Researchers at Cortical Labs in Melbourne published results in 2022 showing that clusters of human neurons grown in a laboratory — called brain organoids — could learn to play a simplified version of Pong. The paper, published in Neuron, demonstrated that biological neural networks can exhibit goal-directed learning behaviour, igniting a debate about consciousness, biocomputing ethics, and the future of AI hardware.

What the Research Actually Shows

The organoids, containing approximately 800,000 neurons, were grown on a multi-electrode array that allowed researchers to both stimulate the neurons with electrical signals and record their activity. The Pong game was "explained" to the organoid through patterns of electrical stimulation representing the ball's position. When the organoid's neural activity corresponded to correct paddle movement, it received a reward signal; incorrect movement produced a different signal.

The organoids improved their Pong performance over time — a result the researchers attributed to the inherent plasticity of biological neurons. The paper was peer-reviewed and published in Neuron; the results have been discussed extensively in the neuroscience community, though the interpretation of what "learning" means in this context remains debated.

The Ethical Questions

The results have prompted questions from bioethicists. Do these organoids experience anything? Are they conscious in any meaningful sense? The researchers are careful to note that the organoids lack the complexity of a human brain by many orders of magnitude. But the question of where the line is — and who gets to draw it — is now part of the scientific conversation.

The Computing Implications

Biological neurons are extraordinarily energy-efficient compared to silicon. A human brain performs roughly 10^15 operations per second on approximately 20 watts of power. The most efficient AI accelerators require kilowatts for comparable performance. If biocomputing can be scaled and controlled, it could represent a fundamentally different approach to AI hardware — though scaling from thousands of neurons to billions remains an unsolved engineering problem.

The 2026 State of the Field

Since the original Cortical Labs paper, several research groups have published follow-up work. Johns Hopkins researchers demonstrated that organoids could learn to distinguish between two different audio patterns — a more complex task than Pong. A team at the University of Melbourne showed that organoids could retain learned behaviours for up to 72 hours, suggesting a form of short-term memory. None of these results have been replicated at scale, and the field remains in early exploratory stages.

The practical path from laboratory organoids to useful computing hardware is long and uncertain. Current organoids are fragile, difficult to reproduce consistently, and require specialised laboratory conditions. The interface between biological tissue and electronic hardware — the multi-electrode array — is a significant engineering bottleneck. And the question of how to "program" a biological neural network in a reliable, reproducible way remains largely unsolved.

Regulatory and Ethical Framework

The International Society for Stem Cell Research updated its guidelines in 2021 to address organoid research, but the rapid pace of development has outpaced regulatory frameworks. Most jurisdictions do not have specific regulations governing organoid intelligence research. The UK's Nuffield Council on Bioethics published a report in 2023 recommending a precautionary approach and calling for international coordination on ethical standards — but no binding framework has emerged.

The Path Forward

The most likely near-term application of organoid computing is not general-purpose computation but specialised co-processing — using biological neural networks for specific tasks where their energy efficiency and learning capabilities provide advantages over silicon. Pattern recognition, anomaly detection, and adaptive filtering are the most promising candidates.

Several research groups are exploring hybrid architectures that combine silicon and biological components: silicon handles the deterministic, high-speed computation that it does best, while biological components handle the adaptive, energy-efficient pattern recognition that neurons do best. These hybrid systems are still in early research stages, but they represent a more pragmatic path to practical biocomputing than attempting to replace silicon entirely.

Australia's Role in Biocomputing Research

Australia is punching above its weight in organoid intelligence research. Cortical Labs, the Melbourne company behind the original Pong paper, has attracted international attention and investment. The University of Melbourne's BioMedical Engineering department has active research programmes in neural-electronic interfaces. The Australian Research Council has funded several projects in this space through its Discovery Programme. For a country of 26 million people, Australia's contribution to this emerging field is disproportionately significant — and positions it well to benefit commercially if the technology matures.

Sources & Further Reading