwise we would have too many neurons. So we have specific molecules to stop uh the uh multiplication process and force the differentiation. Yeah. So we have the reverse problem actually. Sure. right now. Are the cells you're using are they male or female cells or is that irrelevant? Um I I don't remember honestly. Um because I had the same question a few weeks ago and uh and I look with our one neurologist said and uh and she answered to me and I don't remember the answer but yeah. Well then it makes me wonder are they are they Swiss cells or are they like Austrian cells? Where are they coming from? And will we have a competition? Uh so the the two uh suppliers that we are and one is in Canada and one is in the US. Okay. But it doesn't tell me the identity of the donor. I guess it's secret. Uh but you know I could I could use my own. Okay. Uh it's I would be happy I would love actually to use my own uh sales. Um the the only thing that prevented me to do this so far it's the danger for me because um I'm working also at the lab. Okay. And if for any reason I would uh inject uh by making a mistakes uh some of these stem cells of myself into my body. Uh this could be dangerous because that will be recognized as friendly cells. However, they will multiply. Uh, so looks like so not they get to walk past your immune system totally unencumbered. And so if there's something wrong, that would be a problem. This is an interesting moment for young people when they hear about a technology like this because if they were interested in in joining and studying this for the rest of their lives or or just getting involved, what would you tell young people they should study? What should they get smart at in order to to be part of a team like yours? So yes, it's a very good question. Uh first I must say that we have a discord server uh when we have a lot of people coming including teenagers who ask exactly uh this question. Okay. So so I would say I love um students uh who study both things neurobiology on one side and I would say uh artificial intelligence. Okay, that means being able to program from the ground up. Okay. uh uh artificial neural network. So not only use a high level function of PyTorch but really know how to uh I know the mathematics behind okay so and uh and of course read the literature okay uh because there is a lot of things here that have to be known okay uh to be productive but we always have interns every six months we have a new intern uh coming from a master um training here in the neighborhood and they have this double uh training and it's just Perfect. The only thing that all students are missing including sometimes researchers is skepticis skepticism. Okay. When you make experiments, okay, and you think it's success is it is successful probably it's because it's wrong and it is the experimental artifact, right? I I say this because we are a company so we just don't care about publishing. Okay? But most of the science today is aimed towards publishing. So first good result, okay, I publish even if it's not true and nobody will be able to tell if it was true or not. Okay? And but we do care because we want to do things that work. So and and I can tell you firsthand uh many times we thought we had something and it was wrong. Okay. Because you try to redo it and it doesn't come out how you thought. So there was some noise or something. Yeah. Correct. Okay. So read the data. Uh be skeptical about your data. This can be an artifact in many way. There are many ways to be wrong on the numbers. So what where do you see this going? Where is final spark heading? Are you do you want to be the AWS? Do you like in five years, 10 years, do you want to have a throughput capability of some sort of server capacity and start selling that? Yes, absolutely. Wow. Yeah. What what else do you need to keep going? What what what do you need to get there? Money. Money. Are you fundraising? Yeah. Oh, okay. Yeah, that's incredible. I mean, using carbon-based instead of siliconbased anything. I mean, it's it's it's so science fiction. I would imagine you have investors very interested in this just in terms of like the moonshot of what this could possibly mean in terms of our understanding of many areas of the world. Uh yes but VC you know venture capitalist are not only about this it's a they want to to see at least that there are many others who are doing it okay and who are investing so it's so early that it's too early for VCs okay um this this could be cool for sovereign funds yeah for sure have a okay a longer also duration of investment or thing like this so But um initially we are looking for 10 million uh let's say dollars first round $50 million second round with sovereign funds uh we are at the difference here. It could be $50 million first round 300 millions uh second round and these are the kind of of numbers which are compliant with this objective of having uh to beat AWS in 10 years. Yeah. And is it are you finding that the what your customers are using it for is it pretty efficient? Like is there is it not just energy efficient but are they getting value out of it in terms of would they rather use this than an AWS server? Of course. No, there is no question here at this point because the people but but be clear uh the people who are using it they're not using it for computation right okay they're using it basically for doing fundamental research. Ah okay. And you cannot do this on a digital computer. No way. Okay. And the data they get is unique in the world. Yeah. It's very difficult to to do to to make these experiments. So and for them it's very easy. Actually we do the work behind. So to have all the organoids living and putting them and all the microfic 247 etc. All the show pen and the show pen. You got to rotate that around. You're like brain brain mechanics. Well, this this will be the last question for me because I know we're running up on time. I'm just curious. You've spent so much time looking at these neural networks that are sort of live neural networks. They're real ones. Has it given you any insights into your own brain and sort of the machinations of your own brain? So, I would I would say um a bit yes, but it's it's non-scientific. It's just personal. Okay. Um I spent really we a lot of time you know under the microscope looking in those neurons and looking at the spikes and things like this and um I mean I also uh did the same for rat neurons for by the way okay um and there's not there are not so much difference between rat neurons and human neurons okay from from a phys under the microscope first okay and if you look at the signals al also okay and the general way it works is very similar okay I would love to say that we are somehow way way no but the science doesn't show this and this is in according with all literature there's no major difference okay and this comes back to your initial question about uh looking it at machinery okay and um the problem is that uh if you think a neuron is a machine because actually after all I want to use it so somehow it's a machine okay if I add several machines is it still a machine okay I mean obviously I would say yeah probably yeah I don't know so and then then what's next what is free will in this consideration okay do do machine have a free will really no they don't okay so where do I go from it I I don't know But I just tell you I'm a bit lost here. It's so everybody is going to to think his own way about this. But uh think that uh there are already some physicists like uh I know tens of years ago like pen rose for instance who look at neurons and he was a famous physicist. All right. So perfectly respected according to science uh physicist. Okay. And and he went into uh questioning about neurons and uh free will. Okay. And his only research was to say, well, okay, maybe there are some quantum effects in neurons and maybe there is a free will that I can hide between the quantum effects. Okay. And uh but there is absolutely no science as far as I know which is um compliant uh with either connections between free will and quantum mechanics and quantum mechanics and neurons. So there are two connection here which are not proven. So wow. So so the big takeaway is humans probably don't have free will. I guess I don't know just an engineer. Okay. Just all over a morning cup of coffee by the way. Yeah. The problem I have you know I I tell you um there are things that I know okay for sure. For instance I stimulate with one micro amper my neurons. This is true. This cannot be discussed. Okay. and and most of the interesting questions are not the things I know right unfortunately. Okay. So, and it's very easy uh to get out of his out of my field of expertise and go where I'm not expert at all and I know actually I don't know. Okay. So, I can imagine a lot of things but they have no value actually. Yeah. Well, it's fun. It's fun to hear you imagine anyway because it you know it just connects to the humanity of of each of us. It's I think all of us are just wondering. Yeah. Yes. Well, that being said, I I think it's a good place to wrap. Matt, do you have any other questions? Yeah, Fred, this was awesome. Thank you so much for sharing about what you guys are doing. It's fascinating. I mean, it is it is a straight out of a science fiction novel. And I can only think when more people start getting on this and as you start scaling, what's going to be uncovered by being able to utilize this? like just both in terms of like processing and learning, but then just for humanity to to the point it's what do we do when we throw more neurons together and what can emerge out of it even I didn't even know we had 400 different types of neurons in our brains and so let's re let's do a bunch of combinations and see what happens like that sounds awesome sounds like we should be doing that so thanks for doing all the work that you're doing yeah seriously no it's it's a pleasure you know but there is still one big mystery for me okay Um this idea of what we're doing today has been in science fiction for long time. Okay. Right. Yeah. So people have been dreaming about it but have not done it. Okay. Uh it's not the first time in history and technology that it it happens. Think about airplanes. Okay. We could have made a glider, a simple glider 2,000 years ago and people were dreaming about it. This was a science fiction at that at that time. Okay. And uh and then we had to wait 2,000 years to go from dream to reality. At least to try. Okay. And why does it take so long to go to to move from dream to try to do it to try to make the dream come true? This is a big mystery for me. And remember uh Final Spark is one company working on this. But we are not the only one. Okay, we are probably we are probably among the the first ones. But it's like airplanes. Airplanes suddenly started to the idea or inventors on different continents different countries at the same time suddenly people started to create airplanes and it's the same now for biological neurons used for bio computing I don't know what is the mystery behind this coming from dreams to trying to make it real it is so interesting the like like when man didn't think they could run the four-minute mile and then they break the four-minute mile and then that year I think nine more people broke the four-minute mile But but the the previous year they didn't even think it was possible. And so showing some showing something about humans, it's very rare for a human to have a new idea. But once that idea exists, then other humans can get a hold of it and kind of get behind it. So it is something strange to your point. It's a little snowball that in humans has to start. But I don't know. I don't know what makes it pick up momentum. I think a lot of times it's the military-industrial complex. I think I think when we need airplanes to go to war, that's when that's when it goes off. So when they approach you and they need brains to help them go fight battles, I I just pray that you'll say no, Fred. But but do what you must. Do what you must. Yeah. Have you been approached by any military to No. No. No. We have not. No. No. Okay. Okay. Well, we're basically very much uh I would say undercover right now either it's really um in scientific publication scientific journal people who are aware of this and you know people who love technical things and podcaster who love about uh new science uh these are the people who are aware of this science but uh most of the people don't know about this. Do do you consider yourself a person that likes to buck the modern like modern thought in society? Because to do something this outrageous, you kind I wonder what your personality is like. Are you the sort of person that just wants to do things contrary or was this just a curiosity and you're just genuinely like as an engineer, you're just curious about it? I'm just uh curious. Uh but uh first no really I love science fiction. Okay. I I would I know and I think I I I feel that you're the same, okay? That that you love science fiction and and I always thought I I watch a science fiction movie or I read a science fiction book and I at the end of the book or the end of the movie this is over and you are back into your life. Okay. And this I don't like at all. Okay. I want to stay in I want that my life is a science fiction movie. Okay. I want to stay in this movie. All right. So um basically we wanted to create the movie. So it's awesome. So now we are in our own movie. So that's we are part of it and that's that's really cool. That's incredible. It's incredible man. You know thank you for pushing the edge of science. I think the the future generations thank you and we thank you for just doing it. It's about uh the future. I mean you know a lot of people uh draw a dark future based on the technology. Uh I think we should see future differently and and here consider also that technology may not be something um a lion or totally different artificial it may be something in between and bio computing may be this example of things which are technology but living just in between floating in between. Yeah, I appreciate that. I appreciate the word of optimism because I think that a lot of people look at tech progressing and it it it goes dystopian very quickly for a lot of people, but but I think they also have to realize that that's how humans have responded to new technological breakthroughs ever since we've had technological breakthroughs. But somehow we're still here. We keep going and there are humans that believe in a better world and are more optimistic about using technology to bring about a better world. And I think that's a huge message that people need to hear. Yes, you're right. At least I guess too many people forget that we'd better live today than 1,000 years ago. That's right. And the difference is thanks to the technology. Okay. We should not forget this. Okay. It's it's that technology that gave you the ability to complain about this technology, you know. Yes. Something I wouldn't want to quit you without telling you something else which is uh I find incredible. Okay. Which happened about one month ago. uh yeah five weeks ago uh you know I told you about that we are doing Python script uh for the situation okay and for uh we realized that um GPT could actually write Python script uh and with GPT agents we could actually uh let um some of the Python script be written by JPT and executed by ChPT and then we basically only have to give a prompt to ch say Okay. Modify the synthetic connections of the bological neurons by finding the right stimulations to do and measure the activity and conclude what to do next and repeat the experiment. Uh it does this. We implement it and every minute it does a new experiment. So we have this running now. It's it's it's kind of alive. It's kind of alive. It's incredible. Yeah. It's just awaiting its prime directive. I mean that that's the thing I'm I'm struggling with with chat GPT is how do you keep a prime directive in his head because it'll go off and then it'll be off on something. It's like the whole to me the whole part about the the brain that's magic is that we can stay focused on something. It'll be really curious to see how you manage to do that as well. It's very focused I tell you because uh basically uh we define a loss function. We say I want this electro to spike at this moment. So it's the objective function is very clear. Okay. We don't know how to accomplish that but we know we have to stimulate sometimes with complex algorithms that we cannot imagine but he can try and and his reasoning is powerful and the difference is that he's not going to get his knowledge from something which has already be written on the internet. He's going to get his knowledge from his experimental data that has decided by itself from trial and error. So this is totally a different story. It's it's an agent. uh it's not only uh and uh when we look at the point because we also told him to explain us what he was doing and why he was doing it and when we see the reasoning he has it's uh it makes a lot of sense for us. It seems like the non-biological robots are going to be the first ones to have to go through this cultural moment of are they alive or not before before your machines. It just seems to me like there will be Tesla bots that were all wondering do they get rights? are they actually, you know, how how much rights do they get? Be even before we get to the hurdle of, okay, now we've made organoids that are operating inside of robots or maybe we have a a combination. I think before we get to that point, we're going to have to say, are robots alive? And where are you going to stand on that? Now that you've thought a lot about this question of free will. Well, it's really directly connected to free will, obviously. Okay. Uh because if we are machines, we have no superiority to a large language model. Period. Okay. So, but this is just an assumption. And once again, I'm out of my sphere of Right. Right. I agree with you. I agree with you, though. Like, what's the difference? We seem like machines. We get rights. When robots can walk around and they don't want to get run over by a car, like, well, probably give them the right to not get run over by a car, I would guess. But we we love to think that we are superior somehow. Yeah. Just just between us, between human beings, you know. We agree on this. Sure. Secret club. We really know. We really do. Yeah, we're gonna need a handshake at some point so we can identify each other. Oh, Fred, this is awesome. Thank you so much for chatting with us. Incredible to learn about what you're doing. Yeah, thank you very much. It was a really a pleasure and very cool exchange. Yeah, that's great. All right, see you everybody. Ciao. Fred, that was awesome, man. Yeah, man. Thank you so much. You know, so much. Oh my gosh, that was so fun.