Biocomputing & Ai, with Dr. Ewelina Kurtys

Speaker A: Hey.

Speaker B: And we are back. Welcome to the Unrivoted podcast where we talk about artificial intelligence, sometimes a little bit of digital transformation people and other technologies. Just, uh, as a reminder, this podcast is brought to you in part by our venture partner, Wingnut Investments, where tightening the roi, John, is only a thumb, thumb screw away.

Speaker A: Um,

Speaker B: yeah, I hope they don't use

Speaker C: those anymore, but, uh, who knows, right?

Speaker B: Oh, John, I have an ample supply of thumb screws, uh, as tchotchkes ready to give out.

Speaker C: Remind, um, me not to visit you, uh, over the holidays then, so I

Speaker B: want to experience it. We'll definitely tighten down on things. John, would you mind introducing our guest?

Speaker C: All right, my pleasure. All right, so after our long break from recording, uh, very excited to welcome Dr. Evelina Curtis. She is a two time founder and CEO and now strategic advisor for a company that I'm very interested to hear about because I've never quite heard anything like it before. And uh, it's almost on the border of science fiction, but I'm sure you're making it a reality. The, uh, company is named final spark. So Dr. Uh, Curtis, thanks for joining us today. Uh, we're glad to have you. So, uh, give us a little background on yourself, a little introduction and uh, fill in all the details that I didn't in that, uh, brief introduction.

Speaker A: Thank you very much. So nice to meet you. Um, I'm Evelina Cortes. I'm a, um, neuroscientist by background, so I've done research in neuroscience. I was always fascinated, uh, with brain, uh, also as a child and later I moved to industry because I was also fascinated to see what is outside academia, what is in the real world. And I started to work with startups and uh, this is how I discovered, uh, artificial intelligence. And I realized how many things, uh, how many great, amazing things we can do with this. So I also become fascinated with that. I studied a little bit, um, I even tried to learn coding, but it didn't work very well for me. Uh, but uh, still I was really fascinated with artificial intelligence and I working for some time on commercial applications, uh, of it. And uh, later I started to work on, even step farther, uh, on the future of artificial intelligence, which we believe is biocomputing. Uh, so Final Spark is trying to make first steps in this new, uh, field. There are only three companies in the world to our knowledge, which are working on this. And this is trying to build computers from living neurons. Ah, so the same neurons we have in our heads, uh, they are working quite well. Uh, this is why we can talk. So at least we can be quite sure that neurons can process information very efficiently. This we know from nature. But now we try to achieve this also in the lab by just having isolated neurons. So there is no connection to the human brain. Uh, and we want these neurons to process information. So this will be new, uh, type of computer which will be biocomputer, which will have at the heart as a processor, it will have living neurons. And this is what we are trying to do.

Speaker B: That is very awesome and we're so happy to have you here. Evelina. Um, you were nice to send me, ah, a copy of your pitch deck, uh, before we met and I went through it and I was looking at from the eye of an investor and thinking futuristic and going, okay, this is a game changer. This is a real game changer. And I think about a couple areas that maybe you can help chime in on. Uh, maybe energy and architecture, maybe you can share a little bit. Um, from your deck it sounds like, um, the bioprocessor is more energy efficient. And uh, that sounds amazing. But let's think about artificial intelligence. With the bioprocessor, will I need a small nuclear reactor in my garage to power it? I mean, give us a clue of the energy requirements.

Speaker A: So by looking at our brains, we can have idea that neurons are very efficient because we can run on banana for the whole day. And to reproduce the workings of the human brain, you would need indeed a small nuclear plant. So with digital computing, if you would like to do the same. So we can know that actually digital computing is very energy demanding. Um, and uh, the energy usage uh, by AI is increasing exponentially. And today we can still manage this, but we can expect that in the future it can be a common problem. So that's why people are trying to figure out, uh, some uh, energy efficient alternatives. And there is a lot of effort actually to improve, uh, software and also working on new hardware. But we believe that neurons are really a revolution, uh, because they are 1 million times, uh, more energy efficient, uh, than uh, digital computers. So we believe it will revolutionize the artificial intelligence. And we believe also that, you know, the AI, which we know today, which is digital, is actually based, uh, is inspired, uh, by living neurons. Uh, so we have today artificial neural networks. So this is all the LLMs or uh, chatgpt or all the actually image recognition, all the stuff. Uh, this is based on artificial and neural networks which are very energy demanding. And we believe that we could run the same, ah, stuff on the living neurons on the real Neurons and that will be much more energy efficient.

Speaker B: Okay, this is, this is really fabulous. Um, John, I'm going to, I'm going to Bogart a little bit of time here and then I'll, I'll hand over. So you celebrated um, the storing of one bit of memory, which is a first step and that's. Congratulations, that's huge. Huge. Um, when you consider I have trouble remembering my wi fi password, little things like that, um, it's going to take a few more bits to um, become interesting and useful. How far away are we from going to more than one bit?

Speaker A: Well, all depends on the uh, investment actually and on how many people will work on this problem because it's very challenging problem we assume because for now we are self funded. So we go very slowly. We are very small team of six people. Uh, but assuming we find an investor, we are currently talking with investors about 50 million Swiss francs, uh, investment which can accelerate our work. And in this case uh, we plan in two, uh, three years, uh, solve the problem of learning in vitro, which is currently the biggest, uh, the most important milestone which we have as the next one. And that means that we want to be able to change uh, connections between neurons. So the same what happens also in our brains when we learn connections rewire so that neurons can learn something. So uh, we would like to be able to process some simple information. So some simple algorithm, something very basic like maybe recognizing some image or sound. So that's, that's the next milestone. Milestone. After this, uh, another milestone will be advanced uh, algorithm. So this will be another uh, three years we expect. So we would like to match the performance of digital computing. And after we are planning another milestone, uh, which will be scaling so we want to. At the moment we have very little structures of neurons for the moment is enough for R D which we are doing. We have little blobs of cells which are around 10,000 each and are around half millimeter diameter. We put them on electrodes. This is very, very small. But in the future we would like to make very huge uh, structures of neurons even 100 meters long. Why not? Because technically it's possible. We don't have to be so small as a human brain. Maybe we can do something huge. So we are uh, envisioned that we will build bio servers. So this will be central computers, biocomputers which will be available remotely, uh, as today, uh, is uh, cloud computing. And we think it makes sense because neurons are very fragile. So they need special condition, they need a very specialized staff to take care about them. So we think that such a central bioservers would make sense. And this will be cheap computational power. And this will happen in around ten years. Uh, so we expect to be profitable in maybe ten years, uh, but we envision a really big profitability even in billions because this technology will be very important. It will be competitive to digital. So you will be able to run uh, maybe LLMs, uh, using 1 million times less energy. So that's the plan for the next 10 years.

Speaker B: Got it, got it. Last segue and then John, uh, will be teed up for you. Um, when I think about reliability, quality control and potentially uh, scaling the future of a bio cloud literally is where you're going here. Yes, you have to keep these organisms alive. So the elements of um, temperature, humidity, whatever, uh, oxygen, CO2 level, et cetera, um, it's basically a high tech spa if you think about it.

Speaker A: Yes, absolutely.

Speaker B: How do you stop from getting too comfy, uh, and make this actually this magic work? Because part of this is environmental control. I mean that is part of uh, the light, the shine on the investors, uh, to say that that's understood. But how would you handle that?

Speaker A: Uh, well, I think that keeping a stable condition of neurons, it's uh, engineering challenge is not the biggest challenge we have. Ah, so this is all doable. It's actually done in the labs all over the world because neurons are used very widely for biomedical research. So um, there is a lot what has been done about how to take care of them. So I think this is not the biggest challenge because the biggest challenge is how to program them. How, uh, because we don't know how exactly they encode information. Uh, we know that they do this in space and time. So it matters where, for example in the brain and when uh, the neurons are active. Uh, but we don't know what does it mean exactly. So that's the biggest challenge. But uh, keeping them alive maybe, um, can be fully automated because actually a lot of things in our lab are already automated. So we will uh, definitely work in that direction, uh because it can increase the robustness of the work. Uh, so uh, the biggest challenge is programming them.

Speaker B: Understand. So John, I'm going to tee this up for you. Think about the neurons unionizing and going on strike. But that's just the potential.

Speaker C: I mean if we look at a long enough timeline, 10 years is what it sounds like is the goal to reach some level of, maybe even sentience, uh, among um, you know, neurons used in a biocomputing, uh, uh, standard, which is a little bit scary. So um, you know, a few questions that you know, come to mind obviously. So you mentioned earlier, obviously that um, artificial neural networks that power deep, uh, learning LLMs, uh, basically any form of what we call M, whatever term we use synonymous with AI, uh, is driven by the neural network design. And there's a lot of different architectures for neural networks that exist besides what's used in like uh, a transformer model for LLMs. So thinking about that, how you know, neural networks were designed to mimic the function of the way the brain works in terms of electrical impulses sent between synapses. So in your design, uh, of building a biocomputing network, are you trying to use a similar architecture, uh, like a neural network? Or is that even like a question that's not even worth asking? How are you, once uh, you build the bio element, uh, how are you trying to architect them in a way to create that computing, uh, environment, uh, if that makes any sense. I don't know if I'm making sense. I said not to.

Speaker A: It does make sense actually. Neurons are much more complex. On our website, uh, finalspark.com uh you can see, uh, images of our neurons from electron microscopy, which is really high uh, magnitude, um, resolution. And um, you can see that the connections are very messy and they are in 3D. Uh, so this is much, much more complex. There are also a lot of recurrent connections which are not so common and a little bit difficult in uh, digital computing. So we can expect actually that theoretically there is much more possibilities because you have a very complex 3D structure. Also the same way is uh, our brain built, so we could expect that it has. Maybe this is one of the reason also why it's so energy efficient. Uh, because it is very complex and there are many, many more connections than in silicon neural networks.

Speaker C: Yeah, it, you know, brings up another question too of, of the comp. So if you have the complexity and, and it sounds like one of the questions right now is how do we program something that is highly complex that clearly, you know, I've never heard of anything even closely resembling this idea, which makes it very exciting to me, um, actually using uh, maybe you know, computer based artificial intelligence, um, neural networks, uh, as a means of discovering the correct or the optimal way of programming the bio computer. Is that ever any consideration of using maybe what we call AI today as a guide or a uh, helper, like an AI pair programmer, uh, in order to help determine how to program a neuron based biological computer?

Speaker A: Absolutely, because I think uh, artificial intelligence and also LLMs especially are Ah, Very helpful today. Um, I think it's, um, everyone should use them. It's risky not to use, and they can be helpful in many aspects. Also when you do research or technical work, also during coding, uh, they can save a lot of time. Uh, yes. And it's also in the direction of automation, which is a very good approach because you can have much more control on the work, and it can be done with smaller, uh, teams. So, yes, definitely, um, digital AI is a great help in building, uh, biological AI.

Speaker C: All right, now, the last. The last formal question, and we'll see where this, where this one goes. Um, so, you know, the Turing Test has been around for a long time as a means, and I think some people and I would maybe tend to agree that we've most LLMs today could probably pass the Turing Test and have some level of artificial intelligence that, uh, Alan Turing set out with his hypothetical test many, many, many decades ago. Um, but thinking about creating, uh, a biological computer now, there's already a lot of talk around guide rails for LLMs, like, you know, how do we control AI from getting out of control and doing something horrible like starting a nuclear war or eliminating humans? Or maybe those two things are in the same. The same element. But, um, thinking about, you know, and I'm interested as you, being a neuroscientist, thinking about the aspect of sentience or consciousness, which is still a very, uh, difficult topic and a lot of unknowns in the scientific community about what exactly creates consciousness. What are your thoughts about creating, you know, a biological computer that is basically a human created brain and how that could potentially affect positively or negatively, um, you know, our ability to kind of harness, uh, this. This power, you know, based off of what we're using now, silicon chips and neural networks, as opposed to an actual biological representation of a human brain, which all three of us are conscious beings, um, most of the time. What are your thoughts? Yeah, except for. Yeah, right. Martin gets like that when I ask long winded questions. But, um, so I'm interested, you know, from a neuroscience perspective, from your, you know, perspective at Final Spark, um, how you think the development of a biological computer and its connection to consciousness. What, uh, are your thoughts on. On where that pathway might lead?

Speaker A: So first you were talking also about the AI alignment, uh, like, how to avoid that it will do something against us. So I think this problem is the same in digital and biological computer. It doesn't matter the hardware. But when AI is more intelligent than us, then, indeed, uh, we have to think about how to, uh, how to teach it to be on our side. So that's one thing when it comes to the consciousness is question which we get a lot, uh, from many people. Uh, people send us emails every day with questions about our project. Some of them are about consciousness. So we are actually not experts in that because um, neuroscience, you uh, know it's still. Well, I, uh, know there are some theories, but actually consciousness you cannot prove with the scientific method. So actually the only consciousness you can be sure about is your own. And anything else is uh, actually not measurable. So this is more topic for philosophers, we believe. And actually we reach out to many to encourage them to think about biocomputing because we think, um, we hope that they will be interested, uh, to help us to answer questions about consciousness and any other actually ethical, uh, questions, uh, which always come, which are new, uh, because every new technology is bringing some new questions. So we hope that philosophers will help us with this. We actually get in touch with some research groups which planning some um, project. But it's not funded yet. So it's too early to talk about this. But we hope it will launch soon. And uh, we hope that uh, uh, that they help us to answer this question because we really want that the work which we do is accepted by society and that people are not afraid of what we do. But uh, that uh, it will be aligned with all the ethical expectations.

Speaker C: And then. I'm sorry, Martin, one more question real quick. Quantum computing versus biocomputing. What are your thoughts between uh, what one might offer versus the other or obviously have a preference toward biological computing. Which is uh, great to hear, but, uh, I'm interested in quantum computing versus biocomputing and how there might be a relation or maybe they're, you know, very closely related. I don't know.

Speaker A: So actually what they have in common is that they are both in unconventional computing. So new ways of computing which is totally different than digital, uh, because hardware is totally different than digital. But of course in quantum. It works totally different. And quantum computing, uh, is expected to be good, uh, in cryptography and faster operations, which is exactly the opposite of the neurons because neurons are rather slow but efficient in some complex tasks. So I generally believe that in the future we will have mix of different hardware. And also of course the digital will still stay, but we will have quantum. We will have, um, biocomputing many, many other. Many different chips which will be optimized for specific task. Which is actually, I think the trend which is already we can see, uh, in industry and also in academia that people are Trying to build specific chips for specific tasks. Ah. So that they will be also more energy efficient and some other efficiencies. So I believe that um, uh, there is a space for quantum, for biocomputing, which are actually uh, totally different and they are complementary also with the digital computers because each of these will have some advantages for some tasks.

Speaker B: This is awesome. Um, as we wrap up here, I'm thinking of this one last area which may be humorous, but um, it's probably a reality. If you can imagine a farm of neurotic. Neurotic neuron computing. And it's a large array, um, as a neighbor of people next door and et cetera. Is this farm of neurons likely to make auditory sounds? Are we expecting to hear cow noises or mooing? I mean, what can we expect from this?

Speaker A: Actually you can imagine that animals, when they make sound is because they have not only neurons, but they also have some organs which help them to make sounds. So here we have only neurons. So we don't build organism which can make a sound. Um, and actually it's also important that we don't try to reproduce the brain, but we make totally different structure with the same building blocks as brain. Um, but uh, when it comes to sound, actually it's interesting because we get a lot of requests also from artists because uh, this uh, this work is um, stimulating, um, imagination, also from artists. And actually they sometimes have ideas to make music based on the activity of neurons. However, it's more like, uh, you have some frequencies, uh, some activities of neurons, uh, and you can try to make music out of this. But that's uh, not because uh, neurons are making sound, but because we can uh, translate these signals to the sound. Uh, but uh, no, definitely. I know that people often, you know, when they see that we have living neurons, they always a little bit know how to say this? They a little bit make them like human. They see them as human. But we have only, uh, we have only living cells. So there is no other things of living organism. So, uh, they will not make sounds, I guess, unless we make them, uh, with hardware.

Speaker B: All right, so we're going to hook up the speakers and the amplifiers and we're going to create a concert in the future.

Speaker A: Um, yes, I think this is possible even today with the technology which we have. It's just not, not useful. And actually I can say that the music so far is not very beautiful. If you just take a raw signal from neurons. Uh, it's a little bit. It's not so pleasant to listen because I heard some samples already from some people.

Speaker B: I guess you haven't been to a heavy metal concert lately, but, uh, yeah,

Speaker C: you lose your hearing after that, so then there's nothing. There's nothing left to hear after you leave one of those. Yes.

Speaker B: Yeah, it all started when I was a tin. Anyway, moving on. Um, I want to, first off, say, Evelina, thank you for joining us on the Unrivited podcast. And, uh, we look forward to tracking your progress in the future. And on behalf of John and I, we really appreciate your time and wish you the best of, uh, good fortune fundraising and promoting bio or neurotic community computing with your neurons. Thank you so much.

Speaker A: Thank you.

Speaker C: All right, thanks.

Speaker B: Bye. Bye.