Researchers have demonstrated that biological brain cells grown in a lab can learn to perform computing tasks, blurring the line between biology and artificial intelligence.
A dish of roughly 800,000 living brain cells has learned to play a simplified version of the classic video game Pong. The system, dubbed DishBrain by its creators at Cortical Labs, is not a simulation. It is a biological neural network, a cluster of human and mouse neurons grown on a microelectrode array, wired into a computer and trained using feedback signals. When the cells moved the paddle correctly, they received predictable electrical stimulation. When they missed, the signals became chaotic. Within minutes, the cells adapted their behavior to keep the stimulation stable.
This is the clearest demonstration yet that biological neurons outside a living body can exhibit goal-directed learning. The underlying research, published in the journal Neuron, shows these cells processed sensory input, adapted to feedback, and modified their activity to achieve a defined objective. That is, in functional terms, the same basic loop that underpins machine learning.
The AI industry faces a physical problem. Training large language models and other advanced systems requires enormous data centers running thousands of GPUs. That hardware consumes significant electricity, generates substantial heat, and costs hundreds of millions of dollars to build and operate. Sam Altman at OpenAI has publicly discussed the energy constraints facing advanced AI, and Nvidia's most recent generation of data center GPUs has become one of the most sought-after and expensive pieces of hardware on the planet.
Biological neurons operate on a radically different energy budget. The human brain, containing roughly 86 billion neurons, runs on about 20 watts of power, roughly the same as a dim lightbulb. A modern GPU cluster training a large foundation model can draw megawatts. If even narrow, specific computing tasks could be offloaded to biological substrates, the efficiency gains would be substantial. As the National Tribune recently highlighted in its coverage of the research, the convergence of neuroscience and computing is now moving from theory into tangible engineering experiments.
The implications extend beyond power consumption. Biological neural networks process information differently than silicon chips. Neurons fire in complex, asynchronous patterns, adapt structurally over time, and exhibit forms of noise tolerance that digital systems struggle to replicate. For tasks involving pattern recognition, real-time adaptation, and learning from small datasets, biological systems have inherent advantages that conventional architectures cannot easily match.
The Road From Lab to Industry
Cortical Labs, the Melbourne-based startup behind DishBrain, has attracted venture backing to explore commercial applications. The company has raised seed funding to develop what it describes as biological processing units, chips that integrate living cells with traditional semiconductor components. The immediate goal is not to replace GPUs or run consumer software, but to target specific applications in robotics, autonomous systems, and adaptive control where learning speed and energy efficiency are critical constraints.
The challenges, however, are considerable. Keeping neurons alive outside a controlled laboratory environment requires precise temperature regulation, nutrient delivery, and contamination prevention. Scaling from 800,000 cells to systems capable of handling commercially useful workloads presents significant engineering problems. The cells also have a finite lifespan, which means any biological computing system must account for degradation, replacement, and maintenance cycles that silicon does not require.
There are also questions about reproducibility and reliability. Biological systems are inherently variable in ways that digital chips are not. Two batches of cultured neurons may develop different internal connections, respond to stimuli at different speeds, or produce slightly different outputs. For applications requiring deterministic, repeatable results, that variability is a problem. For applications that prioritize adaptability over precision, it could be an advantage.
Despite these hurdles, the broader trajectory is worth watching. Organizations like the European Human Brain Project and the BRAIN Initiative in the United States have spent years mapping neural circuitry and developing tools to interface with biological tissue. Cortical Labs is applying those tools to a specific engineering problem, and their progress suggests that biological computing, once confined to speculative academic research, is entering an early commercial phase. Companies building AI infrastructure, particularly those investing heavily in custom silicon, should monitor whether hybrid biological-digital architectures can move beyond novelty into viable niche applications.
