BioComputing: when neural networks are made of real cells

For over seventy years, computing has followed an apparently linear trajectory: increasingly smaller transistors, increasingly faster processors, increasingly denser memories.

Modern artificial intelligence has accelerated this race, but it has also made one limitation evident: training and using increasingly large models requires growing amounts of energy, specialized infrastructure and cooling systems. From this arises a question that, until a few years ago, seemed like science fiction: what if, instead of imitating the brain only with software, we tried to directly use living neuronal cells as part of a computing system?

BioComputing, also called wetware computing or biological computing, explores precisely this frontier. In the laboratory, neurons cultivated on electronic interfaces can receive stimuli, generate signals and modify their behavior in response to the environment. Companies and research groups are already experimenting with programmable platforms in which biological neural networks communicate with silicon chips and software APIs. We are not looking at a replacement for the traditional computer, but at a new category of hybrid systems, still experimental, that could change the way we think about learning, energy efficiency and adaptation.

1. What BioComputing really is and why research is interested in it

A biological neuron does not function like a transistor. It does not simply execute a sequence of binary instructions: it receives thousands of signals, integrates information over time, modifies the strength of its connections and reacts dynamically to experience.

When many neurons are cultivated together, they form networks capable of producing complex electrical activity. By connecting these networks to microelectrodes, researchers can read the signals generated by the cells and send new stimuli in response.

The most interesting part is the closed loop. The software presents the biological network with a situation, measures the response and returns feedback. Over time, the network can adapt its activity. In this sense, the system is not “programmed” in the same way as a traditional application: rather, it is trained through interactions. It is a logic close to reinforcement learning, but the substrate that learns is not a numerical matrix on a GPU: it is living neural tissue.

The most advanced platforms today integrate neuronal cultures, sensors, electrical stimulation and software environments that make it possible to create reproducible experiments. The value for research is not only computational. These systems can also become experimental models for studying learning, plasticity, the effects of drugs or neurological diseases under controlled conditions.

2. The potential advantage: learning while consuming much less

The human brain is often cited as an extreme example of efficiency: it performs sophisticated cognitive activities with an energy consumption on the order of only a few tens of watts. The data centers required to train the largest artificial intelligence models, on the other hand, operate on completely different energy scales. The comparison is not direct, because the brain and GPUs perform different tasks and operations, but it highlights a fundamental characteristic of biology: processing and memory are deeply integrated.

In classical digital systems, data and computation continuously travel between memory and processor. This transfer has an energy cost. In biological networks, on the other hand, the state of the network is embedded in the connections themselves. Learning physically and chemically modifies the system. If this property could be exploited reliably, some classes of adaptive problems could be addressed with much greater efficiency than with conventional hardware.

However, it is essential to avoid premature enthusiasm. Biological neural networks are difficult to standardize, have intrinsic variability, require precise environmental conditions and still do not offer the predictability of an electronic system. BioComputing today is above all a research platform. Its true potential is measured by its ability to open new paths, not by replacing corporate servers tomorrow.

3. From the biological neural network to the hybrid computer

The most realistic future is not a laptop “made of brain”, but a hybrid architecture. Silicon will continue to manage reliable memories, communications, security, input/output and deterministic computation; the biological component could be used as a coprocessor for tasks in which adaptation, pattern recognition and response to incomplete contexts are more important than arithmetic precision.

Such a system requires genuine interface engineering. Digital data must be translated into stimuli that can be interpreted by cells, neuronal activity must be transformed into information useful to software, and shared measurement criteria must be established. Models are also needed to understand when the network has learned, when it is degrading and how reproducible its behavior is across different experiments.

This is where a topic very close to risk management comes into play: reliability. A biological component can change over time, react unexpectedly or lose performance. Before imagining industrial applications, it will be necessary to develop validation, monitoring and control methodologies comparable to those used for critical systems, but adapted to a living and non-deterministic element.

4. The ethical and governance questions that come before the market

BioComputing raises questions that the technology sector cannot postpone. A neuronal culture is not a complete brain and there is no evidence that current platforms possess consciousness. However, as biological systems become more complex, research will have to define clear limits, protocols and ethical criteria. What types of tissue can be used? What level of complexity requires more stringent ethical review? How is the life cycle of a culture documented?

To this are added aspects of biosafety, sample traceability, protection of experimental data and responsibility. If a hybrid system makes a decision based on variable biological behavior, who guarantees that the result can be verified? Governance will have to evolve together with the technology. It is a perfect example of innovation in which risk is not an obstacle to eliminate, but a dimension to be designed from the outset.

For Research & Development teams, the message is clear: observing BioComputing today means simultaneously studying computer science, neuroscience, adaptive systems, ethics and risk management. Even if commercial adoption is still far away, the skills developed now may become decisive when this technology leaves the laboratories.

Conclusion: a new definition of “computer”

BioComputing forces us to reconsider a deeply rooted belief: that computation must necessarily take place on silicon circuits. The first platforms with real neurons show that it is possible to create a dialogue between software and biological tissue and to use that dialogue for learning and adaptation experiments. The path toward reliable industrial applications remains long, but the direction is scientifically concrete.

For WRM Solutions, following this evolution means looking beyond the hype surrounding artificial intelligence and asking how infrastructures, reliability criteria and governance models will change. True innovation does not consist only in building more powerful systems: it consists in understanding in advance what new risks, opportunities and responsibilities will emerge when even the material with which we build computers changes.