Latest / Art of the Question / Can We Build a Computer From Living Brain Cells? - Dr. Ewelina Kurtys - #18

Transcript

Can We Build a Computer From Living Brain Cells? - Dr. Ewelina Kurtys - #18 · Art of the Question

• Julius Moorman: How did you end up working ⁓ on a bio computer? So ⁓ what is your background and how did you get involved? So I did PhD in neuroscience actually in the Netherlands in Groningen, which was very nice place for that. I was always fascinated with the brain and after my PhD, I also did a postdoc in London, but I was Also very curious always to leave academia, but I didn't know what to do. And I was exploring, ⁓ considering different options. then London is actually a great place for that because you can network a lot and you can learn from people something new. So I discovered the world of startups. I started to work in a small company in London, which was doing something related to my research. but they also used a lot of AI. And this is how I learned about AI. And I became so fascinated with the topic that I left my research and I started to work on commercial applications of artificial intelligence. I learned some basics of programming and how AI works. ⁓ And later I started to work also on the future of AI, which is actually Final Spark. So this is kind of combination of neuroscience and artificial intelligence. I met the founders at one of the conferences in London and I got really excited about the project. I thought, okay, we have to do something together because it was such a great topic, neuroscience from my research, but also artificial intelligence from my industry interest. So I thought it's a really good blend for me. And this is how I got involved in the project. Very cool. And before we get into what the project entails, ⁓ what fascinates you about the brain? Why have you decided to dedicate your life ⁓ studying this magical organ? So when I was considering my PhD topic, someone gave me really good advice. Someone said to I still have options open. And after you choose PhD, you still have also, but it kind of defines you for later. And it is helpful to choose wisely. And someone told me, think about it, because I was in biomedical research, biomedical field, someone told me, think about the human body. What excites you? What is the most exciting topic you would like to study? And then for me, the choice was obviously that it's the brain. And I'm very happy. Because I think when you do research in the lab, it doesn't really matter that much if it's a lung, liver or brain, know, it's a work is often the techniques are often similar. And but I think neuroscience gives you this additional spark, which is exciting. And also later, totally on the personal not professional, I become also interested in psychology, because I realized after I left academia and started to work with people, I realized that half of the success is actually psychology. And then I started to study ⁓ for myself reading some books, you know, it's also kind of connected to brain. And it's so funny that actually now when you look at the newest research in psychology, you can also explain them with the neuroscience and it's all, you know, it's all explaining each other. So with neuroscience, can explain psychology and other way around, which is kind of actually logic because it's ⁓ It's actually looking at the same organ but from different perspective because neuroscience is more looking on let's say, chemical things which happen in the brain while psychology is looking at the behavior but at the end is the results of this process which is also chemical in the brain. So it's all nicely connected. So yeah, I just thought it's the coolest topic from biomedical research and I never regretted the choice. Yes. And actually even AI is also inspired by neuroscience. So even my work in AI somehow is connected. Even when I was working only on the software AI, that was also kind of connected to neuroscience because neural networks are inspired by living neurons. Yeah, I was going to ask that next, but I think it's also very interesting to pick a field like obviously We haven't figured out everything from the human body. There's still a lot more biology to explore, but I think, I think the brain is probably still after a hundred years of neuroscience. We've come a long way, but it's still so much we don't know about, about how it works. So I think you also chose an interesting field there with, ⁓ endless more, ⁓ things. to explore. Would you, would you agree? Yes, absolutely. think especially for clinicians, it's really a still black box in many cases. I was doing always preclinical research. So that was more on the cells or animals. never worked with humans on my research, but still, I think it's really, really difficult and still a lot of obvious things maybe we know, but they are not really explained by ⁓ science. So yes, a lot, a lot to be answered yet. What is one thing ⁓ that we still don't know about the brain, for example, can you name something that might seem obvious? But one thing I can tell you, which is also connected with final spark, this is really like Holy Grail at the moment for many people in neuroscience, but also some other fields which are inspired by neuroscience, also computer, so we don't know how neurons encode information. So basically we don't know the language. We can read the signal from neurons, but we do not know really what does it mean. So when you measure signal from the brain, you can make some simple classification. So for example, if I show you 10 images after I can detect with for example, imaging, which image you think about. So that's just classification when you have limited number of options, but I could never figure out what you are thinking about if I don't have this previous knowledge. That means I cannot decode the signal from your brain into some specific information. And this is really difficult also for computing because there are a lot of new technologies like neuromorphic computing. This is the field when ⁓ engineers try to inspire themselves by human brain and they try to use software, digital still, on the dead matter, no neurons, but still they try to make it as in the brain. And that's also a difficult question for them. And this is also a difficult question for Final Spark. We try to program real living neurons for AI. And this is really a challenge. So we know a lot how neurons process spikes, which are electrical signals, but we do not know what does it mean. So interesting. And before we get into Final Spark, I was also just interested. You mentioned that artificial intelligence is kind of modeled after the brain or at least very similar. Can you maybe explain how AI is similar to the way our brains work? So when we look at the really basics, basics, then we know that AI is actually neural networks, artificial neural networks, which actually at the end, these are just the mathematical equations which are connected to each other. But people like to visualize this in such a way that you have neurons, nodes connected and this is kind of, it looks like a neuron. Although neurons in nature, they are much more messy. They are not so nicely proportional, but the way how we usually draw it for artificial intelligence is very organized, but this is inspired because you have, it looks like neurons all connected to each other. ⁓ So yeah, so that's the reason why it's on the basics inspired. And also actually there are a lot of things which you can learn from brain. and implement in digital also. even I know that we heard this recently that ⁓ Yann LeCun, one of the fathers of artificial intelligence, he now believes that large language models is that alley and that at the end we have to build the models which are more inspired by neuroscience. So for example, which will learn more in the way. So it's not only about this architecture of neural networks, but also that the algorithm will work more like a brain, like for example hierarchical. So brain is always dividing information into hierarchies. So yeah, there are many things and also there are many things which we maybe today can try to implement. Like for example, people try to make a chip where you have memory and processing closer to each other because in the computer it's far away. Let's ⁓ farther than in the brain, in the brain you have in one place memory. and computing. people try to do chips, new type of chips where this is closer, you use less energy. So, there are many little steps you can do to learn from a brain. And I think ⁓ more we know about neuroscience, more we can implement also in the software. So, these are really a lot of little things which can be introduced. Yeah, and it's interesting what you mentioned there about the potential limitations of artificial intelligence. I really want to do a separate episode on that topic because I remember when CHED GPT just came out, it was just this watershed moment and everyone was blown away by its capabilities. And then GPT-4 came about and we were freaking out because We were like, okay, if this is going to continue like this, if we just feed it more data, it continues to become exponentially smarter. And for a while you saw all these people ⁓ kind of sounding the alarm bell and AI is it's dangerous. This is going to be the end of us. But then when GPT-5 was released, the differences weren't as good anymore. It seemed that we kind of like hit a plateau at least. from the outside, ⁓ a user perspective, that's what it seemed like. And then, yeah, I feel like most of the improvements were regarding fine tuning and regarding direct implementation. So for example, that it was able to directly work with your Gmail if you're using cloud, for example, but those like exponential growth in the smartness, at least how it feels. to me from, from a user that doesn't know much about how AI works. It feels like we kind of hit a, yeah, sort of a plateau there, but it's something I'd like to do a whole ⁓ separate episode about, but I was just like, now that we're talking, is that something that you share this, this opinion or this feeling, or do you think there's still a lot more progress to be made the way we're currently using and scaling LLMs? To be honest, I don't know because I have no idea how much more we can do with the large language models. I think this is really deep expertise. What I can say from the user expertise, I have the impression that ChalGPT is really getting better. For example, you don't have to worry about the prompt so much. In the past, we were talking about prompt engineering. Today, I have the impression it doesn't really matter that much. I mean, of course, you have to give essential information, but you don't have to worry so much about the style, for example. And I see it's kind of like guessing a lot of things much better than before. So yeah, I think it's still impressive, but how far we can get with the chat GPT, I really don't know. Important limitation here is not only what technically can be done, but also the cost, because it's really expensive. So maybe it could be that ⁓ some of the creators of large language models like OpenAI, might decide, you know, maybe to grow slower, just because it costs so much money. Because today, the better model, the more money it costs for servers. So that's also the problem we are solving with Final Spark because living neurons are one million times more energy efficient, comparing to digital. So I think limitations of child GPT and other large language models are not only what is technically possible, but also the cost. Maybe sometimes ⁓ some improvements are not worth the cost, but I don't know. This is really a speculation. It's a good segue into, ⁓ we're finally going to talk about Final Spark. So for the people that don't know what Final Spark is, can you explain to me what you are doing at Final Spark and what the company is doing in a broader sense? Yes, so I'm a strategic advisor at FinalSpark and I work on the commercial side. So that means I left the laboratory and now I work on anything that is related to our external connections like investors searching for some maybe collaborations, partnerships, or most importantly, we have also growing numbers of number of clients because our laboratory is available remotely. So this is very special and we are very happy with this. So we don't have final product yet, but people can connect to our lab and this is already working kind of product. So this is cool. And what final spark is doing. So final spark is building computers from living neurons. That's the objective of the project. So this final goal we haven't achieved yet. We don't have yet the computer, but we have a laboratory, which is very robust and you can run there high number of experiments very reliably. on how neurons process information. So how they process spikes, these electrical signals. And in our lab, you can programmatically stimulate neurons with electrical or chemical signals. So you can send them electrical spikes, it's just electricity through electrodes, or you can send them chemical signal like for example, dopamine or serotonin, because this is also how neurons are communicating in the brain. So that's what we do. And we try to program them, we try to make them some tasks. And the final objective is to make super biocomputer, which will be one million times more energy efficient compared to silicon and will be able to replace silicon in some tasks. Of course, not everything because I generally believe the future will be a lot of different chips, a lot of variety of computers. ⁓ But it will be a cheap alternative for silicon for, for example, generative AI. because neurons are quite good in analyzing complex problems and generating ideas like in our brains, but it costs them much less energy than silicon. So we believe that will be good use case for neurons, but we expect it will take us around 10 years to build the computer. So it's still long timeline. And we might also discover some other applications on the way. Yeah. I remember just first hearing about final spark and just the concept of biocomputing and I was just like blown away. It seems like something straight ⁓ out of a science fiction movie. And what really helped me understand it is to have a sense of understanding, basic understanding of, how the brain works. can you maybe elaborate a little bit about. ⁓ what's actually happening inside our brains when we process any information or formulate any thoughts. And then I'm talking mostly about the electrical signals and the chemical signals that we receive there as well. Yes, so neurons are processing information differently than silicon. In silicon we have zeros and ones and logic gates. Neurons are totally different. So it's different substrate. So they process spikes, which is electrical activity. on the surface of neurons and also they can be stimulated by chemical signals. That's why in our lab we have both chemical or electrical signals and neurons basically save information in the way how they connected. there are many, of each neuron is making many, many connections with another one. So they are all forming such a network, which is very complex and If neurons are well connected, that means they are more likely to exchange signals. And this is also memory. So they process the spikes and this is processing of information, but the connections is the memory. So that's in a nutshell how it works. And also there are many things about the brain and neurons. For example, they have a lot of recurrent connections, which are possible in digital, but much more difficult. They also have hierarchical way of analyzing information as I mentioned before, which also saves energy. ⁓ For example, when we see images or especially when we see moving ⁓ environment around us, like any kind of movie or when we go by car, ⁓ our brain is registering what is changing. While computer ⁓ in a standard way is analyzing every pixel with the same importance. We have more efficient way. So yes, there are a lot of ways how neurons process information more efficiently, but on the heart of this is totally different. Like we don't have logic case, but we have spikes. So these are basically the signal is encoded in time and space. So that means that neuron is important when and ⁓ in where it's active. This is an information. So in the brain it matters where exactly and also when we have neurons in the lab, we have 3D structures in our lab, which we put on electrodes and we communicate with them in both ways. So we can send them signals and receive signals from neurons and it also matters where and when they are active. So this is temporal data. Yes. And from my memory is that our brains use around 20 Watts of power, which is comparable to an iPhone charger or something. And with this very small amount of energy, we are able to do everything. We're having this conversation. We were allowed to play sports, to catch a football, to drive, to jump, whatever we need to do. which is relatively low energy when you think about it, because if we were to simulate such a scenario with a computer, we would need like a million times as much energy, correct? Yes, absolutely. That's correct. Yeah. No, okay. I still remember pretty well. So then the idea of ⁓ a biocomputer is that we somehow use this efficient way of computing and somehow, ⁓ yeah, reenact those signals in a petri dish with something you call organoids and then hopefully be able to translate that ⁓ signal back into something useful. So if we ⁓ focus on the organoids ⁓ first, like What's actually, does the bio computer consist of? What is it and which cells are used? How do you grow these? So we get the cells from human skin. So basically this is already commercialized method. So you can buy ⁓ commercially stem cells, which are derived from human skin. ⁓ And you can. theoretically produced any type of cells out of them, any type of human body, human body, they are human stem cells. And this is a method for which was awarded the Nobel Prize 15 years ago. And this actually enables us to do this kind of research because before you had to, you had to kill often animals to get neurons from the brain and this is still done for research. Sometimes it's necessary, although it's ⁓ of course less and less being used now. But we wouldn't be able to do this for ⁓ a ⁓ biocomputer. So yes, so we get huge number of neurons quite easily, quite cheaply. The processor is already commercialized. So many people ask for that. So yes, the consent of donors and everything is sorted already by provider of the cells. And yes, we are very happy. Because of this, we can produce huge amount of living neurons in the lab. So we have a biologist, Flora, in the lab now. We started without biologists, so that was a little bit robust, but now we have real biologists and she can do something what is called differentiation. And this is the process when you can get some specific cells, cell type from stem cells. And then we put them on the shaker so that they connect to each other because this is something natural what happens between neurons when you put them close to each other, let's say, and they will connect. This is just biology. So you don't have to do anything. And they will naturally form this 3D connection. So they form a neurosphere. This is ⁓ around 10,000 neurons in each neurosphere connected in three-dimensional. Usually we use cortical neurons, but we are also experimenting with other types of neurons and other types of cells which are also in the brain, like for example, microglia or astrocytes which are taking care of neurons. to make sure that they are healthy and happy. So yes, we do play with different combinations because this is really big problem for us to solve still. We are not sure what is the perfect cell composition for our biocomputer and there many, many types of different neurons in the brain. So yes, we are also experimenting with different versions. Yeah, and these cells, these stem cells, they are turned into neurons and if you put these neurons together, they form this sphere basically that happens automatically. You don't really have to do anything for that. then you keep that, what did you call it again? The shaker? Well, yeah, but the ball itself, is it the organoid? Organoid, organoid. Yes, okay, so the organoid sits in this petri dish and then it's connected to something that allows you to send it electrical signals and ⁓ chemical signals, right? Yes. Yes. So electrical signals we send through electrodes and chemical signals we release to the medium and medium is this specific special liquid in which neurons are partly immersed so that they can get all the nutrients, oxygen, everything what they need for living. Yes. Okay. So that's how you basically communicate with the organoid, right? That's how you, that's the plan is to, if you want to give it an assignment or a task to compute, that's how you stimulate the organoid in order to ⁓ hopefully have it produce some sort of signals that we can then later translate, right? So how does the translation look like? What are we looking for there? So this is actually the challenge, which is not solved yet. ⁓ So we generally search for patterns in the signals for some relationship between what we send to neurons and what we measure outside as an output. But we still don't know how ⁓ neurons encode information, so we don't know how to decode the signal. So this is still an area of active research. But yes, we generally search for patterns. So there's a lot of signal processing in this, analyzing signals and trying to find some relationship between input and output. Yes, very cool. let's assume for a second that we figured out all problems connected to, to biocomputers and that it works perfectly. Um, can you take me step through step? What happens if I give it a task, for example, to calculate a 10 times 10, for example, I asked the biocomputer, what is 10 times 10? What are all the steps that are happening? throughout the entire system before I get my answer. And I know it doesn't work perfectly yet, but we will just ignore that. In this scenario, everything works perfectly. Yes. So in the future, when it works, you will probably have interface, which will be digital as we have today in our lab. So basically today you communicate with neurons by writing Python script or by writing prompts. Like to the large language model. Yes, because it's also connected. So I think in the future will be similar. Maybe, who knows, maybe it will be prompt based. So basically you will write some kind of task on the chat, like please multiply 10 by 10. And then this will be sent to our hardware. So basically today you just send signals. You don't send words or numbers because it's too complex at this stage of the development. So today you say just, I'll stimulate neurons with this kind of signal for this kind of amount of time. But in the future will be the same. We just will be able to encode the numbers and the words into signals. So at the end will be always the same. You write on your screen computer, which is still digital, you write a task and this will be translated to the signal, electrical or chemical or both. So neurons will be triggered. And they will process inside. So inside the real living network, they will process this information and they will produce an output, which is in form of neuronal activity, which will be measured by electrodes. And this will be sent to digital computer, analyze and decode it into your answer. So today we cannot encode and decode. So basically we just exchanging signals. You send signal, you receive signal, but in the future, you will have a question, question will be translated into signal, output signal will be translated to the answer. So yes, so this is in a nutshell very simple. And of course we will have a lot of layers. So like today, you know, you don't need to know Verilog or other very low level ⁓ programming language in order to program computers because you have high abstract layer. So we also are aiming for this, that There will be abstract layers so you don't have to really understand how neurons work and you will be able to program them. And actually the future probably will be prompt based, so no more coding as we know it for today still. Also for digital computers, maybe probably we will be just saying with words, maybe not even typing. think speaking will be as much easier. I already talk a lot to chat GPT, so I think this is the direction. And because it's faster. So I imagine the future in such a way that we will be giving prompts by saying them most of the time. And then computer will ⁓ analyze them or a bio computer or other type of computer. So that's how I see the future. And we already see this clearly. Yeah. And I think if we make the comparison to artificial intelligence, I think it works, would work very similarly for, for, for people to understand. Like if you currently ask JGPT a question, you need to be connected to the internet. So the only thing that is going, that's happening is you're basically sending a request to open AI servers and all the computing power and the calculations is happening locally. at the level of the servers of OpenAI or Anthropic if you're using cloud. So your computer isn't actually doing any of the calculations, assuming that you're not running the model locally, which most people aren't, of course. So in that sense, it's similar and more like a cloud-based solution. That's what we believe because neurons are very fragile. it will be probably possible to build something ⁓ maybe some edge device or portable computer, why not? But we believe more in bioserver because we think it's easier to build. We also can build something huge. We think of building some structure will be 100 meters long, even of neurons. So yeah, we think it's much more exciting to build huge bioserver. So indeed it will work in the same way as today cloud computing. So that will be bio cloud. Cool. And of course, ⁓ we spoke about the energy efficiency of neurons and how our brains are like a million times more efficient than a computer is. of course, if we are using this in a way like, of course, most of the heavy lifting is done by running the actual computations, right? That's why AI is currently using so much energy and why people are worried about the water usage. It's because locally at the server level, that's where the heavy compute power is. But of course you're still do making the request from a computer and also the bio servers have to be called. They have to have maybe some sub staff. So there is some energy being used in order to keep them alive, I assume. Yes, absolutely. You don't have to cool them. Sorry, I interrupt you. Here is actually a different type of problem because yes, indeed, silicone servers, they heat up a lot, but neurons are actually, you need to keep them at around 37 degrees Celsius because this is body temperature, so this is best for neurons. So yeah, so actually you don't have this necessity to cool them with incredible amount of water as it is. for digital server. Yes. But yes, the app cost is in the life. You are right. And we take this into account. And even if you take into account this cost, we made some hypothetical calculations based on our current cost of the lab. It still will be much cheaper, at least hundreds times cheaper to keep neurons instead of silicon. Yeah, that's incredible. And I mean, if you just put it in, in The desert of Dubai, example, you don't really need much energy to warm them because 37 degrees seems pretty regular there. This idea of a biocomputer, is this something that is completely new or when did this idea first arrive? Because it sounds so incredible. So basically, final spark was founded in 2014 by the Jordan and Martin Kutter. are two Swiss entrepreneurs. have PhDs in signal processing. So this is applied mathematics and they had the dream to build thinking machine. And they tried of course, fundamental research in digital AI, but they realized that it's so expensive. So they decided to be more disruptive, to try to be more disruptive. Because as a small startup, if you really want to make a difference, you have to do something different. Because digital AI is already a chase of ⁓ budgets. So it's not so much ⁓ for startup, at least if you want to build a superior model. So you need a lot of money. And indeed, when we see the success of large language models, this is in big part due to the huge cost which were spent to build them. So, yes, the founders of Final Spark decided to go to try living neurons. Our first employee actually, Jean-Marc, he had some experience working with Professor Markham, which is a famous neuroscientist in Switzerland. So they thought it's a good starting point. Let's try neurons. They will use a little energy. And it seems to be crazy at that time. seemed to be crazy. ⁓ but yeah, there are other people also in the world. are happy to have some competitors which proves that the idea makes sense. ⁓ probably, or maybe not proofs, but, ⁓ let's say it increases the chances. If you would be the only one who have this idea, then maybe not. ⁓ but because there are other people as well thinking now, and now it seems to be really more and more popular. So it's really, it's really interesting how it grows. Of course we live in the bubble, so for me, okay, it's very popular, biocomputing, but I know for general public not yet. But we see more and more awareness. And actually when you put in Google Finance Spark, you find like thousands, thousands, like maybe 20,000 of hits. We also contributed to this, of course. We try to promote the topic. But yes, I think it's getting more and more traction. It's interesting how it's growing. So yeah, I would say it started 15 years ago maybe. Actually, one of our advisors, Steve Potter, he did some kind of similar research on living neurons already 30 years ago. ⁓ Not sure he was really like directly talking about computer because that was more to understand how neurons process information in the brain. But that's kind of like basis also for our work today because the questions are similar. And also what we do today can be useful for neuroscience or medicine. Because when we understand how neuroscience called information, we can build computer, but we can maybe also cure some of the diseases. Or maybe we can upload information into brain so that you can learn your language. So yes, so I would say it's very young, very young field. Yes. And compared to the, from when it was just an idea to today, what were some of the significant breakthroughs that were made over these last 15 years? Because I remember when I visited the laboratory in Switzerland a couple of years ago. ⁓ It was not from Final Spark, believe, but there was this video going around of ⁓ the biocomputer playing a game of Pong, for example. know Final Spark had the demo with the butterfly that you could play, ⁓ moving the butterfly. more recently, I also saw a YouTube channel ⁓ made a video where ⁓ Yeah, neurons or human brain cells were playing doom the video game. So I believe they're there. Yeah. I don't know if you've seen that one. But those are some examples to, yeah, seeing this idea working or at least signs of it working. so what are some of the breakthroughs of these past few years? Well, so I would say the progress is quite. basic and it's maybe hard to understand for general public because there a lot of technicalities. So well, we built a very robust laboratory, which is now used all over the world by researchers and also companies who want to try this cutting edge technology, which is still in the making, but people are interested to know how it works. To have some idea and to do some exploratory research. So that's a big success. We also publish a paper in Frontiers, technical paper, which ⁓ is ⁓ top 1 % of the papers read in this journal. So we are happy. For that, we also have two patent families and a lot of, as I mentioned, a lot of hits on Google. So I think we increased a lot the awareness about this new field. ⁓ For the technical part, we managed to store one bit of information in our neurons. We also were able actually recently to do some closed loop experiments with large language models. So today you can employ a chat GPT or some other model to work in your lab because you work with living neurons by just writing coding Python. So who said human have to do this? It can be done by AI. So today we can set up AI for closed loop experiments. So AI is writing a script. in Python programming language, the script is executed. So it means neurons are stimulated physically in the lab. And then the answer from neurons is measured and it's sent back to the large language model, which is analyzing the results and adjusting the script. So maybe it will be more likely to achieve some goal. And so far we managed to increase or decrease the activity of neurons with that way, with this closed loop way, which is very basic. But that's the stage of the development of the technology which we have today. So what we try to do is really difficult because we try to do general computing machine. We don't try to make some specific task work, like for example, some game. We want really to figure out the secret for general computing and this is very difficult. ⁓ But yes, we're making small steps towards and I think Generative AI is very helpful to speed up the research. Yes. Yeah, no, can imagine. And what ⁓ is the biggest problem that needs to be solved at this moment? If you had to pick something, ⁓ what would you need to solve in order to turn this into a consumer product, for example? Learning in vitro. So how to make a neurons learn on the dish? So this is again, it comes back to the problem, which we already discussed earlier, encoding information. So we have to know the language because we know we cannot code neurons in such a way like we code digital computers. So we need to find a totally new language of programming, like totally new approach of programming. And this is difficult because this requires change in thinking and some new framework. This is always the hardest part. So, yeah, so this is the hardest that will be learning in vitro how to be, how to change connections between neurons. Yes. So that they can learn new information and perform some tasks for us and that we can decode the signal and encode the signal into neurons. That would be, that would be, that, that will change a lot in biocomputing and also neuroscience. Yes. Yeah. Because in neuroscience you mentioned there, is that even understood how we do that in our current in our own brain? No, no, no, of course not. Because neurons are always the same. if we knew this for brain, we would also knew this for in the dish. So no, we don't know what exactly the spikes mean. Like from the brain, you can guess based on the location, like type of activity, but no, you cannot like read your mind. Nobody can read, but I'm sure it will be possible in the future and not far away future. What makes you sure that it's possible? there any indications that indicate that? Because somehow our brain is the best proof that neurons can process information ⁓ in a reliable, repetitive manner. because we have different brains, but we all speak the same language. So somehow it can be encoded and decoded. So it's just a matter of time when we find it out. So there is no doubt that neurons are working, that they can process information, that they can encode information. And this information can be decoded also into something we would understand with our eyes or ears. ⁓ So there is no doubt. And why I think it will be done soon is because, well, I think there is a lot of technological progress which allows this research. For example, the fact that we can use neurons from human skin. ⁓ means that we can do a huge number of experiments today, which was not possible 20 years ago. And also I think artificial intelligence, generative AI has also a role here to accelerate the research. think a lot of things can be done faster now. Also in biocomputing because of AI tools. So, and there so many people also putting effort on these questions, ⁓ including us, but not only. So I really believe it's a matter of time. Hard to say how much, but yes, I think it's not that far. I think maybe it will be faster than quantum computing, but this is a longer discussion. also not an expert in quantum. So yeah, but I think within few years we will know for sure more about how neurons encode information. Yes. Yeah, it's super interesting. if we, you briefly mentioned it there already, but if we crack this code, if we find out how we do encode information, you'd say it might also be possible to like download languages into the brain potentially. Yes, theoretically yes, but it's a really a theoretical idea. It's not the objective, the final spark, but it could be a side effect of our work. Yes. Yeah. Super cool. And ⁓ well, of course ⁓ we are talking about, ⁓ yeah, the organoids are alive. How long do they stay alive? They usually stay alive around three months. But we had some success of nine months longer. We know for sure in nature they can be alive even hundred years in our brains. So it's just a technical challenge how to make the condition in the lab so comfortable. as they have in the brain so they can live so long. But for now it's around three months. ⁓ For some people that might sound a bit, how can I say it, something that is alive, some people might be worried, are these organoids, could they potentially be conscious or if they aren't conscious right now, if they get developed enough and smart enough to do all these complex computations in the future, ⁓ would that potentially also mean that you've created a living brain in a petri dish ⁓ in a sense? if it is that, what would that mean to you practically? So yeah, I personally believe not, but I don't generally believe that machines will be conscious. But there many people who do believe this. And what I can say, we actually reach out to many philosophers. ⁓ I was even at conference in Netherlands, I think two years ago, ⁓ to try to encourage philosophers to work on this topic. And we really hope that soon there will be some papers written by philosophers about this, because we are of course not really ⁓ experts in this, this humanistic and ethical aspects. We are very technical people, but yes, I think it's important to ask these questions and ⁓ search for answers also for digital AI. Yes, they are interesting, interesting questions. So, yeah, but only thing I can say that we don't, because many people say that we make a brain in the lab, which is not true, because we just use the same building blocks as in the brain. But what we make is different type of structure, we never will be able to ⁓ reproduce the complexity of the brain and that's not the purpose to build a brain in the lab, but we use the same building blocks. yes, some people believe in consciousness and that's why it's good to engage philosophers on that. We don't have any scientific method to do this. It can only be considered theoretically by philosophers. Yeah, I've had I had Henry Schevlin on the show the other day and he was recently hired by Google Google DeepMind. He's a philosopher, fantastic philosopher, ⁓ specifically interested in AI consciousness. ⁓ So very similar topic and I had such a great discussion with him about it. And ⁓ yeah, the interesting thing about consciousness is that even in the field of philosophy, isn't a singular definition of what it means to be conscious. According to some people, consciousness is like a switch, like it's binary. You either are consciousness or you're not consciousness. And then there's also people in the field who believe that consciousness is a spectrum so that you can be a little bit conscious or a lot conscious. So, ⁓ yeah, it's definitely, yeah, it's probably one of the hardest questions in philosophy that hasn't been solved. yeah, because, yeah, consciousness is also a very personal experience, of course. So it's very hard. Like there is no scientific method to prove consciousness, basically, ⁓ which is a problem. ⁓ But I thought that was super interesting as well. Can you talk about, ⁓ because I think this is really cool about the collaborations you have with the universities because ⁓ it currently already is accessible through the cloud basically or from a distance like the laboratory that you have in Switzerland. People can access it from the United States or anywhere in world basically. Can you maybe elaborate on that and ⁓ what the universities are doing and how you are collaborating with them? Yes, so everyone has different ideas and people come to us with ideas so we never tell them what to do. And yes, you can connect to our lab remotely. You do experiments by writing Python script or writing prompts on your personal computer wherever you are. And this is possible since a long time. We had early adopters which were universities. Indeed, we started with that. We selected eight universities out of 34 who submitted the proposals, ideas for a project. So from that, we chose some early adopters and the first group already published an article recently, which is also on our website, finalspark.com, so you can check it. And they have published A paper where they try to encode like the letters for the blind people on the Braille, which I read from the signals from the robot encode to the neurons. Of course, this is still hard encoding. So it's not like how neurons really work. It's just, we make some kind of assumption that, okay, this signal will mean this, another signal, something else. But still it's a great example that you can do something meaningful with neurons and we are very happy they published this. And there a other papers in the making, for example paper about connections between neurons which are very, very complex but you can somehow mathematically approximate these connections. Like based on what you measure on the surface of neurons you can see how they are connected inside this round organoid. ⁓ so yes, we also started to have paid, ⁓ commercial clients, which is really nice for us. So companies and big and small started to write to us and, ⁓ we started to rent the access, ⁓ to our lab, which we call. ⁓ neuro platform. And today you can subscribe our lab for a one month. a minimum and you can get access to real living neurons and you can run experiments in this. So you don't have to worry about biology. But, let's say all this hard work, which is to keep them alive and to build all this. But you can directly run experiments on living neurons. Yes. Super interesting. So basically by building this laboratory and having this remote access to the laboratory, you allow other universities and other companies now to do science in your lab from across the globe. Yes, absolutely. That's super interesting. And ⁓ do you feel like the amount of papers on biocomputing and the interest is growing over the years? Because I feel like it is. Yes, definitely it's growing. Biocomputing is still a very diverse field. Some people don't really work on living neurons. Some they work on DNA, for example, computing, which is also called biocomputing. So yes, there's a lot of variety in the field, but definitely we see increasing popularity of our work of trying to program living neurons. This is definitely a big change. A few years ago, nobody heard about this. Now more and more people are excited and want to try and they are aware of the research. Yes. Yeah. Is there something ⁓ most people get wrong about a wetware and biocomputers that when they hear that they have a misinterpretation about it? Yeah. The biggest thing I would say is the fact that people say that we make brains, which is not true. We use the same building blocks of the brain. think in that also, you know, stimulates all this discussion about consciousness, which I think is just anthropomorphizing the matter. So yeah, yeah. That's, think, the biggest misconception I hear. And even scientists are saying this often, which I think it's very bad because it only increases this... I don't know, it can bring wrong conclusions to people about what we do. yeah, I think we shouldn't say that. No, no, no, no, can imagine. ⁓ when ⁓ would you ⁓ call it? Let's say in 10 years from now or maybe as long as it may take, ⁓ when do you think ⁓ that this is a success? Well, for a startup success is when it becomes profitable. So actually, we might become profitable before we do the computer because we already have a lot of clients who use our lab. So that's an important metric of success. And another metric is to build a computer. So to build a real bioserver where people can connect and perform real tasks, which will be competitive to the silicon servers. Yeah. And you mentioned earlier that currently it is able to do one bit of information to encode one bit of information. In the future, is the size of the organoids going to grow or are just more organoids going to be connected to each other? No, think at this stage it's more understanding how we should stimulate neurons and how we should decode information. So it's not so much the problem of size. I think for this level of research, which we do now, this small organoid is enough. We believe in the future when we learn to program them, it will matter the size and of course, maybe type of neurons and many, many things and numbers of connections like also in our brain it matters. But for the moment, I don't think it's for this fundamental research which we are doing. I think it doesn't matter that much. Cool. And now what's next for Final Spark? Is there anything in the pipeline that ⁓ Yeah, that is exciting to talk about. Well, we want to ⁓ increase the number of users of clients, which we have. I think this will ⁓ help us to go further, to move forward because we have still limited resources. know, it's a small startup, only six people. So, and we are constantly developing our technology further. So we are making new patents, new original innovations. So this is important. So yes, we want that more people know about the project, that more people use our lab and that we invent more and more solutions related to biocomputing, which we have done already a lot of solutions in the lab, which are how to make this biocomputer better, how to keep neurons alive for a long time and things like this. yes, so we have a lot of technical challenges which are solving and we also have commercial challenges, which how to become profitable. So yes, it's still long way, but yes, we enjoy that. So exciting journey. Yes. Yeah. I mean, if you were looking for, if you weren't looking for a challenge, you would have probably picked a different profession because it seems you're literally trying to solve one of the hardest problems ⁓ we currently have in science. So I think that deserves a lot of ⁓ praise. Before we wrap it up today, is there anything we haven't discussed that you think is still worth mentioning or would be interesting to mention? So what I would say at the end is I encourage everyone to check our website, finalspark.com. There many ways to connect with us. We have also Discord channel, have newsletter which we send every six months, it's not often. But if you want to be up to date on the research. Then it's very good idea to subscribe. If you are a technical, especially person, it's good also to join the discord. We have a lot of resources, which are online for free. What is also interesting, recently we have developed, well, we have prepared a data set, which we're sharing to people. Whoever wants to get it, can get it for free and they can analyze some data offline, of course, for the... live connection with neurons, you need to subscribe to Neuroplatform, but ⁓ offline data you can get for free. And we have a lot of researchers who did already some analysis on them. So yeah, I think this is really good start also for any engineer who would like to start with biocomputing or just try. ⁓ I think this is good. So yeah, we have a lot of resources, also a lot of blog articles on our website. ⁓ Not everything is technical. Also you can read about the future of computing, how we see this. So yeah, so I recommend to check finalspark.com. I think this is good starting point if someone would like to know more. Yeah, I will definitely put it in the show notes. One more question regarding the offline data. So that is everyone can download that for free and they can basically see ⁓ what the input was and what the reaction to that input was, for example, so it was maybe an electrical spike or dopamine or serotonin, whatever, and then they see what the reaction to that is. Yes, absolutely. And then they can try to analyze this search for patterns, which is down the biggest secret or try to visualize the data, anything, anything. Yes. And this is like, if you would do experiment and after the experiments you have data, you can analyze. So this is just after the experiment. So yeah, which was done by someone else. But yes, this I think is really exciting and this is very new. So we don't share this data set since long time. So I think this can even be used for a publication because it's something and something new and not many people use this kind of data. So yes. Awesome. Well, I really appreciated it, Evelina. It was very interesting to link up again. ⁓ yeah, it's incredible what you guys are doing and what you are trying to solve. ⁓ Say hello to Fred for me if you see him. yeah, we'll be in touch. ⁓ yeah, like I mentioned, I will put everything, the website and where to reach you in the show notes. So. Thanks a lot. was amazing. Thank you.