Biocomputing: computers made from brain cells

Computers are usually made of silicon, transistors and memory chips. In biocomputing, however, living tissue comes into play. Human nerve cells or three-dimensional brain organoids are connected to electronic interfaces. This allows researchers to feed in electrical signals, measure cell responses and influence neuronal networks through feedback. However, the goal is not a human brain in a computer. Rather, biological networks are intended to take on special learning, research and signal processing tasks. Projects like DishBrain, Brainoware and the commercial CL1 show initial practical approaches. Nevertheless, the technology is still in its early stages and raises not only technical but also fundamental ethical questions.

The most important thing in brief

Biocomputing uses living nerve cells as part of a hybrid system of biology, electronics and software.
Organoid intelligence, or OI for short, focuses on three-dimensional brain organoids made from human stem cells.
Electrode arrays transmit electrical stimuli to the cells and read out their reactions.
The CL1 works with cultured neurons on a silicon chip, but not with a three-dimensional brain organoid.
The biggest hurdles are reproducibility, scaling, durability, reliable performance metrics and ethical rules.

What biocomputing and organoid intelligence mean

Biocomputing is a generic term for information processing using biological systems. These can include DNA molecules, proteins, individual cells or neural networks. In the area of organoid intelligence, the focus is on living nerve cells. The term “wetware” separates this biological component fromHardwareand software. However, it is not a complete brain in a petri dish. The often used term “mini brain” is also scientifically misleading. Brain organoids only replicate selected properties and developmental processes of brain tissue.

Organoid intelligence is an interdisciplinary research field. It combines stem cell biology, neuroscience, bioengineering, computer science and machine learning. A research roadmap presented in 2023 describes how three-dimensional brain organoids could be coupled with input and output interfaces. In the long term, such systems are intended to replicate learning and memory processes. The approach should also help to study brain diseases in more detail. At the same time, the developers emphasize that ethical questions must be part of the research from the start. The technical term OI therefore does not stand for an existing human-like intelligence, but rather for a research goal. That goes from the basicTechnical article on organoid intelligenceout.

Expression	Short definition	Technical role
CL1	Cortical Labs’ commercially available biocomputing platform with cultured neurons on a silicon chip.	Enables experiments with neural information processing, feedback and learning.
Electrode array	Arrangement of numerous microelectrodes that can send and receive neuronal signals.	Forms the interface between living cells and electronics.
Organoid	Three-dimensional cell structure that is grown from stem cells and recreates certain tissue properties.	Serves as a biological model as well as a possible computing substrate.
Wetware	Living biological material that processes information or is part of a technical system.	Adds a biological level to classic hardware and software.
OI	Abbreviation forOrganoid Intelligenceor organoid intelligence.	Refers to the use of brain organoids for learning, research and biocomputing.

How living hardware is created from stem cells

Many brain organoids arise from induced pluripotent stem cells, iPS cells for short. To do this, researchers can, for example, take skin or blood cells from an adult. These cells are first returned to a stem cell-like state. Growth factors then cause them to develop into nerve cells and other cell types. Under appropriate culture conditions, the cells organize themselves into three-dimensional structures. This can result in synaptic connections and spontaneous electrical activity patterns. TheMax Delbrück Centerexplains such organoids as models that make certain processes of the human brain accessible in the laboratory.

However, a brain organoid is not a reduced human brain. It lacks, among other things, a complete anatomical organization, body, sensory organs and a normal blood supply. In addition, a single organoid usually only represents certain cell types or properties of a brain region. Its development does not completely follow the processes in the human body. This limits both medical comparisons and possible computer applications. At the same time, this simplification is useful for research. This allows scientists to observe individual development processes, diseases or reactions to active ingredients in a controlled manner.

Larger organoids require a reliable supply of oxygen and nutrients. Without blood vessels, the nutrient solution only reaches a limited extent inside the tissue. Therefore, larger areas of cells in the center can die. Microfluidic systems are intended to solve this problem. They supply nutrients and growth factors and remove waste products. The OI research roadmap names such artificial supply systems as a prerequisite for larger and longer-lasting organoids. Early concepts envisaged systems with several million cells, while smaller models are often well below this size. However, more cells do not automatically mean more intelligence, because their networking and controlled control are also crucial.

How living nerve cells process information

Nerve cells communicate through electrical and chemical signals. If the voltage on your cell membrane changes sufficiently, an action potential is created. This signal influences other cells via synapses. Repeated activity can change the strength of certain compounds. This synaptic plasticity is an important basis of biological learning. Biocomputing attempts to make precisely this adaptability technically usable. No program is written directly into the cells like with a normal processor.

The connection to the electronics is usually made via a microelectrode array. Some electrodes send electrical impulses to the cell network. Others measure when and where neurons become active. Software first translates digital information into stimulation patterns. The cell responses are then recorded, filtered and converted back into digital values. This creates a closed feedback loop. In the future, flexible and three-dimensional electrode structures should reach a larger part of an organoid than flat chips. One presented by the EUKirigami concept for electrode arraysshows how an electrode cover can be placed around three-dimensional tissue.

“Training” has a different meaning in this context than in a classic AI model. The researchers cannot send error correction directly through the tissue using backpropagation. Instead, timed stimuli and feedback change the dynamics of the cell network. Beneficial responses can be confirmed by predictable signals. Inappropriate reactions, on the other hand, can be associated with unpredictable or changed stimulus patterns. Whether this has really resulted in learning must then be checked using objective performance tests. A mere change in electrical activity is not sufficient evidence. Fatigue, stress, cell growth or fluctuations in the measurement can also change the signals.

Another approach is so-called reservoir computing. The biological network serves as a dynamic reservoir that converts time-dependent input signals into complex activity patterns. Often only the digital evaluation layer is specifically trained. The cell network itself does not have to carry out a clearly defined calculation rule. Rather, it provides its nonlinear dynamics and short-term “memory”. This makes the method particularly suitable for speech, time series and other signal tasks. Nevertheless, this form of computing cannot yet be equated with freely applicable artificial intelligence.

What differentiates CL1, DishBrain and Brainoware

The CL1 is offered by Cortical Labs as a programmable biological computer. Inside, living neurons grow in a nutrient solution on a silicon chip. This chip sends electrical impulses and records neuronal responses. The associated Biological Intelligence Operating System, or biOS for short, creates a digital environment and translates its states into stimulus patterns. The neurons’ responses can, in turn, change the simulated environment. According to the manufacturer, the CL1 also contains recording technology, controls and its own supply system. This is intended to maintain the cell culture under suitable conditions for up to six months. The current product description also mentions a touchscreen, USB ports and a cloud connection for experiments. This information comes from theofficial CL1 product description.

However, it is important to differentiate from organoid intelligence. The CL1 uses a cultivated neural network on a flat chip surface. It is therefore not a computer with a three-dimensional brain organoid. Terms such as “Synthetic Biological Intelligence” are also initially names of the manufacturer. They do not replace independent proof of performance. When it was introduced, the device was priced at around $35,000. A cloud version is intended to give research groups access without having to operate a full cell culture laboratory themselves.

The scientific predecessor DishBrain connected human and mouse-derived neurons with a simulated Pong environment. The system transmitted information about the ball’s position as electrical stimuli. The activity of the cell cultures then influenced the virtual racket. In certain experimental conditions, the performance of the cultures improved. The 2022 inNeuronpublishedDishBrain studyinterpreted this as an indication of targeted adaptation. However, it does not follow that the cells understood the game like a human. The experiment also provides no evidenceconsciousnessof cell culture.

Brainoware is another approach. A three-dimensional brain organoid was actually used here. A high-resolution electrode array transmitted spatially and temporally structured signals. The system used the dynamics of the organoid as a biological reservoir. In experiments, it was used for language classification and nonlinear equation prediction. The results showed that biological networks can in principle be used for reservoir computing. However, they have not yet proven a universally applicable organoid computer. The details were published in 2023Nature Electronicspublished.

system	Biological basis	Technical approach	Level of development
DishBrain	Flat cultures of human and animal nerve cells	Closed-loop feedback with a Pong simulation	Scientific proof of concept
CL1	Neurons cultured on a silicon chip	Programmable stimulation, measurement and integrated cell supply	Commercial research platform
Brainoware	Three-dimensional human brain organoid	Biological reservoir computing via an electrode array	Experimental research prototype
OI concept	Scalable and long-term supply of brain organoids	Learning-enabled organoid computer interfaces	Long-term research program

What opportunities biocomputing actually offers

An important area of application is research into neurological diseases. Organoids can be created from cells from specific patients. As a result, they carry part of the donor’s individual genetic characteristics. Researchers could investigate how epilepsy, dementia or developmental disorders affect neuronal networks. Drugs could be tested directly on human cell models. This could supplement certain animal experiments or reduce them in the long term. However, a complete replacement of all animal models cannot be automatically derived from this.

Biocomputing could also provide new insights into learning and memory. In artificial neural networks, only mathematical models of biological nerve cells are active. A living cell network, on the other hand, shows real synaptic plasticity, biochemical regulation and self-organization. Researchers can observe how drugs, diseases or environmental conditions change these processes. This makes biocomputing primarily a new type of neuroscientific measuring instrument. This benefit is likely to become practically relevant sooner than a powerful biological universal computer.

The potential energy efficiency is also generating great interest. The human brain handles complex tasks with a power consumption of approximately 20 watts. However, it cannot be directly deduced from this that an organoid computer would automatically be more economical than a modern chip. In addition to the cells, today’s systems require pumps, temperature control, sensors, measuring amplifiers and digital evaluation. In addition, there is sterile production and a continuous supply of nutrients. Fair comparisons must therefore consider the entire system. Reliable energy benchmarks for typical computing tasks are still missing.

In the long term, biological networks could serve as special accelerators. Applications would be conceivable for time-dependent signals, adaptive controls or the recognition of complex biological patterns. This would not make normal PCs, data centers and graphics processors superfluous. Exact mathematics, data storage and reproducible program execution remain strengths of digital technology. A realistic future model is therefore a hybrid computer. The digital side takes over control and evaluation, while the wetware handles selected adaptive tasks. Organoid intelligence would therefore be more of a biological coprocessor than a replacement for CPU, GPU or cloud.

Limits, ethical questions and realistic outlook

Biological hardware is more difficult to standardize than a silicon chip. Cell lines differ in their genetic origin, developmental state and culture conditions. Even two organoids from the same cell line can form divergent networks. In addition, living cells change during operation. They grow, form new connections, age or die. A task can therefore produce different results after several weeks than at the beginning. Such fluctuations would be a significant problem for industrial applications.

Added to this is the limited input and output. A human brain has billions of nerve cells and an enormous number of synaptic connections. An electrode array only records a small section of this. In three-dimensional organoids, flat electrodes primarily reach the outer surface. Flexible sheaths and implantable probes are intended to improve coverage. However, they can affect or damage the delicate tissue. This means that the interface often becomes the actual bottleneck of the system.

There are also no recognized standards for computing performance. FLOPS, clock frequency and memory size cannot be meaningfully transferred to a living cell culture. Instead, researchers need standardized tasks for learning time, accuracy, stability, energy requirements and transferability. These tests would also have to take into account control cultures without training. This is the only way to distinguish true adaptation from random fluctuations. Independent repeats by other laboratories are equally important. Spectacular demonstrations with video games do not replace reproducible benchmarks.

Ethically, the focus is primarily on the possible development of characteristics of sensation or consciousness. There is currently no reliable evidence that today’s brain organoids have human consciousness or experience pain. Still, the question could become more pressing as systems become more complex, long-lasting and more connected to sensors. Researchers would then have to define which activity patterns are considered warning signs. Termination criteria for experiments would also be required. The OI research roadmap therefore proposes “embedded ethics”. Experts from research, ethics and society continuously support technical developments.

Other questions concern the people whose cells are used. Genetic and possibly medically relevant information can be derived from iPS cells. A donor might have consented to a disease study but not to the use of their cells in an adaptive computer system. Therefore, consent, data protection and possible commercial exploitation must be clearly regulated. It is also unclear who owns trained biological models and their results. TheNuffield Council on BioethicsAt the same time, he warns against using terms like “mini brain” to create false ideas about the level of development.

New perspective:The future of biocomputing may depend less on the number of nerve cells than on a new concept of “biological versioning.” With normal software, a specific version can be archived and later run identically. Living hardware, on the other hand, changes daily. A complete experimental data set would therefore not only have to contain program code. He would also need to document cell line, donor background, age of culture, culture medium, temperature, electrode contact and activity history. Biocomputers therefore require a kind of biological life course. Without this CV, results could neither be reliably compared nor later reproduced.

Conclusion

Biocomputing does not turn living nerve cells into a thinking PC. But the combination of cell culture, electrodes and software opens up a remarkable field of research. In the short term, it should primarily improve medications, disease models and basic research. In the long term, biological networks could take on special learning and signaling tasks. However, reliable benchmarks, reproducible cell cultures and strict ethical guidelines are crucial. The real breakthrough would therefore not be the artificial mini-brain, but rather a controllable hybrid computer that uses biological adaptability sensibly, economically and responsibly. Until then, it remains fascinating research, not a computer revolution.

Bewerte den Beitrag hier!

[Total: 0 Average: 0]

Nico Nuss

Website

Author Nico Nuss has been working on mobile computing and automation software since 2001. Drawing on his experience and strong interest in future technologies, he focuses on robotics and AI.