DishBrain Pong: Living Neurons Learn to Play and Drive Neuromorphic Computing Forward

Overview

In this video, researchers show living human neurons learning to play a video game, Pong, using a multi-electrode array and electrical feedback rather than traditional code or hardware. The field of neuromorphic computing is described as a brain inspired approach that could redefine how machines think, learn, and interact with the real world.

• DishBrain demonstrates plasticity in a biological information processing device.
• Random noise input with predictable reward signals guided learning, a practical test of the free energy principle.
• Commercial biology based computing with CL1 highlights new directions for AI, neuroscience and drug discovery.

Introduction to neuromorphic computing

The video introduces neuromorphic computing as a design philosophy that copies the brain's architecture, where memory and computation are not strictly separated and information is carried by spikes rather than a central clock. It contrasts this with traditional von Neumann systems, emphasizing energy efficiency, parallel processing, and the brain like ability to learn from small data samples.

The DishBrain Pong Experiment

At the core of the video is the DishBrain experiment, where living human neurons, grown from stem cells, are cultured on a multi electrode array and taught to play Pong. Neuronal activity serves as input to the game, while electrical stimulation provides feedback. The team tested a theory known as the free energy principle, exploring how high entropy, random feedback can improve performance when paired with consistent correct responses. The results showed the neurons learned to hit the ball more reliably, with learning speed that surprised the researchers and underscored the plasticity of neural systems.

Biology as a Computer: architecture and efficiency

The discussion pivots to how brains process information differently from silicon chips: processing and memory are inseparable in the brain, signaling via spikes, and learning via synaptic plasticity rather than backpropagation through layers. This leads to neuromorphic systems that use artificial neurons and synapses to emulate biological computation, offering orders of magnitude greater energy efficiency for tasks like pattern recognition and adaptive response in dynamic environments.

From Pong to a Commercial Biological Computer

The narrative then shifts to Cortical Labs, the company behind DishBrain, which evolved Pong experiments into a commercial platform named CL1. The CL1 device grows living neurons on a multi electrode array, with inputs and rewards delivered electrically. The platform targets neuroscience research, drug discovery, and studying learning and memory, promising a cloud like access model where researchers can run experiments remotely. A central challenge remains: identifying effective reward signals that reliably drive learning in living tissue and ensuring the system can be tuned for higher functionality over time.

Broader context: players, potential and challenges

The video positions neuromorphic computing as a major rethink of computing since the transistor, with other approaches including silicon chips like Intel LIH and IBM TrueNorth, as well as photonic and biotech oriented paths. It also discusses how neuromorphic systems could advance medicine by enabling more accurate memory models, drug screening, and disease modeling. The narrative cautions that biology is inherently messy and unpredictable, thus requiring careful design of interfaces, feedback, and safety considerations as the technology matures. The broader question remains how such living processors will integrate with existing AI ecosystems and what new capabilities they unlock for edge devices, robotics, and space exploration.

Conclusion and future directions

Neuromorphic computing using living neurons is framed not as a replacement for silicon, but as a complementary approach that could transform AI, neuroscience, and memory related research. The CL1 platform, ongoing research into reward signaling, and cross disciplinary advances in genetics, neurobiology, and computing hint at a future where computers think with a biological substrate as well as through silicon, potentially changing medicine and our understanding of learning itself.