Computer from brain cells – how FinalSpark creates living AI | Ukrainian News

"It won’t escape from the lab," assures Fred Jordan, co-founder of the Swiss company FinalSpark. Over the past five years, FinalSpark has invested more than $3 million in the development of a "living computer" made from living human brain cells.The artificial intelligence based on the bioprocessor could cost consumers at least ten times less than traditional models like ChatGPT, which run on GPUs and CPUs, according to FinalSpark. The reason is – significantly lower energy consumption.So when could the living AI become a reality? Where is the company looking for investors? Fred Jordan, who runs the company with Martin Kutter, discussed these topics and more in an interview with LIGA.net.What does "living computer" mean?Scientists Fred Jordan and Martin Cutter founded FinalSpark in 2014 to develop a "thinking machine" using traditional hardware. But after several years of research, they concluded that the best machine for processing information already exists – the human brain, so they decided to create a computer based on human brain neurons. A "biocomputer" or "bioprocessor" is a cluster of human brain cells – mostly looking like a sphere or "neurosphere" – with electrodes connected to it. Through these electrodes, a person can stimulate the neurons to perform required tasks. The neurons used by FinalSpark are not taken directly from the human brain. They are made from stem cells, and the stem cells are obtained from human blood cells. Currently, the company employs six specialists, including two co-CEOs, and three more serve as advisors.Pros and cons of living computerWhy did you bet on human neurons, considering that even a simple calculator can solve a mathematical problem in a second that no human could ever calculate in their head? Where does this faith in the ordinary human neuron come from? We don’t aim to make computations using biological neurons. We are targeting a specific type of operation, which is used by AI. And most of the AI today works by simulating neurons.We want to replace those simulations of neurons with real neurons. It only applies to AI. Not cryptographic computation, not Windows – this is not going to work well on a biocomputer.However, processing information, like what you are doing right now with your brain, could be replicated or simulated with an artificial neural network and could also work with biological nervous system networks.It’s connectionism. When you process information mathematically, you have units of processing which are connected together. And this concept set up like 50 years ago and it's the underlying concept of neurons whether they are biological or artificial.What are the advantages and disadvantages of living neurons compared to artificial ones?The main disadvantage is speed. Cortical neurons will always be slower. Because biological neurons work with an actual electric potential, and a spike generated by the neuron. Communication between biological neurons takes 2 milliseconds. Transistors can simulate neurons much faster.So biocomputer won’t be good at a processing of images in the Tesla car, where you have to process information very quickly. Or analyzing faces of a crowd in real time with speed like 1 milliseconds per one identification. But for AI companions, where AI slows down to time of real human thinking, it’s okay. Or generating images: it doesn't matter if the image is generated in 1 millisecond or 1 second. So for generative AI, we can do at least the same quality of work as classic AI.The main advantage is energy consumption. To simulate one biological neuron you should use thousands of artificial neurons. Your brain is consuming 20 watts. You have about 100 billion neurons. So, the power efficiency of artificial neurons or simulated neurons compared to biological neurons is about 1 million. So, at the end of the day, for instance, Amazon web services powered by biological neurons will consume 100 times less energy which is the most important recurring cost of computer centers today.Financial advantageEnergy efficiency is great, but isn’t there a risk that the consumers will behave as consumers and go to an AI which can answer quicker? They could not care about energy efficiency and such things, they will care only about the time they will spend on this conversation with ChatGPT or another bot. So, do you think about ways to give your system the same speed as usual generative models?What is the first thing consumer take care of? The price. My price can be 100 times lower.But creating human cells is also very pricey and complicated.It's not complicated when you learned how to do it. And I think there's nobody in the world that can create a microprocessor on its own. You need a full industry for this, not only a factory. So it’s much more complicated to create a processor than a biological neuron network. And if your product consumes 100 times less energy, it 100 times less cheaper and so not only the initial investment is lower but the running cost is lower.How much money have you invested in this project from its start until the end of 2025? And how much do you plan to invest in 2026? What specifically the 2026 investments will be used for? In the past five years, we have invested about 2.5 million Swiss francs.In 2026, we plan to invest 500,000 Swiss francs. Main goals – to automate work with microfluid and connecting neurons to electronics. For automating biology part we design robots. In electronics we want to increase the number of organoids available for the for researchers. So right now we have 2-6 systems, I would love to have 20-50 systems.To reach the final stage of developing a biocomputer, we need to attract around $50–60 million in investment, including hiring approximately 20–30 more specialists.Read alsoA "fully Ukrainian" drone is still a long way off: how Ukraine is reducing its dependence on ChinaThe funds you have already invested – are they your own capital, or did you raise external funds?The main source of financing today is another company I co-founded – Alpvision. And there are a lot of individual people that write us with proposals to invest in our company from $3,000 to $1 mln. Meanwhile we have preliminary discussions with many possible big investors, like venture investors, the sovereign funds of Singapore and Saudi Arabia.It also can be a big business angel: if someone gives us even $10 million then it will be easier for me to find the remaining $40 million because of the interest of the others.What is the hardest part of negotiations with major investors?To build reliable relationships with the people, give them confidence. It takes time. But I hope we will have some progress in 2026.How long to wait for the launch of living AIWhat could be the first, the simplest commercial application of your product?We have a commercial application for two years already: this is a tool for researcher called the Neuroplatform. People pay monthly fee which ranges between $500 and $5000 to use our neurons remotely. It gives us about 10,000–50,000 Swiss francs per year.This is for science and it's great, but what can be a commercial use for ordinary people?This will be a new, our own chatbot which will cost way less than traditional models. Like $1 per month instead of $10 per month. But we are very far from this. Bioserver will be available in maybe 10 to 12 years.Why so long and what exactly you will be doing during this years to create it?We are creating a new science, a new industry from scratch. So 10 years is already very ambitious. First of all we have to learn these neurons.The main problem of classical AI was learning, it took decades to make it work. Training is the key. And training biological neurons is totally different from training artificial neurons.So when you train artificial network you have just set the right mathematical value for each neuron in the matrix. But in biological neural network we don't have this matrix. We have connections between neurons which are called synapses and the problem is that we don't have simple way to change these values.As a human being you don't know how exactly you learn, how each synapse changes connections for that process in your brain. You don't do this consciously. So we have to recreate this process of changing the synapses in order to output the right answers.If I show you a dog or a cat, you'll say correctly what is it: a dog or a cat. This is obvious for you, because you have the right synaptic connections in your brain.But some small kids can’t, because they don't have yet information in their neurons about the words "dog" and "cat" and don’t have right connections with the images of these animals.So this is the biggest challenge that we have to solve for biocomputing is to be able to train the synaptic connections between the neurons to output the right thing for the given input.Read alsoSurviving the blackout. How much Kyivstar, Vodafone and lifecell invest in energy independenceDo you conduct some scientific research on this by your own or do you work with other scientists who explore these connections?There are about 10 universities who are using our system for free. We gave them such access because they are working on the same problem as us. We meet with them every month to discuss the research progress.But the most critical part of work is still on our side – researching, how to stimulate the neurons in order that they learn something.How to make a living computer: details of the processHow did you extend the lifespan of the organoids from a few hours to four months? And what the maximum of such lifespan?It's already seven months, not four. And at the final stage of developing a biocomputer I believe it should be 100 years, like a human.To prolong the lifespan at this point we optimized the way the medium is brought to the cells. Speed, composition, many environmental parameters like air temperature.Where do you take the stem cells which then transform into neurons? Could it be the cells from your own body?These cells are produced by several labs in the world. We buy cells typically from Canada or the US.Converting cells from blood into stem cells requires specific protocols and we don't have the expertise to do this, although we could learn how to do it. But we don’t have to buy stem cells very often: we can buy 1 million every three years and multiply them as much as we want.I could use my own cells, that would be fun. But if I use my own cell and then accidentally insert mutated cell in my own body during the lab work it could be dangerous for me.Can you describe the process of transforming these cells from stem to neuron?We put cells in the medium, add specific molecules. For example, the molecule BDNF (Brain-Derived Neurotrophic Factor) forces cells to stop multiplying and become something else.If you put the right molecules at the right time every day in the medium for a few weeks, the cells will transform their type several times. You have to be careful because mistakes can lead to creating cells of different organs like eye, hair or bone.What does your laboratory look like, what does it consist of? We have two labs. There is a usual cell culture lab, where we have a centrifuge, a laminar flow hood, microscope, incubators. And the electrophysiology lab where we bring our neurons and put them on the electrodes: here we have things which are bit less usual, like micropumps to have the fluids coming; electronics to measure the actual potential and to send some signals; 3D printers; laser cutter; microscope.And recently we bought our first PCR machine for measuring which exactly genes are expressed. We make neurons from stem cells, but there are about 400 types of neurons. So in order to know exactly which type of neuron we have, whether it is well functional or not, we have to check a number of genes that are expressed and for this we bought this new machine, which was our dream.Read also"We started with $30,000". How the production of bionic prostheses is organized: CEO of Allbionics showsYou have four patents for your inventions. What exactly are they for and what they give you in practice? There are two patent families. The first one is fundamental: about the concept of biocomputer and microfluid for keeping alive and training the neurons. The second one is about technical details about the microfluidic. For example, we found that getting rid of bubbles in microfluid can prolong the lifetime of neurons.Filing those patents costed us about $10,000. In practical use patents give us nothing in short term. But they can help to attract investment in the future.What is the most complex operation your biocomputer can perform at the moment? Nothing impressive. Like storing one bit of information or reacting to specific electrodes when you stimulate it with electricity, like a basic circuit.How high is the risk that this artificial brains will develop to the point of spinning out of control? What is your biggest, fantastic fear of this machine?I have absolutely no fear of this machine. It won’t escape the lab – it's hard enough to keep it alive.

Alive AI: how FinalSpark makes a computer out of human brain neurons – interview with the CEO

Comments