Sean's Blog

Biological Computing

Edit on GitHub

Biological computing using living neurons has crossed from science fiction into commercial reality. The first biological computers are now available for purchase, using lab-grown human brain cells to process information with high energy efficiency. These systems represent a fundamental shift from traditional silicon-based computing toward hybrid biological-digital architectures that leverage billions of years of neural evolution.

This emerging field, termed Organoid Intelligence, uses the computational power of actual brain tissue grown from human stem cells. Unlike neuromorphic chips that simply mimic neural behavior, biological computers employ real neurons that form functional networks, learn through experience and process information through natural synaptic connections.

Basic Functionality

Biological computing systems work by growing brain organoids from human induced pluripotent stem cells in laboratory conditions. These three-dimensional neural cultures, typically 0.5 to 5 millimeters in diameter, contain between 10,000 and 800,000 living neurons that spontaneously develop electrical activity and complex oscillatory behavior similar to human brain tissue.

The technical architecture integrates these living networks with multi-electrode arrays containing 8 to 384 electrodes that both stimulate and record neural activity. Advanced microfluidic life support systems maintain organoid viability for up to 100 days, a significant improvement from initial lifespans measured in hours. Communication occurs through electrical stimulation and chemical signals using neurotransmitters like dopamine and serotonin.

The systems leverage natural neural plasticity mechanisms for learning. Correct responses receive predictable electrical patterns while errors trigger chaotic stimulation, mimicking reward-based learning in biological brains. Information storage occurs through structural and functional changes in neural networks, enabling pattern recognition and memory formation through synaptic plasticity.

Early Adoption

Biological computing has moved beyond laboratory demonstrations into practical applications across healthcare, research, and technology sectors, though adoption remains in early stages.

Pharmaceutical companies are exploring biological computing for drug discovery and neurological disorder research. The technology enables disease modeling for conditions like Alzheimer's, epilepsy, and Parkinson's disease using human-relevant neural responses. This approach offers alternatives to animal testing while providing more accurate data for pharmaceutical development.

Healthcare applications include medical device development, brain-computer interface research, and personalized medicine approaches. The systems enable real-time monitoring of drug effects on neural networks and enhanced understanding of brain disease mechanisms.

Current limitations present significant adoption barriers. Biological components survive only 100 to 180 days on average, requiring regular replacement. Systems remain limited to thousands or hundreds of thousands of neurons, far below brain complexity levels. High costs, with CL1 units priced at $35,000, limit accessibility for many potential users.

Regulatory and ethical concerns constrain adoption. Ongoing debates about potential consciousness in biological computing systems require new ethical frameworks governing the use of human neural tissue. Using biological neural networks opens up many hairy ethical concerns and has an eerie similarity to the plot of The Matrix. Before we understand the emergence of (self-)cosciousness, we should not rush the adoption of this bio-technology as it may be an inherently evil application.

Applications for Artificial Intelligence

Biological computing systems may offer critical advantages over classical deep learning approaches, addressing fundamental limitations. These living neural networks demonstrate superior generalization abilities, natural robustness against adversarial attacks and adaptive learning mechanisms.

Generalization represents the most significant advantage of biological neurons. Classical deep learning models often fail catastrophically when encountering data distributions different from their training sets. A neural network trained to recognize cats in photographs might completely fail when presented with cartoon drawings of cats. Biological neurons, however, demonstrate remarkable ability to extract meaningful patterns and apply learned concepts to novel situations through natural abstraction mechanisms developed over millions of years of evolution.

Research at Johns Hopkins University showed that brain organoids could adapt their learned behaviors to new environmental conditions within hours, while equivalent artificial neural networks required complete retraining. The biological systems maintained performance across different stimulus patterns by leveraging inherent plasticity mechanisms that classical AI lacks. Adversarial robustness emerges naturally in biological systems. Traditional deep learning models are notoriously vulnerable to adversarial examples, where tiny imperceptible changes to input data cause complete misclassification. These attacks exploit the brittle decision boundaries learned by artificial networks. Biological neurons, by contrast, process information through noisy, analog mechanisms that naturally resist such manipulations.

FinalSpark demonstrated this robustness by showing that organoid-based pattern recognition systems maintained accurate classification even when input signals were corrupted with noise levels that completely broke equivalent digital neural networks. The biological systems' inherent stochasticity and redundant processing pathways provide natural defense against adversarial perturbations.

Continual learning without catastrophic forgetting represents another key advantage. Classical AI systems suffer from catastrophic forgetting, where learning new tasks erases previously acquired knowledge. This limitation requires complex architectural modifications and careful training procedures to overcome. Biological neurons naturally support lifelong learning through synaptic homeostasis and distributed memory representations.

Cortical Labs' experiments showed that organoids could sequentially learn multiple pattern recognition tasks while retaining performance on earlier learned behaviors. The biological systems demonstrated graceful interference patterns rather than the complete knowledge destruction seen in traditional neural networks.

Key Startups

Cortical Labs, founded in 2019 as a spinout from Monash University, leads commercial biological computing development. The Australian company pioneered the DishBrain technology that learned Pong. CEO Dr. Hon Weng Chong and Chief Scientific Officer Brett Kagan have positioned the company at the forefront of wetware-as-a-service offerings. Cortical Labs launched the CL1 in March 2025 - the world's first commercial biological computer. Priced at $35,000 per unit, the system contains 800,000 lab-grown human neurons with integrated life support maintaining cell viability for six months.

FinalSpark, established in 2014 by Dr. Fred Jordan and Dr. Martin Kutter, developed the first remote-access biological computing platform. The Swiss company, currently seeking $50 million in Series A funding, has created a unique dopamine-based reward system for training organoids and extended operational lifespans from hours to over 100 days.

Koniku, founded in 2015 by Nigerian neuroscientist Dr. Oshiorenoya Agabi, focuses on biological detection systems. With $1.8 million in funding and partnerships with Airbus, the company developed Konikore processors that combine biological neurons with silicon for detecting explosives, drugs, and diseases. Their systems achieve parts-per-billion sensitivity with response times under 10 seconds.

Conclusion

Biological computing using living neurons has transitioned from theoretical research to commercial reality, offering a sustainable path toward more intelligent, adaptive and energy-efficient computing systems. While technical challenges around lifespan, scalability, and standardization remain significant, the field demonstrates extraordinary potential for revolutionizing drug discovery, neuroscience research, and artificial intelligence applications.

The convergence of stem cell technology, advanced electrode interfaces and artificial intelligence has created a new computational paradigm that leverages natural neural evolution. As these systems mature and scale, they promise to address the energy crisis in computing while unlocking new forms of machine intelligence that complement traditional silicon-based approaches. The first commercial biological computers mark the beginning of a transformation that could fundamentally reshape how we approach computation in the coming decades, though the ethics of it all remains unclear at best and evil at worst.

References

Generalization and Biological vs Artificial Neural Networks

Adversarial Robustness in Biological vs Artificial Systems

Catastrophic Forgetting and Continual Learning

Energy Efficiency and Biological Computing

#AI