A computer made partly from living human brain cells sounds like something from a science-fiction movie. Yet biological computing is no longer just an idea for the distant future.
Researchers are now experimenting with systems that connect living neurons to electronics, allowing biological cells and conventional computers to exchange information. In one of the most unusual demonstrations to capture public attention, living neurons were connected to a computer system capable of interacting with video games—including demonstrations involving the legendary game Doom.
But the reality is more complicated than the headline.
A collection of human neurons is not simply sitting inside a computer, consciously holding a controller and playing a video game. These systems combine living cells with electrodes, software, conventional processors and carefully designed feedback systems. The biological component is only one part of a much larger machine.
That does not make the technology any less extraordinary.
The human brain performs an astonishing amount of computation while consuming relatively little energy. Meanwhile, artificial intelligence is pushing modern computing infrastructure toward an enormous demand for processors, data centers and electricity. That has led some researchers to ask an unusual question: Could biology eventually teach us a better way to compute?
The answer remains uncertain. Biological computers are still experimental, difficult to build and surrounded by major scientific questions. But they also raise questions that are even more difficult to answer.
If living human neurons can learn, adapt and interact with machines, where does computation end and biology begin? Could increasingly complex biological systems one day become genuinely intelligent? And perhaps most importantly, how would we know if something inside the machine was actually experiencing anything at all?
To understand why these questions are becoming increasingly serious, we first need to understand what a biological computer actually is.
What Is a Biological Computer?
Before we can understand how living neurons could interact with a video game, we need to understand a much simpler question:
What exactly is a biological computer?
When most people hear the word computer, they imagine silicon chips, processors, memory and software. Traditional computers process information by controlling the movement of electricity through billions of tiny electronic components called transistors.
A biological computer takes a very different approach.
Instead of relying entirely on silicon, it uses biological material as part of the computing system. Depending on the experiment, that biological material might include DNA, proteins, bacteria or living cells. One of the most unusual areas of this research involves living neurons—the specialized cells that carry and process information in the nervous system.
Neurons communicate using a combination of electrical and chemical signals. When large networks of neurons interact, their activity can produce extremely complex patterns. In a biological computing experiment, researchers can place these cells on or near electronic devices capable of both recording their activity and sending stimulation back to them.
This creates a two-way connection:
Computer → sends information to the neurons
Neurons → respond with electrical activity
Computer → reads that activity and converts it into useful information
The result is not simply a “brain replacing a computer chip.” At least with today’s technology, these systems are better understood as hybrid biological-electronic systems.
The conventional computer still performs many essential tasks. Software may process data, monitor the experiment and translate information into signals that the neurons can receive. Meanwhile, the living cells respond according to their own biological properties, creating activity that researchers can measure and potentially use as part of a computational process.
That distinction is important.
When headlines say that “brain cells are playing a video game,” the image can be misleading. The neurons do not have eyes, hands, a screen or a conscious understanding of game rules in the way a human player does. Instead, the computer translates parts of a digital environment into patterns of stimulation, while the biological cells produce responses that can influence what happens next.
In other words, the machine acts as a translator between two completely different worlds:
The digital world of computers and software
and
The biological world of living cells.
This is what makes biological computing so fascinating. Scientists are not simply trying to build a smaller computer. They are exploring whether the information-processing abilities developed by living systems over billions of years can work alongside the technology humans have built from silicon.
And to understand why anyone would go through the extraordinary difficulty of keeping living cells alive inside a computing system, we need to look at one of the biggest problems facing modern technology:
Energy.
Why Are Scientists Interested in Brain Cells?
If conventional computers already work incredibly well, why would anyone want to put living neurons inside a computer system in the first place?
The answer starts with a problem that is becoming increasingly difficult to ignore:
Computing is getting hungry for energy.
Modern artificial intelligence depends on enormous amounts of computing power. Training and running increasingly capable AI models requires large numbers of processors working together inside data centers. Those machines need electricity not only to perform calculations, but also to power memory, networking, cooling and the rest of the infrastructure around them.
And the demand is growing.
The more capable AI becomes, the more computation researchers often need. Larger models, bigger datasets and more complicated tasks can require enormous computing resources. This has created a strange situation: some of the biggest technology companies in the world are no longer thinking only about faster chips or better algorithms. They are also thinking about where all the electricity needed to run those machines will come from.
That is where biology becomes interesting.
The human brain is an extraordinary information-processing system. It contains roughly 86 billion neurons, connected through an enormous network of synapses, yet the brain operates on roughly the power of a small light bulb.
That does not mean the brain is simply a more efficient version of a modern GPU. The two systems work in fundamentally different ways, so comparing their raw performance directly can be misleading.
But the difference is still fascinating.
A conventional computer performs calculations using precisely engineered electronic components. The brain, meanwhile, processes information through a constantly changing network of living cells that communicate with one another through electrical and chemical activity.
And the brain does something else that researchers find particularly interesting:
It learns from remarkably little information.
A modern machine-learning system might require enormous amounts of training data to recognize patterns reliably. Humans can often learn from only a handful of examples.
A child does not need to see thousands of photographs before understanding what a dog is. After seeing a few dogs, the brain can begin building a flexible concept that works across different breeds, sizes, colors and environments.
That ability comes from the brain’s biological architecture—not simply from a faster processor.
So researchers are asking a much bigger question:
What if some of the principles that make biological brains so efficient could be incorporated into computing?
The goal isn’t necessarily to grow a biological replacement for your laptop or GPU.
Instead, biological computing could eventually become a complement to conventional computing, potentially useful for particular problems where living neural networks have interesting advantages.
But there is an important reality check.
A few hundred thousand neurons in a laboratory are nowhere near a human brain. They do not suddenly inherit human intelligence, memories or understanding. Keeping biological tissue alive is difficult, controlling it is difficult, and connecting it reliably to electronics is an enormous engineering challenge.
So why bother?
Because scientists are not only interested in building a better computer.
They are also using these systems to investigate how biological intelligence itself works.
If researchers can grow human neurons, connect them to electronics, stimulate them, record their activity and observe how their behavior changes through learning, they gain a new experimental window into the nervous system.
And that could be valuable far beyond computing.
It could help researchers investigate how neurons learn, how neurological diseases affect human cells, and how biological systems respond to different environments or treatments.
In other words, biological computing sits at the intersection of two very different ambitions:
Build computers that learn from biology.
And:
Use computers to understand biology.
The strange thing is that the first experiments in this field were much simpler than today’s demonstrations.
Scientists didn’t begin with Doom.
They began with something much smaller: a dish of neurons connected to a machine.
How Did Scientists Teach Neurons to Play a Game?
The idea of connecting living neurons to a computer sounds futuristic, but the basic concept is surprisingly simple.
A neuron is an information-processing cell. It receives signals, responds to them and communicates with other neurons through electrical and chemical activity.
So researchers began asking a strange question:
What happens if you give those neurons a digital world to interact with?
The first challenge was obvious: neurons cannot plug themselves into a USB port.
They need an interface.
Researchers can grow neurons in laboratory cultures and place them on arrays containing tiny electrodes. These electrodes can perform two important jobs: record electrical activity from the cells and deliver electrical stimulation back to them.
That creates a communication channel between the biological system and the computer.
Imagine a video game where the computer needs to tell the neurons what is happening.
Instead of showing them a screen, the computer can translate information from the game into patterns of electrical stimulation. The neurons respond with their own electrical activity, which the electrodes detect and send back to the computer.
Now you have a loop:
GAME → COMPUTER → ELECTRODES → NEURONS → ELECTRODES → COMPUTER → GAME
And suddenly, the neurons can become part of the system controlling what happens next.
But there is still one enormous problem.
How do you teach a collection of cells what you want it to do?
You cannot sit down and explain the rules.
You cannot show the neurons a tutorial.
And you certainly cannot give them a controller.
Instead, researchers can create a feedback system.
The basic idea is similar to learning through consequences. The computer can provide different patterns of stimulation depending on what happens.
In the famous Pong experiments, for example, the biological system was given information about the position of the ball and paddle through electrical signals. When the system produced a useful response, the feedback could be made more predictable. When things went wrong, the stimulation could become less predictable.
The neurons did not need to understand the concept of Pong.
They simply responded to the patterns of activity around them.
Over time, their behavior could change.
This is one of the most fascinating properties of biological neural networks: they are plastic. Their connections and activity can change in response to experience.
That doesn’t mean a dish of neurons suddenly becomes a tiny human brain.
It means the biological network can adapt.
And that distinction is crucial.
Researchers aren’t giving the cells a list of instructions saying, “Move the paddle left when the ball goes left.”
They are creating an environment in which certain patterns of activity produce more favorable feedback than others.
The biological network then becomes part of the process of finding a useful response.
This is fundamentally different from writing a conventional computer program.
A programmer could simply code the rules of Pong directly into software. A biological system, however, can be allowed to adapt its activity through feedback.
That makes the experiment interesting not because neurons are secretly tiny gamers, but because researchers are testing whether living neural networks can participate in a learning system.
And this idea didn’t appear overnight.
Scientists had already spent years experimenting with living neurons connected to machines.
Some of the earliest demonstrations were extremely primitive compared with today’s systems. Researchers were essentially trying to answer one question:
Can living neural tissue control something outside itself?
The answer appeared to be yes.
But another question immediately followed—and it was much harder to answer:
Were the neurons actually learning, or were researchers simply interpreting random biological activity as meaningful behavior?
That question would follow the field for years.
And eventually, researchers would design experiments specifically to test whether the biological cells were actually contributing something to the system.
That is where the story gets much more interesting.
Part 5 — DishBrain and the Pong Experiment
The breakthrough that put biological computing on the map did not involve Doom.
It involved a game that is almost comically simple by today’s standards:
Pong.
In 2022, researchers led by Cortical Labs described a system called DishBrain, in which living neurons were connected to a computer through a high-density array of tiny electrodes. The cultures contained roughly 800,000 human and mouse brain cells. The neurons were grown directly on the electrode array, allowing researchers to both stimulate the cells and record their electrical activity.
The choice of Pong was deliberate.
Pong is simple enough that the researchers could reduce the problem to a few basic pieces of information: where the ball was, where the paddle was and what happened after the system moved the paddle.
That made it possible to turn a video game into a controlled neuroscience experiment.
Turning a Video Game Into Electrical Signals
The neurons couldn’t see the Pong screen.
Instead, the computer translated information about the game into electrical stimulation delivered through the electrode array.
Think of it as giving the biological network a very crude sense of its digital environment.
The computer essentially provided information about what was happening, while the neurons generated electrical activity that could be measured and translated back into actions in the game.
That created something called a closed-loop system.
The loop looked roughly like this:
Game state → electrical stimulation → neurons → recorded activity → computer → game state
The important part is that the loop went in both directions.
The computer wasn’t simply recording a dish of neurons doing random things. What the neurons did could influence what happened next in the virtual environment, which then changed the information fed back to the neurons.
That feedback was central to the experiment.
Researchers reported that the cultures showed apparent improvement in the Pong task within minutes under the closed-loop conditions, while comparable control conditions without the structured feedback did not show the same learning pattern.
And that is where the experiment becomes much more interesting than the headline:
The researchers weren’t just asking whether neurons could produce electrical signals. They were asking whether those signals could change through interaction with an environment.
But How Do You Reward a Dish of Neurons?
This is probably the strangest part of the entire experiment.
You can’t tell the neurons:
“Good job. You hit the ball.”
There is no tiny coach standing beside the petri dish.
Instead, the researchers designed the environment so that successful and unsuccessful outcomes produced different patterns of stimulation.
The approach was influenced by the free energy principle and active-inference ideas used by the researchers. In simplified terms, the system was designed around the idea that biological networks can respond to changes in their environment and seek more predictable states.
When the system performed successfully, the feedback was structured and predictable.
When it performed poorly, the feedback became less predictable.
The neurons weren’t told what Pong was.
They weren’t given a written strategy.
Instead, they were placed inside an environment where their activity had consequences.
Over repeated interactions, their collective activity changed.
That is what made the result scientifically interesting.
Was It Really “Playing Pong”?
This is where we need to slow down.
Saying “800,000 brain cells learned to play Pong” is an effective headline, but it can create the wrong mental picture.
There wasn’t a tiny conscious player sitting inside the dish thinking:
I need to move the paddle.
The researchers built a complete biological-electronic system around the neurons. The computer supplied information, electrodes communicated with the cells, and software interpreted the resulting activity.
What the experiment demonstrated was much narrower—and arguably more interesting.
The neural cultures could participate in a closed-loop environment and modify their activity in a way that improved performance on the task. The researchers described this as synthetic biological intelligence.
That is very different from demonstrating human-like intelligence.
It also doesn’t prove that the cells understood Pong.
And it certainly doesn’t prove that the cells were conscious.
Those distinctions became extremely important because the study itself used the word “sentience” in its title, triggering considerable discussion about what that term should mean when applied to laboratory-grown neural cultures.
But before getting into that philosophical minefield, there is another problem we need to solve.
If a dish of neurons can participate in Pong, what happens when you replace Pong with something much more complicated?
That question eventually led researchers toward a far more ambitious demonstration.
Doom.
Part 5 complete
Article progress: ~1,850 words.
I also tightened the scientific wording here compared with the original YouTube script. For example, instead of presenting “the neurons hate chaos” as an established fact, we explain the feedback mechanism and the theoretical framework more carefully. That’s important if we’re building a site readers can trust.
Part 6 — From Pong to Doom
Pong was a remarkably simple test.
There was a ball. There was a paddle. The objective was straightforward: keep the ball in play.
But if researchers wanted to push biological computing further, they needed a more demanding environment.
And eventually, they chose one of the most famous stress tests in computing history:
Doom.
Released in 1993, Doom is a first-person shooter in which the player navigates a three-dimensional environment, fights enemies and tries to survive. It is dramatically more complicated than Pong—not because modern hardware struggles to render it, but because interacting with the game requires a much richer stream of information and actions.
In 2026, Nature reported on Doom’s unusual role in scientific research, including its use in experiments involving neurons in a dish.
The demonstration came from Cortical Labs, the company developing the CL1, a biological computing platform designed to keep living neural cultures alive while allowing electronics and software to communicate with them.
Cortical Labs describes the CL1 as a closed-loop system: living neurons are cultivated in a nutrient-rich environment on a silicon-based platform, while the system can send electrical information to the neurons and record their responses.
That architecture is important because the neurons aren’t operating in isolation.
The CL1 essentially creates a bridge between two kinds of computing.
Silicon handles the digital world.
Neurons provide the biological component.
And the two continuously exchange information.
So How Does Doom Get Into the Picture?
Imagine the game running normally on a computer.
The computer knows where the player is, what is happening around them and what actions are available.
Instead of a human sitting at the keyboard, however, the biological system becomes part of the control loop.
Information from the simulated environment can be translated into electrical stimulation delivered to the neural culture. The resulting neural activity is recorded through the same general interface and can then be interpreted by the surrounding computer system.
The neurons aren’t looking at a monitor.
They’re receiving an electrical representation of information from the digital environment.
And that distinction matters.
The biological system doesn’t need to understand the concept of a demon, a weapon or even the word Doom. What matters is that the neural network receives signals and produces activity that can be incorporated into the larger system.
Cortical Labs says its CL1 platform uses a programmable, bidirectional stimulation and recording interface, allowing software and biological neural networks to interact in real time.
That is the real technological achievement.
The remarkable part isn’t simply getting an old video game to run.
The remarkable part is creating a machine in which living neurons can participate in a digital environment.
A Biological Computer Is Still a Hybrid Computer
This is where the headline can become misleading.
You might hear:
“Human brain cells are playing Doom.”
That sounds as if someone took a tiny human brain, connected a keyboard to it and left it alone to play a video game.
That’s not what is happening.
The system contains conventional computing hardware and software alongside living neural tissue. The electronics provide the interface, the simulated environment and the mechanisms required to translate information between the digital and biological systems.
The neurons are one component of the machine.
Cortical Labs itself describes the CL1 as a combination of “hard silicon and soft tissue”, with neurons interacting with a simulated world through its biological intelligence operating system.
That makes the experiment much easier to understand.
Think of the neurons not as a replacement for the entire computer, but as a living processing element inside a larger computing architecture.
And that leads to the question that matters most.
If conventional software is still doing so much of the work, how much is the biological system actually contributing?
That’s where things get uncomfortable.
Because the most interesting question isn’t whether the neurons can produce useful signals.
It’s whether those signals represent something genuinely different from what a conventional computer could produce.
And to answer that, we have to look inside the system itself.
Part 7 — Is the Biological System Actually Doing the Thinking?
This is where the story gets much more interesting.
It is easy to hear that human neurons are playing Doom and imagine a tiny biological brain sitting inside a machine, making decisions on its own.
That is not what the system is.
The reality is much more complicated—and, in some ways, more interesting.
The Doom experiment is a hybrid system. Living neurons are connected to conventional computing hardware and software. The digital system handles the game and translates information between the game and the biological network. The neurons receive electrical stimulation, produce measurable electrical activity, and those signals can then be fed back into the digital system.
Think of the entire system as having three major layers:
1. The digital world
This is where Doom actually exists.
A conventional computer runs the game and knows what is happening inside the virtual environment.
2. The biological layer
This is the living neural network.
The neurons receive electrical stimulation and produce patterns of electrical activity—often referred to as spikes—that can be recorded by the electrode array.
3. The translator
Something has to convert information between those two completely different systems.
The digital environment has to become electrical stimulation that the neurons can receive. The resulting neural activity then has to be converted into information that the computer can use.
So the system looks more like:
DOOM → SOFTWARE → ELECTRODES → NEURONS → ELECTRODES → SOFTWARE → DOOM
That is very different from putting a brain inside a computer and letting it figure everything out.
The Computer Is Still Doing a Lot of the Work
This distinction became especially important when the code behind the Doom-neuron experiment was made publicly available.
The published implementation shows that the CL1 functions primarily as a neural hardware interface. The game logic and machine-learning components run on the surrounding computer system, while the CL1 receives stimulation commands and returns recorded neural activity.
In other words, the neurons aren’t responsible for rendering Doom.
They aren’t looking at pixels.
They aren’t independently deciding what constitutes an enemy.
And they aren’t running the entire learning algorithm inside the biological tissue.
The conventional computer still provides enormous amounts of computational infrastructure around them.
This matters because it changes what the experiment actually demonstrates.
The impressive question isn’t:
“Can a brain replace a computer?”
The experiment doesn’t show that.
A much more interesting question is:
“Can living neural tissue become a useful computational component inside a computer system?”
And that is a question the experiment is genuinely designed to investigate.
So What Are the Neurons Actually Contributing?
This is the difficult part.
A neural network isn’t simply a collection of miniature processors executing instructions. Its behavior emerges from the activity and connections of many cells.
The experiment therefore treats the neurons as part of a learning loop.
Information is encoded into stimulation.
The neurons respond.
Their activity is recorded.
Software interprets that activity.
The resulting action changes the digital environment.
Then the new environment produces another round of stimulation.
And the cycle continues.
That gives the biological network an opportunity to influence the behavior of the overall system.
The publicly available Doom project even includes ablation tests designed to investigate whether the neural signals are actually important. In these tests, the biological spike signals can be replaced with zeros or random values. If the system could continue learning normally without meaningful neural information, that would suggest the surrounding software was doing more of the work than the biological component.
That is an important scientific idea:
If you remove one component and the system still works exactly the same way, that component probably wasn’t doing very much.
So the real scientific challenge isn’t simply getting neurons connected to a game.
It’s demonstrating that the neurons provide a meaningful contribution that cannot be explained entirely by the conventional software surrounding them.
And that brings us to an important distinction that gets lost in almost every sensational headline.
Learning Is Not the Same as Understanding
A system can change its behavior based on feedback without understanding what it is doing in the human sense.
A thermostat can respond to temperature.
A reinforcement-learning algorithm can improve its behavior after receiving rewards.
A neural network can recognize patterns without having a human-like understanding of those patterns.
And living neurons can change their activity in response to stimulation.
None of these facts, by themselves, prove consciousness.
This is why saying “the brain cells learned to play Doom” needs some qualification.
There is a meaningful sense in which the biological system can participate in learning a task.
But that does not automatically mean the cells understand the game, know that they are playing it, or experience the game in anything resembling the way a human player does.
Those are entirely different claims.
And once you separate learning, intelligence, and consciousness, the story becomes much more complicated.
Because scientists can measure neural activity.
They can measure changes in behavior.
They can test whether removing the biological signal affects performance.
But measuring subjective experience is another problem entirely.
And that is where this technology starts crossing from computer engineering into one of the oldest unanswered questions in neuroscience:
What does it actually take for something to experience the world?
Before we get there, however, there is another technological limitation we need to understand.
The neurons in these systems are usually grown in relatively simple arrangements.
But the human brain isn’t a flat sheet of cells.
It is a three-dimensional structure containing layers, regions and extraordinarily complex connections.
And researchers are increasingly interested in something that looks much more like that:
miniature 3D brain-like structures called organoids.
Part 8 — Why 3D Brain Organoids Could Be the Next Big Step
There is a problem hiding underneath all of these experiments.
The human brain is not flat.
A living brain contains different types of cells arranged in three dimensions, organized into layers and regions, with an enormous number of connections forming throughout the tissue. A relatively simple layer of neurons growing across a chip can reproduce only a small fraction of that complexity.
That is why another technology has become so important to biological computing:
brain organoids.
A brain organoid is a three-dimensional cluster of cells grown in the laboratory that can develop some features of brain tissue. Researchers can create these structures from stem cells and study how neural cells develop, organize and interact.
They are sometimes described as “mini-brains,” but that phrase can be misleading.
A brain organoid is not a miniature human brain.
It does not reproduce the full structure, complexity or capabilities of an actual brain. Instead, it is a laboratory model that can reproduce certain aspects of human neural development and organization.
And that distinction matters enormously.
Why Grow Neurons in 3D?
Imagine trying to understand a skyscraper by looking only at a single floor.
You might learn something useful, but you’d miss the elevators, staircases, different departments and the way all those floors connect together.
Something similar happens when scientists study neurons in a simple two-dimensional culture.
A 2D neural network can still be remarkably useful. Its cells can form connections and communicate with one another, and researchers can place electrodes across the culture to record and stimulate activity.
In fact, this relatively simple arrangement is one of the reasons systems such as DishBrain and CL1 can interact with computers.
But a 3D structure can provide a more complicated biological environment.
Researchers working with organoids can investigate things such as cellular organization, development and more complex neural networks. University of Queensland researcher Ernst Wolvetang has described 3D organoids as having more cell types and more complex neuronal networks than the relatively simple 2D cultures used by Cortical Labs.
And this creates an intriguing possibility.
What if researchers could combine the advantages of both approaches?
3D biological complexity + electronic communication.
The Problem: How Do You Talk to a 3D Brain?
This sounds easier than it is.
The electronic interface used to communicate with neurons is generally much easier to build when the cells are spread across a surface.
A flat layer allows electrodes to sit underneath or alongside the cells and interact with a large portion of the network.
But imagine putting a three-dimensional ball of neural tissue on top of that same flat electrode array.
The electrodes can interact with the part closest to them.
Much of the organoid is physically farther away.
That means researchers face a fundamental engineering problem:
How do you communicate with a three-dimensional biological network without only talking to its surface?
Solving that problem could require better electrode technologies, more sophisticated interfaces and improved methods for delivering and measuring signals throughout biological tissue.
Researchers working on organoid intelligence have specifically proposed scaling up brain organoids while simultaneously increasing the number of electrodes, improving real-time interaction algorithms and connecting the organoids to more sophisticated input and output systems.
If those pieces eventually come together, the biological system could become considerably more complex than the relatively simple neuron cultures used in early experiments.
From Neurons on a Chip to Organoid Intelligence
This idea has already been given a name:
Organoid intelligence, or OI.
The basic concept is not to build a biological computer that copies a conventional computer exactly.
Instead, researchers are interested in taking advantage of the natural information-processing properties of biological neural tissue.
The goal would be to create systems in which organoids receive information from their environment, process that information through biological activity and produce measurable outputs.
That could eventually mean connecting organoids to:
- computers
- sensors
- robotic systems
- digital environments
- other biological systems
The field is still highly experimental, and many of the proposed applications remain research goals rather than commercially proven technologies. But the roadmap is fascinating because it combines stem-cell biology, neuroscience, computing and artificial intelligence into a single research direction.
And there is another reason scientists care about this.
A more sophisticated biological model might not only be useful for computing.
It could become a tool for studying the biology of learning itself.
Researchers could potentially investigate how neural networks change when they learn, how diseases interfere with those processes and how different treatments affect biological information processing.
That could make the computer part almost secondary.
Instead of asking:
“Can we build a computer out of a brain?”
researchers could eventually be asking:
“Can we use a living neural system as a new experimental model for understanding intelligence?”
But there is a huge obstacle standing between today’s experiments and that future.
The more complex these biological systems become, the harder it may become to answer a deeply uncomfortable question:
At what point does a collection of living neural cells become something we have an ethical responsibility toward?
And that question becomes even more important when those cells are connected to a machine that is deliberately giving them experiences, feedback and goals.
Part 8 complete.
Part 9 — Could Biological Computers Actually Solve AI’s Energy Problem?
There is a reason biological computing has attracted attention beyond neuroscience.
It isn’t just because putting living neurons on a chip is strange.
It is because energy has become one of the biggest constraints on modern computing.
Artificial intelligence has pushed this problem into the spotlight. Training and operating increasingly capable AI systems requires large amounts of computation, and that computation ultimately requires electricity. Modern AI infrastructure also brings additional demands from memory, networking, cooling and other data-center systems. Recent research reviews describe the energy cost of large-scale AI as a significant technical and societal challenge.
And this creates an interesting comparison.
The human brain performs continuous learning, adaptation and information processing while operating at roughly 20 watts. Researchers studying biological intelligence therefore see the brain as an important source of ideas for making computing more energy efficient.
But there is a catch.
A biological computer is not automatically an energy-saving computer.
Keeping living tissue alive requires a support system. The cells need appropriate conditions, nutrients and environmental control. The electronics surrounding them also consume energy. A complete biological computing platform therefore has to be evaluated as a whole rather than by looking only at the neurons themselves.
That distinction is important because it is easy to make an impressive-sounding comparison:
Human brain: about 20 watts.
Modern AI: enormous amounts of electricity.
Therefore, biological computers must be the solution.
It doesn’t work that simply.
The brain and an AI data center are doing fundamentally different things, using fundamentally different architectures.
The real question is whether researchers can capture some of biology’s efficiency without needing to reproduce the entire human brain.
Why Biological Neural Networks Are Interesting
One reason researchers are excited about biological systems is their ability to learn and adapt.
Traditional machine-learning systems generally depend on enormous computational resources and carefully designed training processes. Biological neural networks, by contrast, evolved to operate continuously in messy, uncertain environments while learning from relatively limited information.
A recent review of synthetic biological intelligence describes this research as an attempt to combine living neural tissue, hardware and software to investigate biological information processing and potentially create new computing architectures.
That doesn’t mean a dish of neurons can currently compete with an NVIDIA GPU at rendering graphics or training a giant language model.
It can’t.
The potential advantage lies somewhere else.
Biological systems could offer a fundamentally different way of processing certain kinds of information—particularly tasks involving adaptation, noisy signals and learning.
Researchers are therefore not necessarily trying to build a biological replacement for every computer.
They are exploring whether biology could become another type of computational hardware.
The Goal May Be Hybrid Computing
This is perhaps the most realistic way to think about the technology.
The future might not be:
Biological computers replace silicon computers.
It could instead be:
Biological systems + conventional computers = a new type of hybrid computer.
The silicon hardware could handle tasks it is exceptionally good at:
- numerical calculations
- data storage
- communication
- running software
- controlling experiments
The biological component could contribute its own strengths:
- adaptive behavior
- neural plasticity
- learning from feedback
- processing information through biological networks
This idea is already reflected in modern research into synthetic biological intelligence and organoid intelligence, which combines biological neural networks with engineered hardware and software.
And the concept is no longer purely theoretical.
Cortical Labs’ CL1, for example, is designed as a closed-loop platform in which living neurons interact with software in real time. The company describes the system as combining living neural networks with silicon hardware and provides a controlled environment designed to keep the neurons alive for extended periods.
That is a very different proposition from building a giant computer entirely out of cells.
It is more like adding a new kind of biological processing layer to a conventional machine.
Could This Eventually Make AI More Efficient?
Possibly—but we are nowhere near being able to say that it will.
Research into brain-inspired computing is already exploring how principles from biology could lead to more energy-efficient artificial systems. Scientists are studying everything from neural signaling and architecture to learning strategies in an effort to understand why biological brains can achieve sophisticated behavior with remarkably low energy consumption.
But there are several enormous engineering challenges.
Biological systems are:
Fragile.
They require carefully controlled environments.
Variable.
Living cells don’t behave like perfectly identical transistors manufactured in a factory.
Slow to maintain.
A silicon chip can be manufactured, shipped and replaced relatively easily. Living neural tissue requires biological maintenance.
Difficult to scale.
A modern processor can contain billions of transistors packed into an extraordinarily small space. Building and controlling a large, stable biological neural network is a completely different challenge.
And then there is perhaps the biggest question of all:
What exactly should we use biological computers for?
If a biological system requires specialized equipment, life support and careful maintenance, it doesn’t make sense to use it simply to perform calculations that a conventional processor can already perform cheaply and efficiently.
The technology needs a problem where its biological properties provide a meaningful advantage.
That’s why researchers are also interested in applications beyond conventional computing.
Biological neural systems could potentially become tools for studying learning, neurological disease, drug responses and other aspects of human biology. Reviews of organoid intelligence highlight potential applications spanning neuroscience, biomedicine and information processing.
So the biggest opportunity may not be creating a biological version of today’s computers.
It may be creating a completely different category of machine.
A machine where living tissue isn’t merely being studied under a microscope.
It is actually participating in the computation.
And as that idea moves from laboratory experiments toward commercial platforms, another question appears:
Who is actually building these machines—and are biological computers already becoming a real product?
Part 9 complete.
Part 10 — The Companies Turning Biological Computing Into a Real Product
For years, biological computing existed mostly inside research laboratories.
Scientists could grow neurons, connect them to electrodes and run carefully controlled experiments—but that was very different from having an actual computing platform that other researchers could use.
That is beginning to change.
A small number of companies are now trying to turn biological computing from an experimental concept into something researchers can actually access.
Two names are particularly interesting: Cortical Labs and FinalSpark.
They are approaching the problem from different directions.
Cortical Labs: Building a Computer Around Living Neurons
Australian biotechnology company Cortical Labs is perhaps the most visible company in this field.
Its approach is relatively direct:
Put living neurons on a silicon chip and make them part of a programmable computing system.
The company’s CL1 platform is designed to maintain living neural cultures while allowing software and electronics to communicate with them. Cortical Labs describes the CL1 as a “code-deployable biological computer,” with neurons cultivated on a silicon-based interface that can send and receive electrical signals.
The important word here is closed-loop.
The computer can send information to the neurons.
The neurons respond.
The system records that response.
And software can use the resulting information to influence what happens next.
That makes the CL1 less like a traditional computer sitting on a desk and more like a living computational component integrated into a conventional computer architecture.
Cortical Labs says the CL1 is designed to keep its neural cultures alive for up to six months while providing the electronics and life-support systems needed for experiments.
And the company isn’t positioning the technology purely as a gaming experiment.
Its stated goals include research into neural processing, learning, drug effects and biological alternatives to some forms of animal experimentation.
The company has also moved beyond the idea of requiring every researcher to own a physical machine.
Its Cortical Cloud is designed to provide remote access to biological computing resources, allowing researchers to work with biological neural networks without operating their own specialized laboratory.
That is a significant change in the accessibility of the technology.
Instead of:
Build laboratory → maintain neurons → build electrode system → run experiment
the long-term model could become:
Write code → connect to biological hardware → run experiment
That sounds surprisingly similar to cloud computing.
Except the “server” contains living neurons.
FinalSpark: Putting Living Neural Tissue Online
Cortical Labs isn’t the only company trying to make biological computing accessible.
Swiss company FinalSpark has taken a different approach with its Neuroplatform.
Rather than focusing primarily on selling a biological computer for researchers to operate locally, FinalSpark provides remote access to living neural tissue through an online platform.
According to the company’s current Neuroplatform offering, researchers can remotely access brain organoids, stimulate and record neural activity, use a Python programming interface and store experimental data.
The idea is surprisingly simple:
You don’t need to own the biological hardware.
FinalSpark maintains the biological systems.
Researchers interact with them remotely.
That could remove one of the biggest barriers to experimenting with biological computing: the specialized infrastructure required to keep living neural tissue healthy and connected to measurement equipment.
The company says its platform provides 24/7 access to neural organoids and supports real-time neural stimulation and recording. It also says researchers can interact with the system programmatically through Python.
That turns biological computing into something that starts to resemble a conventional developer platform.
Instead of buying a machine and learning how to maintain its biology, a researcher can potentially interact with biological neural tissue through software.
Biological Computing Is Starting to Look Like an Industry
This is perhaps the most important development.
The field is no longer just about spectacular laboratory demonstrations.
Companies are beginning to build infrastructure around the technology.
Cortical Labs is developing physical biological computers and remote computing services.
FinalSpark is providing remote access to living neural systems.
Universities and research groups are beginning to experiment with these platforms as well. For example, in January 2026, Reply announced a collaboration with the University of Milan involving experimental research using Cortical Labs’ CL1 platform to investigate learning and information processing.
And in August 2026, the National University of Singapore announced a collaboration involving NUS Medicine, DayOne and Cortical Labs to develop a biological data center prototype, including demonstrations of CL1/Cortical Cloud systems and real-time neural activity.
That doesn’t mean biological computers are about to replace GPUs.
They aren’t.
The technology is still young, experimental and surrounded by major engineering challenges.
But something important has changed.
You can now build infrastructure specifically designed for computing with living neural tissue.
And once a technology becomes accessible to researchers outside the original laboratory that invented it, the possibilities can expand much faster.
The next question is no longer simply whether neurons can learn.
It is whether this strange new form of computing can become useful enough to justify the complexity of keeping living tissue alive inside a machine.
And that brings us to the most controversial question in the entire field.
Because the more sophisticated these biological systems become, the harder it becomes to dismiss one possibility:
What if the cells aren’t merely processing information?
What if, in some limited way, they are experiencing it?
Part 10 complete.
Part 11 — Could These Brain Cells Actually Be Conscious?
This is where the story stops being purely about computers.
It becomes a question about what it means to feel anything at all.
A dish containing living neurons can receive stimulation. Its activity can change. It can participate in a feedback loop. In some experiments, its behavior can improve on a task.
But none of those observations automatically tells us whether the cells experience anything.
And that distinction is critical.
Learning Is Not the Same as Feeling
Imagine two completely different systems.
The first is a machine-learning algorithm that becomes better at a game after millions of training attempts.
The second is a biological neural network that changes its activity after receiving feedback.
Both systems can appear to learn.
But learning alone doesn’t establish consciousness.
Consciousness is about subjective experience—the possibility that there is actually something it is like to be that system.
Does it experience anything?
Does anything feel good or bad?
Is there an internal point of view?
Those questions are extraordinarily difficult even when the subject is a human being.
With a laboratory-grown neural network, they become much harder.
The Problem: We Can’t Directly Measure Experience
Scientists can measure electrical activity.
They can measure changes in neural connections.
They can observe behavior.
They can stimulate cells and record their responses.
But there is no scientific instrument that can simply display:
CONSCIOUSNESS: 73%
We normally infer consciousness from behavior and from what we know about human brains and other organisms.
That’s already difficult with patients who cannot communicate.
It becomes even more difficult when the biological system has no body, no language and no conventional behavior.
Recent work in neuroethics emphasizes that questions about the possible sentience or consciousness of brain organoids remain scientifically and philosophically difficult, while also warning against sensational interpretations of what current organoids can actually do.
And this creates a strange problem.
A neural culture might produce activity that looks interesting without necessarily having any subjective experience behind it.
Does Responding to Stimulation Mean Something Is Being Felt?
Not necessarily.
A living system can respond to its environment without possessing anything resembling human consciousness.
A plant responds to light.
Single-celled organisms respond to chemical gradients.
Neurons respond to electrical and chemical signals.
So when researchers observe neurons changing their activity in response to stimulation, the observation itself doesn’t prove that the cells are experiencing pleasure, discomfort or pain.
This distinction is especially important when discussing experiments in which neural activity is influenced by different kinds of feedback.
It is tempting to describe the system in human terms:
reward
punishment
pain
pleasure
But these words can become misleading if we forget that they are descriptions of what the experiment does to the system—not proof of what the cells subjectively experience.
A signal that researchers call a “reward” does not necessarily mean the neurons feel rewarded.
Likewise, an unpleasant or unpredictable signal does not automatically mean the neurons suffer.
What About Brain Organoids?
This is where the ethical discussion becomes more serious.
Brain organoids can contain multiple types of neural cells and can develop patterns of electrical activity. They can model aspects of human brain development and function, which is one reason researchers are interested in them.
But today’s organoids are still extraordinarily different from complete human brains.
Recent reviews point out major limitations involving their size, organization, sensory integration, vascularization and overall complexity. One 2026 analysis argues that current cortical organoids lack several features that would be needed to make a strong case for sentient experience, and concludes that the evidence does not currently establish a meaningful likelihood of sentience.
That doesn’t settle the philosophical question forever.
It does, however, give us an important distinction:
“Could something like this eventually become conscious?”
is very different from:
“Is this particular organoid conscious right now?”
The first is a possibility scientists can debate.
The second requires evidence.
And at present, that evidence is not there.
Why Scientists Still Take the Question Seriously
If current systems are unlikely to be conscious, why discuss the issue at all?
Because the technology is changing.
Researchers are developing increasingly sophisticated ways to grow neural tissue, connect it to electronics and expose it to controlled environments.
As those systems become more complex, the ethical questions could become harder to dismiss.
A 2026 consensus paper from the Asia Pacific Neuroethics Working Group specifically examines the potential moral status of human brain organoids, including questions about sentience and consciousness, while emphasizing the need to distinguish genuine scientific concerns from sensationalized portrayals.
The concern isn’t necessarily that today’s tiny neural cultures are secretly suffering.
It is that we should decide how to recognize and respond to ethically relevant biological systems before the technology becomes capable of producing one.
That is a very different argument.
It is a precautionary question:
What evidence would convince us that a biological computing system has become morally significant?
And that question becomes particularly difficult because consciousness itself is still not completely understood.
Intelligence, Consciousness and Sentience Are Different
These three ideas are often mixed together, but they shouldn’t be.
Intelligence
A system’s ability to process information, learn, solve problems or adapt.
Consciousness
The existence of subjective experience or awareness.
Sentience
Generally, the capacity to have subjective experiences, particularly experiences with some positive or negative quality.
A system could potentially demonstrate learning without consciousness.
It could potentially perform an intelligent-looking task without having a human-like inner experience.
And something could potentially have some form of experience without possessing anything resembling human-level intelligence.
That’s why the question isn’t:
“Are the neurons smart?”
The deeper question is:
“Is there anyone—or anything—in there to experience what is happening?”
And science doesn’t currently have a simple answer to that question for increasingly complex biological systems.
The Most Uncomfortable Possibility
There is a particularly unsettling scenario hidden inside this technology.
Imagine that biological computing becomes dramatically more sophisticated over the next several decades.
Neural cultures become larger.
Their connections become more complex.
They interact with increasingly rich environments.
They receive sensory information.
They learn continuously.
They develop persistent internal states.
At some point, researchers might have to confront a possibility that today’s experiments do not require us to seriously consider:
What if the system has crossed a threshold where subjective experience becomes plausible?
We wouldn’t necessarily know the exact moment it happened.
There would be no flashing light.
No notification saying:
“Consciousness detected.”
And that’s what makes the ethical question so difficult.
We already struggle to determine consciousness in situations where communication is impossible. Researchers have developed methods to detect signs of awareness in some unresponsive patients, showing how difficult consciousness can be to infer even when the subject is unquestionably human.
With a biological computer, there may be no face, no voice and no established behavior that tells us what it is experiencing.
That doesn’t mean we should assume consciousness.
It means we should avoid pretending that the question is trivial.
For Now, the Science Is Much Less Dramatic
The most responsible conclusion is also the least sensational.
There is currently no good evidence that the neural cultures used in these biological computing systems are conscious in the human sense.
Current brain organoids and neural cultures remain limited models of nervous systems, and recent scientific and ethical analyses argue that present-day systems fall well short of the characteristics normally associated with consciousness.
So the headline:
“Scientists created a conscious brain in a computer.”
would go far beyond the evidence.
But the opposite extreme would also be premature:
“There is absolutely no possibility that biological computing could ever raise consciousness-related ethical concerns.”
As the systems become more complex, that question may become increasingly important.
And this is one of the strangest things about biological computing.
The technology could become valuable even if the neurons never become conscious.
It could help researchers study disease.
It could provide new models for testing treatments.
It could reveal new information about learning.
And it could potentially inspire new forms of energy-efficient computing.
That brings us to perhaps the most practical reason for pursuing this technology in the first place:
What can we actually do with living neural systems that could benefit science and medicine?
Part 11 complete.
Part 12 — Why This Research Could Be More Important for Medicine Than Computing
The most exciting future for biological computing may not actually be building computers.
It may be using living neural systems to understand the human brain.
That sounds like a subtle difference, but it could be enormous.
A conventional computer can simulate many aspects of biology. But a simulation is still a simulation. Human neurons are complicated living systems, and some of their behavior emerges from interactions between cells that are difficult to reproduce perfectly with software.
This is one reason researchers are developing human brain organoids and other neural models.
Instead of asking a computer to imitate a biological system from the outside, scientists can study human-derived neural tissue itself.
A New Way to Study Brain Disease
Consider a neurological disease.
Researchers might want to know:
- Which cells are affected?
- How does their electrical activity change?
- How does the disease alter communication between neurons?
- What happens when a potential treatment is introduced?
- Does the treatment restore normal function—or merely change the appearance of the cells?
Traditional laboratory models can answer some of these questions.
But they also have limitations.
Animal brains aren’t human brains.
And conventional two-dimensional cell cultures don’t reproduce the full three-dimensional organization of human neural tissue.
Brain organoids can occupy an interesting middle ground.
They aren’t complete brains, but they can reproduce certain aspects of human brain development and disease in a three-dimensional environment. Researchers are already using them to investigate neurological disorders and potential treatments.
That could be particularly valuable for diseases where human biology differs significantly from commonly used animal models.
What If We Could Test Drugs on Living Human Neural Systems?
This is where biological computing becomes especially interesting.
Imagine that researchers have a neural culture derived from human cells.
They can expose it to a drug.
Then instead of asking only:
“Did the cells survive?”
they could potentially ask:
“Did the neural network still function normally?”
That is a much richer question.
Neurons don’t merely exist.
They communicate.
They generate electrical activity.
They form networks.
They adapt.
They process information.
A treatment could therefore appear harmless when looking at simple cellular measurements while still disrupting the network’s functional behavior.
Researchers are beginning to investigate exactly this idea.
A 2025 study used the DishBrain system to examine how anti-seizure medications affected the information-processing performance of neural cultures. The researchers tested three anti-seizure medications—phenytoin, perampanel and carbamazepine—on human neuronal cultures with experimentally induced hyperactivity.
That is a major conceptual shift.
Instead of using neurons merely as samples to observe, researchers can potentially use neural networks as functional test systems.
The question becomes:
Does the treatment make the biological network work better?
This Could Matter for Neurological Diseases
Diseases such as Parkinson’s and Alzheimer’s are particularly difficult to model because the human brain is extraordinarily complex.
Researchers have already developed brain organoid models for investigating these diseases.
For example, human midbrain organoids are being studied as models of Parkinson’s disease, including applications in disease modelling, drug screening and therapy research.
Human brain organoids are also being investigated in Alzheimer’s research. In 2025, researchers reported using human brain organoids to investigate disease-related pathology and identify a potential therapeutic target.
The advantage is not that organoids perfectly reproduce a human brain.
They don’t.
Their value is that they can reproduce specific aspects of human biology that researchers want to investigate.
And that makes them useful experimental tools.
Patient-Derived Cells Could Make This Even More Powerful
Here’s where the technology gets particularly interesting.
Researchers can generate induced pluripotent stem cells—often called iPSCs—from human cells and then differentiate them into specialized cell types.
That creates the possibility of studying disease using cells carrying genetic characteristics associated with a particular patient or condition.
Now imagine combining that approach with biological computing.
Instead of testing a drug on an abstract laboratory model, researchers could potentially create neural tissue representing particular biological characteristics, connect it to an electronic interface and measure how its functional behavior changes when exposed to different treatments.
In principle, that could move medicine toward something more personalized.
Not:
“Does this drug work on the average laboratory model?”
But:
“How does this biological system respond to this treatment?”
We are not at the point where biological computers can replace clinical trials or provide routine personalized treatment decisions.
But the underlying direction is already being explored.
Reviews published in 2025–2026 describe human organoids as increasingly useful platforms for disease modelling, drug screening and toxicity studies, while also emphasizing substantial limitations involving reproducibility, complexity and standardization.
It Could Also Reduce Some Dependence on Animal Models
This is another potentially important application.
Animal models have played an enormous role in biomedical research.
But they aren’t perfect representations of human biology.
A drug that works in an animal may fail in humans because the underlying biology isn’t identical.
Human-derived organoids could provide another layer of testing between basic laboratory experiments and human studies.
That doesn’t mean:
Organoids replace animals tomorrow.
It means researchers can potentially build a larger toolbox of models that more closely represent human biology.
Recent reviews describe organoids as increasingly valuable human-relevant models for disease mechanisms, drug efficacy and toxicity, while also emphasizing that they still have important limitations and require further validation before broader adoption.
That last point is important.
A promising laboratory model isn’t automatically a clinically reliable one.
Researchers still have to prove that results obtained from these systems actually predict what happens in patients.
The Strange Connection Between Medicine and Computing
And this brings the whole story back to biological computing.
The same neural network that researchers might study as a computational system can also become a biological model.
You could have:
Neurons → electrical interface → computer → controlled environment
Then change one thing:
Add a drug.
Now you can measure how the biological network changes.
Change another:
Introduce a disease-related genetic mutation.
Now you can investigate how neural function changes.
Change another:
Alter the electrical environment.
Now you can study how neural networks adapt.
Suddenly, the “biological computer” isn’t just a weird alternative to silicon.
It becomes a laboratory instrument for studying the brain itself.
And that may ultimately be the technology’s most valuable contribution.
But There Is Still a Huge Problem
Biological systems are messy.
Two neural cultures won’t necessarily behave identically.
Organoids can vary from batch to batch.
Their development isn’t perfectly synchronized.
They don’t reproduce every feature of a real human brain.
And maintaining living tissue is far more complicated than running software on a silicon chip.
Researchers therefore face a major challenge:
standardization.
If two laboratories run the same experiment, they need to be confident that differences in their results aren’t simply caused by differences between their biological cultures.
Recent reviews specifically identify variability, reproducibility and scalability as important obstacles to the broader use of organoids in drug discovery.
Solving those problems could be just as important as making the biological networks more sophisticated.
Because eventually, the real test isn’t whether a biological computer can do something spectacular once.
It’s whether scientists can make it do the same experiment reliably thousands of times.
And if they can, biological computing could evolve from an astonishing demonstration into something much more useful:
a new scientific instrument.
But where does this technology go from here?
Will biological computers remain specialized laboratory tools?
Could they become a genuine alternative to certain types of computing?
Could organoids become dramatically more sophisticated?
And what happens if researchers eventually combine living neural networks with increasingly powerful artificial intelligence?
Those are the questions that determine whether the experiments we’re seeing today are just fascinating scientific curiosities—or the beginning of an entirely new computing era.
