My conversation about FinalSpark, with a polish blogger, Artur Kurasinski

Dr Ewelina Kurtys, Strategi Advisor at FinalSpark

The Future of Computing—Built from Living Cells

The work we’re doing at FinalSpark—where I’m proud to serve as a Strategic Advisor—is starting to draw increasing attention. And it’s no wonder. After all, we’re building computers out of living neurons. Who wouldn’t want to learn more about a project this wild?

In recent weeks, I’ve had the chance to speak about our work in several Polish media outlets. I’ve done podcasts, radio interviews, and written Q&As. I’m also planning to appear at the 3rd Congress Science for Society in Warsaw, where Dr Maciej Kawecki will present his new documentary on FinalSpark and our research into biocomputing.

One of my most recent conversations was with Polish tech blogger Artur Kurasiński. We talked about where we are today and where biocomputing could take us. The potential is huge: it’s believed that biocomputing could reduce AI-related energy consumption by a factor of a million. Yes—a million.

Right now, we’re still at the early prototype stage. Our “computers” are tiny spheres of brain cells—called neurospheres—about half a millimeter wide and made up of roughly 10,000 living neurons. You can see them live here. They don’t look like much yet, but they’re already showing promise as energy-efficient information processors.

A neurosphere: prototype of a biocomputer made by FinalSpark.

Curious what this all means for the future? Want to hear the full conversation?

You can read the original Polish version here: Zapomnij o krzemie – nadchodzi era komputerów z żywych komórek

And below, you’ll find the full English translation of our discussion.

Artur Kurasinski: What is FinalSpark and what does it do?

Eelina Kurtys: FinalSpark is a Swiss startup building computers from living neurons. It was founded in 2014 by Fred Jordan and Martin Kutter, scientists and entrepreneurs aiming to build a "thinking machine." After several years of working with digital technologies, they concluded that digital systems consume too much energy to support the development of thinking machines. They shifted their focus toward a complete technological transformation by building entirely new processors, where transistors are replaced by real neurons.

Your project involves using living neurons as computational units. What does the process of “programming” such neurons look like?

Currently, we operate a small-scale R&D laboratory. Tiny neurospheres made up of about 10,000 neurons each, roughly half a millimeter in diameter, are placed on electrodes. This setup allows us to send input signals and receive output in the form of electrical impulses. Here's a live view from our lab: https://finalspark.com/live/, showing our hardware via a camera and real-time neuron signal readings. These are real data, not just graphical representations, and can be used for research.

So far, we’ve been able to encode a single bit of information in our neurons this way.

We're also experimenting with neurotransmitters (e.g., dopamine) added to the medium. These neurotransmitters are enclosed in special chemicals (encaged) and can be released with light. We do this to enhance or suppress specific neuronal behaviors, measured by the output of electrical signals.

In the future, we plan to program neurons by sending them electrical impulses and releasing neurotransmitters into the medium — the fluid that constantly flows around the neurons.

The human brain works in the same way — we learn through electrical and chemical signals. However, this process is still not fully understood. No one really knows the exact algorithm behind how neurons learn. That’s why it’s the focus of our current research.

Can you describe how your Neuroplatform works and what types of remote research can be conducted on it?

As mentioned, we use small clusters of neurons placed on electrodes to enable bidirectional communication via electrical impulses. These electrodes are connected to computers where we can program experiments — defining what kind of electrical or chemical signals will be sent and when — and collect data, i.e., electrical signals from the neurons.

All our equipment is connected to sensors that monitor not only electrical signals but also temperature, pH, CO₂ concentration in the air, and other critical parameters that influence neuron condition and experimental success. We also have our own API, allowing full system interaction using Python in a web app.

So, it doesn’t matter if the computer running the experiments is in our lab or on the other side of the world — anyone can connect.

The platform is currently used for experimental research. No one fully knows how to program neurons yet. Teaching neurons — or programming them — is FinalSpark’s main goal. Our users also explore other topics, like estimating connections between neurons based on signals from the outer neuronal layer of the neurospheres.

What are the biggest technical challenges in working with organoids, and how do you solve them?

A major challenge is keeping neurons alive for a long time in lab conditions. We know from nature that neurons can live for up to 100 years in the human brain, but recreating the perfect physiological environment in vitro — temperature, pH, oxygen, nutrients, pressure, etc. — is hard.

Currently, we can keep our neurons alive on electrodes for three months. This is a patented solution based primarily on a microfluidic medium system — continuously delivering nutrients and removing metabolic waste from neurons. In living organisms, this role is played by blood and lymph flow through tissues.

Do your studies lead toward developing a new kind of “programming language” for biological neurons?

Absolutely. The programming language of neurons is still a mystery. Cracking it would be a revolution not only for biocomputing but also for medicine.

How do you measure the computational efficiency of living neurons compared to traditional digital hardware?

For now, we can't really compare in vitro neurons to digital computers. For example, our neurons can currently store just one (!) bit of information. At this early stage, our primary goal is to establish a meaningful relationship between what we send to the neurons (input) and what we get back (output).

Later, we plan to benchmark performance. Looking at the human brain, we expect biological computers won’t be faster or have more memory than digital ones — but they’ll be vastly more energy-efficient for tasks like pattern recognition and complex problem solving.

The main reason we’re working on living computers is that neurons are estimated to be about a million times more energy-efficient than digital computers when processing information.

You claim up to a million-fold energy efficiency improvement with bioprocessors. Can you cite specific studies or data supporting this?

Yes, this has been described by scientists outside of our project — for example, Professor Thomas Hartung of Johns Hopkins University(https://www.frontiersin.org/journals/science/articles/10.3389/fsci.2023.1017235/full).

Do you think biocomputing could address the rising energy costs of training large AI models?

Absolutely. Escalating energy costs of AI is a key problem we aim to solve.

Do you run simulations or projections regarding biocomputing’s potential impact on global CO₂ emissions?

No. But we think it’s possible to analyze projected energy usage or CO₂ emissions from digital AI in ten years and divide it by 1,000 to get a rough estimate.

You say current AI models only “simulate” thinking. How do you define real thinking, and how might biocomputers achieve it?

We don’t yet have a definition of real thinking because no one fully understands how neurons encode information. We only know it’s a fundamentally different mechanism than in computers.

Digital machines encode data in binary — 0s and 1s — via logic gate activation, based largely on linear algebra. But the brain is analog. Neurons encode information across time and space, and their structure is vastly more complex than that of digital circuits. There are many digital neuron models, but none match the complexity of the original.

Are there any signs that your neurons exhibit behavior close to consciousness? How do you define that boundary?

We haven’t observed anything suggesting consciousness. Though, philosophers often explore this topic. There are many publications discussing whether future computers — not just biological — could become conscious.

We also actively engage in discussions with philosophers on the ethics of biocomputers. Last year, I spoke at a conference at Delft University on this topic.

Beyond questions of consciousness, there are also debates about the boundary between humans and machines, and how society perceives radical new technologies.

What ethical issues do you face working with human cells, and how do you handle them?

Research on human cells has been conducted in biomedical fields for decades, and many protocols ensure ethical practices. It’s crucial to confirm that the original cells were obtained with donor consent. Today, human cells for in vitro research are commercially available, and the collection process is fully standardized.

Might there be a future need for granting “rights” to neuron-based biocomputers?

That can’t be ruled out, although I personally don’t support anthropomorphizing — assigning human qualities to everything, including biocomputers.

We’re not trying to recreate the human brain, only using the same basic building blocks — neurons. I don’t believe biocomputers (or any computers) will gain consciousness. But if practice proves otherwise, we’d certainly have to consider legal rights for machines. For now, that’s pure speculation.

Are you planning to commercialize your solutions, and what commercial applications do you see first?

Yes, commercialization is a core goal of FinalSpark. We envision building centralized biocomputing units accessible remotely — similar to how cloud services like AWS work. We hope to offer computational power that’s at least ten times more energy-efficient than digital systems, considering all operating costs of such a technology — including labs with tightly controlled conditions where cells can function long-term.

You already collaborate with 8 universities. Do you plan to expand access to your Neuroplatform to startups or tech companies?

We’ve already done that. We offer subscription options for private clients — and several companies and individuals have signed up. We provide two subscription types: shared (where experiments are scheduled and hardware is shared among users) and dedicated (with exclusive access to a specific neuron cluster). The dedicated version also supports simple experiments via an interface that requires no coding skills.

Check our website. We also grant access to selected artists who envision ways to illustrate this innovation for the public.

What’s your long-term vision? Could biocomputers eventually replace silicon chips as the dominant computing architecture?

We don’t believe biocomputers will completely replace silicon chips. When comparing digital computers to the human brain, each has strengths and weaknesses. It's more likely that biocomputers will complement digital systems.

They’ll likely be better suited for neural networks — the foundation of current AI, which is partially inspired by our understanding of how the human brain works.

Original text in Polish here: https://blog.kurasinski.com/2025/04/zapomnij-o-krzemie-nadchodzi-era-komputerow-z-zywych-komorek/