In a groundbreaking study, researchers have demonstrated that networks of living brain cells grown in a laboratory dish can learn to play a video game more efficiently than some of the most advanced artificial intelligence systems. The research, published in Cyborg and Bionic Systems, details how a system nicknamed “DishBrain” learned to play a simplified version of the classic game Pong with greater “sample efficiency”—meaning it learned more from less experience—than sophisticated deep reinforcement learning (RL) algorithms. This work, led by the Melbourne-based startup Cortical Labs, provides the first direct performance comparison between biological neurons and digital AI under the same time-limited conditions, suggesting that biological systems possess an innate advantage in rapid, adaptive learning.

The experiment was conducted using the CL1, a commercial biological computer that merges lab-cultivated human brain cells with silicon microchips. These cells were grown on a high-density multi-electrode array, which can both send electrical signals to the neurons and read their responses. The researchers created a “closed-loop” environment where the collective activity of the brain cells controlled the Pong paddle. Feedback on whether the paddle hit or missed the ball was sent back to the cells as electrical stimuli, allowing the network to reorganize and improve its performance dynamically. The study found that while AI often needs millions of training attempts, the DishBrain system showed significant improvement and network plasticity in a fraction of the time.

This achievement is part of a broader vision for a new class of information processing devices. In a related perspective paper in Cell Biomaterials, the researchers propose a framework to guide this emerging field, distinguishing between two key approaches. The first, Organoid Intelligence (OI), is a “top-down” method that aims to grow mini-brains that mimic natural brain structures. In contrast, the team formalizes a new “bottom-up” approach called Bioengineered Intelligence (BI), which involves constructing custom neural circuits from the ground up to create highly controllable and task-optimized devices. This research represents a major step forward for the BI pathway.

The development of the CL1 platform, which enables this research and was described in Nature Reviews Bioengineering, marks a critical milestone in harnessing the power of living neurons. “While substantial advances have been made across the field of AI in recent years, we believe actual intelligence isn’t artificial. We believe actual intelligence is biological,” commented Brett Kagan, Chief Scientific Officer at Cortical Labs, in a press release. This research not only advances our fundamental understanding of how brain cells process information and learn but also opens the door to creating powerful, sustainable, and efficient new forms of biologically integrated computing that could one day surpass the capabilities of silicon-based machines alone.

References

• Cortical Labs. (2025, August 12). Brain cells learn faster than machine learning, research reveals. Tech Xplore; Cortical Labs. https://techxplore.com/news/2025-08-brain-cells-faster-machine-reveals.html

• Kagan, B. J. (2025a). The CL1 as a platform technology to leverage biological neural system functions. Nature Reviews Bioengineering. https://doi.org/10.1038/s44222-025-00340-3

• Kagan, B. J. (2025b). Two roads diverged: Pathways toward harnessing intelligence in neural cell cultures. Cell Biomaterials, 100156. https://doi.org/10.1016/j.celbio.2025.100156

• Khajehnejad, M., Habibollahi, F., Loeffler, A., Paul, A., Razi, A., & Kagan, B. J. (2025). Dynamic network plasticity and sample efficiency in biological neural cultures: A comparative study with deep reinforcement learning. Cyborg and Bionic Systems, 6, 0336. https://doi.org/10.34133/cbsystems.0336