S4 | E1 | Computers Powered by Human Neurons with Dr. Ewelina Kurtys - FinalSpark

Alex Hutchings: 00:00:03

Welcome to the Think Data podcast brought to you in partnership with Mydataworks. If you want to stay up to date with the latest breakthroughs and trends in the world of data and artificial intelligence, and if you're curious about some of the strategies that companies and founders use to launch data and AI products, then you're in the right place. Our aim is to bring together a diverse lineup of fantastic guests from the founders, through to accomplished leaders and product owners at some of the most fascinating data and AI companies worldwide. They will each offer you their own unique insight into what it takes to launch and scale a great data business. Thanks for tuning in and I hope you enjoy the episode. Welcome to the Think Data podcast and today I'm really excited to be joined by Dr. Evelina Curtis. She is a strategic advisor to Final Spark. And this company, probably out of the 80 episodes I've recorded, is one of the most fascinating because they're a company that's building computers powered by living human neurons. So I'm certainly not going to try and explain what you do better than Evelina. But in simple terms, Final Spark is exploring a new kind of biocomputing that uses real brain cells instead of silicon chips. As I say, it's over to you really, because I think when we had the initial conversation and obviously the research I've done before this episode, it is a really fascinating use case and highly complex one to get your head around. So a bit about you really, first and foremost, for the listeners who's not come across yourself, it would be great if you could introduce yourself. Then maybe introduce Final Spark and what led you to become so involved in the company.

Ewelina Curtis: 00:01:58

So, I'm Ewelina Curtis. I'm a neuroscientist by background. So, I did a PhD in brain imaging. So, it's neuroscience. I was always fascinated with brain. And then I moved to, actually, I come from Poland. I lived in different countries. And finally, I came to London, to UK. And I was there a postdoc at King's College in London. That was the last time when I did research. After I moved to industry because I always wanted to see what is outside. I was always curious. What else can you do? except for research. And then I started to work, you know, with small companies initially on medical imaging because that was some topic I really knew very well. But when I was working on medical imaging, I realized that there is huge potential in artificial intelligence because I was heavily used for medical images. And then I realized also that AI is much more than medical imaging. You can do much more stuff. Very exciting. So I started to work with startups on AI. on artificial intelligence, on different commercial applications. And now I work even on the future of the AI, which is great because it's kind of combination of neuroscience and AI. These two fields, which are fascinating for me. And I work with Final Spark. And this is a Swiss startup. They try to build computers with living neurons. Living neurons, the same as we have in our brains. So usually we use neurons to study brain, to study, you know, to solve, for example, how... how to treat the diseases or how to prevent aging. But we use living neurons, the same as for biomedical research. We use them for different purposes. So we try to program them. We want to build a computer. And why we want to do this is because neurons are one million times more energy efficient than digital computers. So we want to build super ultra low energy consuming computer for artificial intelligence.

Alex Hutchings: 00:03:52

Wow.

Ewelina Curtis: 00:03:52

And we see that AI is using increasing amount of energy and today is still manageable, but it can become a problem in the future. So there are two directions, either alternative energy sources or alternative computing. So we work on the second option. And what we do in a nutshell is actually very simple, maybe not easy to do, but easy to understand. So we have living neurons and we put them on electrodes because living neurons, they often they communicate also in our brains. through the spikes. So the voltage on the membrane is changing and this can be detected as electric spike also on the electrodes. And actually people did this kind of research since over a century. It's called electrophysiology. When you put the living tissue, you put, for example, neurons on the electrodes, you can measure some spikes, some electric signals. But we do this for different purposes. We want to program them. So what we do is we send them electrical signals. through the electrodes to simulate neurons, stimulate neurons, and we also measure the response. So we measure how neurons respond. how they produce spikes. And they always produce spikes. It's generally spontaneous activity. Living neuron is always active. But they also change the patterns when you start to stimulate them with external current. So what we want to do, we want to be able to stimulate them so that the response makes sense. So this input, output, like in the computer, and that it will process information. And we know for sure that neurons can process information.

Alex Hutchings: 00:05:30

because the proof for that is that we can talk today because our neurons are working so we know for sure it's possible it's just we have to figure out how to do this yeah it's fascinating obviously you've got that heavy research background you've obviously got the phd or this is your world you're operating but obviously taking the research kind of everything you've learned through your thesis and other work and actually applying it to a you know a commercial entity or a business such as final spark or how How have you adapted to that? And when you're looking at kind of taking Final Spark and obviously taking this to market and obviously getting some adopters onto the platform, talk about that journey, because I'm guessing, you know, it's not been that straightforward because it is so advanced, you know, it's so new.

Ewelina Curtis: 00:06:16

Yes, you're right. There are many differences between doing science and entrepreneurship and engineering also. So for example, one huge difference is that we don't always try to understand everything because what is important is that it works. Because if you try to understand stuff, it will take you much longer. And it's not always useful, practically. Of course, science is different. Science is not to be always useful. It's to, you know, extend, of course, boundaries of our knowledge and understanding. And it's very, very good long term. However, if you are working in a startup, you have to be very pragmatic. And you have to get results quickly with a minimum input. So you have to maximize your results with minimum effort. So what is really important, you don't always try to understand everything, but you have to make it work. And I think this applies. I see when I actually work a lot now with engineers, they definitely follow that in engineering. When you are coding or doing some hardware, it really, you don't have to always understand. You just have to make it functioning. And I think the same is in commercial stuff. You always, you have to make it work. And don't always question everything. Because if you start to question always too much, you know, thinking, oh, why is that? then you can lose a lot of time on thinking. And that's actually, let's say, a big change when you move from academia to industry, that maybe you don't overthink so much because it also can be sometimes counterproductive when you overthink too much your action and then you will never move forward. So, and also, you know, this idea when I talk a lot with, actually with researchers, I'm often invited to speak at some conferences or universities. They are very excited, you know, to know what we do. I think it's kind of like a... Sometimes curiosity, you know, when we come from places. But still, I hear it a lot. Also, when I talk with the professors, that they say this idea is too crazy and they wouldn't do this. Why? Because in academia, you do small steps forward to understand, to be more sure what you do. But here in this project, you actually, you make a huge step forward. And if this works, this will... be really a revolution. However, that means also high risk. So that means if you are a professor, it would be, it's very difficult to publish on such idea because it can take you many, many years to make it work. And actually you risk that it doesn't work at all. So there is, of course, huge risk involved, but there are different priorities in startups and in academia. So I think in academia, you have everything more sure, I would say, more predictable. Of course, your experiments might fail. But I would say there is much less risk in this. It's much more risk when you have such a crazy deep tech idea. It's much more unpredictable. So I think that's the biggest difference. Yes.

Alex Hutchings: 00:09:06

Yeah, it's interesting you touched on that because you're completely right. When you're in the kind of the confines of research labs and you're within an institution, the risk of failure is more disappointing, obviously, the time you've invested. but in terms of a commercial entity there's financial implications there's implications it's far bigger so where are you on that journey then because obviously you've been advising them for a while now but how do you know you know where the end point is i guess is the question because like is that is that tricky to gauge yes

Ewelina Curtis: 00:09:36

it's difficult and actually that's a big challenge for fundraising because we are currently talking to investors about 50 million swiss francs investment to accelerate our research which we find really important but this is long timeline Because we expect it in the computer in around 10 years, which is, of course, an estimation. You cannot really predict everything. And this is a huge risk still. So at the moment, what we were able to do, we were able to store one bit of information in neurons. So this is very early stage. And we have done, of course, a lot of great work. We have a lot of patents. We have one publication, scientific, but that's because it's not our priority to publish. But we published one paper, which actually got really successful. It's like top 1% of paper read in the journal when it was published. And so we are very proud of this. But so we've done a lot of things. We built a lab, which is very robust, very reliable. And also we invited researchers from all over the world to work with us because doing experiment in final Spark Lab means actually writing code in programming language, Python. Usually it's Python. And this is why, because everything is connected to digital. So you have digital computers, hardware. And then all is connected to electrodes and electrodes interface with neurons. So that means you can do experiments anywhere. It doesn't have to be in our lab. So that's why we invited researchers, actually, because we hope they publish. There was recently a publication from University of Bristol, actually. They are from everywhere, but we have, I think, three collaborators from UK who get free access and just to let them play. And we hope that this will, such an open innovation approach will somehow stimulate the discussion about biocomputing. I really hope for that. And we are hoping they will publish papers. And what also is very interesting, actually, we have first clients even, although we don't have final product, but people pay us to get access remotely to our lab and to try some fundamental research on how neurons process information. So this is actually similar like with quantum computing. Quantum also doesn't work, but many people are interested to even pay to try it out so that they are ready. when it's really working.

Alex Hutchings: 00:11:56

Yeah. And on that point then, what's been the initial feedback from people? Because obviously, you know, you go back to your point where you're talking to some professors and they said this will never work, would never do it this way. On the other side, you've got people, you've got the theses, the white papers, you've got obviously a lot of interest. What are people saying when they're getting access and they're beginning to try the platform? What is some of the best feedback you've had?

Ewelina Curtis: 00:12:19

Well, actually, I'm now collecting feedback from our users. They're most of the time happy. Actually, I think I never, I don't remember anyone not happy. They actually, they love to work with us because our engineers are really great. They are quick, they are pragmatic, and we always try to support the user. You know, we know it's very difficult sometimes because to work with living neurons is challenging, especially for engineers if they don't know too much biology, but we are always supportive. And our users, I think, are very happy. Of course, the field is very challenging and we all know this. So it might be difficult to get some results. Actually, it happened that some of the users, they just stopped working on this because they didn't get results. So this is normal thing. But I think our users are quite happy because, I mean, it's not our fault that it's so difficult to work with neurons, but anything that depends on us, we try to do our best. So anything like access to the platform. You know, all the support, this I think is working quite well. And I think also that people who come to us, you know, if they are motivated to use Neuroplatform, that's how we call our lab. I think they are, of course, niche always. Niche people is still a niche topic. And these people, I think, are convinced that it's a nice idea. But when you look at the general public or general, even broadly to scientists, many of them are still resistant because it's so new that they cannot. kind of digest this idea. I think it's normal because that was also the case for me when I heard that for the first time. I thought it's maybe crazy, maybe it's great, but it could be both. So it could be either too crazy or a really genius idea. So yes.

Alex Hutchings: 00:13:58

There's a fine line, isn't there, I guess.

Ewelina Curtis: 00:14:00

Yes. And I think it's absolutely normal that you need time to process and to digest this. Many people still think that it doesn't make sense because it doesn't work yet. And it's also a normal thing in startups until it doesn't work. Nobody believes, but later. When it works, people say, I knew it will work. So we are prepared for that. So no problem.

Alex Hutchings: 00:14:20

Yeah, no, it's a fascinating point because I think, yeah, people will go, oh, yeah, I always knew that bit of success. But on the flip side, if it fails, they go, oh, yeah, I knew it wouldn't work. And talk to me about the, obviously, as much as you can about the founding team around Final Spark, because it is an idea and a concept. I'm sure when people listen to this back and look at some of the use cases and white papers on your website, they'll begin to get their head around it. obviously physically speaking, actually understand what it does. But when was that kind of light bulb moment of this is what we're going to invest potentially 10 years and a lot of money and investment into? When did that happen and who was that kind of the visionary behind the business?

Ewelina Curtis: 00:14:58

So the finance park was founded by Fred Jordan and Martin Kutter.

Alex Hutchings: 00:15:01

These are two Swiss entrepreneurs. They have PhDs in signal processing, so it's applied mathematics.

Ewelina Curtis: 00:15:07

And they've done a lot of work in the digital world. another company which is successful. So that's why they can finance Final Spark. And initially they started Final Spark as a dream of many engineers to build a thinking machine. So some kind of superior artificial intelligence. We also call it sometimes general artificial intelligence. So there are a lot of discussions about this. And they've done, of course, they started with the digital, which is natural for them, natural world. But what they realized that it costs too much money actually to build. great models in AI, you need a lot of servers, a lot of actually electricity to pay. And indeed, today, when we see the success of ChatGPT, in big part is because of the huge investment for training, which actually is in a big part the cost of energy. So they realize that for them, it's too big. It's too much money. So they realize that for a small startup like them, to make a difference, you have to be more disruptive to figure out something different. And our first employee, Jean-Marc, had already some experience with living neurons. Although he's an engineer, but he had a chance to do a project with Professor Markram. He's a very famous neuroscientist, also based in Switzerland. And because he had already some experience with neurons, they said, OK, let's try neurons. They are extremely energy efficient. And that's how it started. It started around 2018, 2019. When they started to play with living neurons, they had to learn. from scratch, the biology, so it took some time for them. But yes, we arrived to the fact that we have today a lab and we are running research in biocomputing.

Alex Hutchings: 00:16:47

Interesting. That's fascinating. And in terms of Final Spark and the vision now, obviously, you've got this kind of the conversations happening with VCs, you're beginning to try and raise capital. It is because being in deep tech and being in biocomputing, it takes a special kind of investor to... get it. So in terms of that messaging and the kind of how you're simplifying the objective, but actually making them understand that it's investable. Talk to me about that, because I think that's for a lot of people listening here, they won't be necessarily overly technical. So how are you breaking down the opportunity for a VC here? Is it about getting in early and getting into something that's generally could change the world or?

Ewelina Curtis: 00:17:27

Yes, definitely. It would be a huge revolution. And we also made some estimations that Considering the energy saving, if we build computers using living neurons, that means we will be able to build AI, which will be maybe 100 times cheaper. Wow. So it will be huge competition for digital servers. And we actually expect that even if we capture a fraction of the market, very small, we will have a billion of profit, not only revenue, profit, pure profit. So this is a really attractive idea. But of course, the challenge is that it's long term. And. This is not for every investor. It's for long-term investors. So yes, this is always challenging. And many, of course, it's normal and we respect that. Many VCs, they are focused on projects which can bring some profit within three years or maybe five. So it's not for every investor, definitely.

Alex Hutchings: 00:18:22

No. And in terms of that business model, because I appreciate that will obviously change in time as you get more involved in VCs. have their own advisors, have their own idea of how to commercialize. But what's the commercial model? Would you effectively be building the AI models yourself? Would you license the platform to organizations looking to build their own proprietary AI models? How would that work in practice?

Ewelina Curtis: 00:18:45

So actually, this is still open because it's very early. But what we imagine today is to build BioServer. This will be central computer, which will be available remotely today. Like today, you have cloud computing the same way. And we imagine it will be just competing. So let's say you could use charge GPT or some other LLM on biocomputer or on the digital. And the difference will be the price. That's what we are interested in.

Alex Hutchings: 00:19:13

Yes, because going back to the first thing you said at the beginning about the kind of the cost of AI, the energy inefficiencies. And actually, I had someone on the podcast. About six months ago, I actually was building cooling for data centers for AI because actually the amount of water that's being used to cool down data centers because of AI is going to lead to a potential inflection point with the ecosystem, really. But what would happen, in your opinion, if AI just carried on as it was and no one did anything about the inefficiencies around data consumption, energy costs? What do you think would happen with AI generally?

Ewelina Curtis: 00:19:50

If nothing we do, then there are two options. Either we figure out alternative energy sources, which is to our belief very unlikely, although of course there are some startups who believe in this. Or if we figure out some very efficient way to produce huge amount of energy, then maybe this problem will be solved. And another thing is that I become a... limited. It will not be for everyone and not for every application. Actually, today you have already this problem because sometimes people build apps and they realize later, you know, that it costs more to run the app than to really make money with that. So you cannot use AI for every problem because sometimes it's too expensive. And I've seen some articles about this, actually, that some startups, you know, they made the wrong calculation. And okay, if you use LLM, okay, It's maybe easy, but it's not. it will not be profitable because it costs too much.

Alex Hutchings: 00:20:41

It costs too much, yeah.

Ewelina Curtis: 00:20:42

Yes, and I think that problem can increase because today, actually, we know that, for example, OpenAI, the creators of ChatGPT, they are losing money. I think they're losing like, I don't remember, it was 19 billion or some billion amounts.

Alex Hutchings: 00:20:56

It's a crazy number, yeah.

Ewelina Curtis: 00:20:58

Yes, it's a huge number in billions. I remember, I checked that. And I actually analyzed this with using ChatGPT itself. Okay. how much more expensive it would have to be to become actually to make even profit to just pass the the board you know just to have any profit at all yes the price would have to be i think at least 50 dollars for now of course it depends on the usage because okay some people use more some people use less so it's a bit complicated to predict the profitability unless you limit the subscription to less tokens but Anyway, we know for sure that AI would have to be more expensive to be profitable. And that means in the future, you know, now they are still working for adoption. It's normal. So a lot of models are available for free. But you know, this money are not coming from nowhere. Someone is investing because they hope you will use it more in the future. So I think once this adoption will be really good, I think they will increase the prices. Unless we figure out some cheaper way. of making AI. If not, it will be more expensive and means, of course, not accessible to everyone. So that's a big issue.

Alex Hutchings: 00:22:13

Yeah, I'm glad you said that because I think people listening here, you know, they go on to ChatGPT, Claude, any of the models out there and they're getting instant results. We're obviously seeing a huge transition away from, say, Google, for example. Most search strings are now in ChatGPT, but actually the cost went up and a question for people to listen here. If the cost went tenfold, would you still use it? And actually, I saw a stat yesterday that they've come out, Sam Altman, and said they have a plan to get from 80 million users to 220 million users. And actually, if that happens, the pressure on the compute power, the pressure on environmental impact is going to be huge. So actually, what you're addressing now, yes, it's a five, 10 year plan, but actually. Getting this is going to be important because we can only imagine how many users we're going to have in five, ten years.

Ewelina Curtis: 00:23:05

Yes, absolutely.

Alex Hutchings: 00:23:06

That's really interesting. For people listening here, it's a question I've not asked before, but because you're so entrenched in such a fascinating space, what does a normal day look like for you in terms of the involvement in Final Spark? Because you're based in the UK, but obviously your lab is in Switzerland, your team is over there. What does a day look like for you?

Ewelina Curtis: 00:23:25

They are very different. Not always the same. Sometimes I travel for meetings or conferences or I'm working from home. Actually, I'm the only team member which is remote because I do commercial work, not really R&D so much. So that's actually the exception because generally Final Spark always wants people locally. Especially if you work on R&D, we have a lot of people who want to work with us, but it's not possible to do this remotely. Because for R&D, it's really important that people exchange information daily basis and they see each other. Well, I have a lot of... conversations with people. Actually, I would say my work is about talking a lot and I love to do this. So, yes, I talk a lot with different people. Either we have calls which, you know, are not official or we have calls like with you that we record some podcasts and we do some presentations for universities, sometimes, you know, any kind of listener because we have, there are a lot of people who are interested in what we are doing. So, yes, so I would say my work Salud.

Alex Hutchings: 00:24:26

about communicating what we do yeah i'm not surprised there's a lot of interest in you guys as i said at the introduction it's one of the most fascinating use cases and products i've come across and i recruit in this space but this is very interesting and as much as you can say

Ewelina Curtis: 00:24:42

what does the kind of next 6 12 18 months and beyond look like for you guys all depends on investors because for now as mentioned we are self-funded by fred and martin the founders so That means we have limited resources. Of course, we cannot spend too much, although we are already spending a lot, but we cannot increase too much. We are a team of six people only, so everything is going quite slowly. With investment, a lot of things will change. So for sure, we will increase the team, which is actually the biggest reason for investment. And then a lot of things can change. And what we work currently, what we're doing, we are increasing, of course, the base of our users. We are very happy that our lab is more and more popular. And our challenge for our R&D is learning in vitro. So how to make neurons learn something? We send them some signals and we want them to answer in some consistent way. And that's the biggest challenge currently in biocomputing.

Alex Hutchings: 00:25:40

Fantastic. Well, Evelina, it's been super interesting. I feel we could talk for a long time about this. But everyone who's listening here, check out FinalSpark's website. There's obviously some really interesting use cases on there anyway. Give Evelina a follow. She posts some really interesting stuff on LinkedIn anyway. But yeah, I'm excited for this to go, this to be released, because I think there's been a lot of interest in you guys. And I know when you do raise, there's going to be... additional interest to people joining but if anyone's listening here likes the sound of Final Spark reach out to Evelina but no thanks so much for coming on this morning thank you too thank you