Living brain cells wired into organoid-on-a-chip biocomputers can now learn to drive robots, thanks to an open-source intelligent interaction system called MetaBOC. This groundbreaking project aims to integrate human brain cells with artificial bodies.

Biocomputing is one of the most astonishing frontiers in emerging technology, enabled by the fact that our neurons communicate using electrical signals, the same language as computers. Human brain cells, grown in large quantities onto silicon chips, can receive electrical signals from a computer, process them, and respond. More impressively, they can learn. The concept was first demonstrated in the DishBrain project at Monash University, Australia. Researchers grew about 800,000 brain cells onto a chip, placed it into a simulated environment, and observed as this biocomputer learned to play Pong within five minutes. This project was swiftly funded by the Australian military and evolved into a company called Cortical Labs.

Cortical Labs’ Chief Scientific Officer, Brett Kagan, revealed that even at an early stage, human neuron-enhanced biocomputers learn much faster and use significantly less power than today’s AI machine learning chips, showing “more intuition, insight, and creativity.” Our brains, consuming just 20 watts, run nature’s most powerful computers.

“We’ve done tests against reinforcement learning,” Kagan explained, “and we find that in terms of the number of samples needed before meaningful learning begins, biological systems, even in their basic form, outperform the best deep learning algorithms. That’s remarkable.”

One challenge, aside from ethical concerns, is that the “wetware” components need to be kept alive, requiring nutrition, hydration, temperature control, and protection from pathogens. In 2023, Cortical’s record for maintaining these components was about 12 months.

Similar projects have emerged since, such as Indiana University’s “Brainoware” organoid, where brain cells self-organize into a three-dimensional structure before being connected to electrodes, and Swiss startup FinalSpark, which uses dopamine as a reward mechanism for its Neuroplatform biocomputing chips.

The MetaBOC project represents a significant advancement. Researchers from Tianjin University’s Haihe Laboratory of Brain-Computer Interaction and Human-Computer Integration, along with teams from the Southern University of Science and Technology, have developed an open-source software system. MetaBOC acts as an interface between brain-on-a-chip biocomputers and electronic devices, enabling brain organoids to perceive the world through electronic signals, interact with it, and learn tasks.

The Tianjin team uses ball-shaped organoids, similar to Indiana University’s approach, as their three-dimensional structure facilitates complex neural connections. These organoids are grown under low-intensity focused ultrasound stimulation, which enhances their cognitive capabilities. MetaBOC also employs AI algorithms to communicate with the brain cells’ biological intelligence.

The researchers highlight robotics as a key integration target. They demonstrate that a brain-on-a-chip biocomputer can now learn to drive a robot, mastering tasks such as obstacle avoidance, target tracking, and using robotic arms and hands to manipulate objects. Since the brain organoid perceives the world through electrical signals, it can train in a simulated environment, minimizing the risk to its biological components.

It’s important to note that the exposed brain organoids in the robot images are mockups intended to illustrate future applications, rather than actual brain-controlled prototypes. A more realistic representation of these biocomputers is provided by images from Cortical Labs, showing what these brain-on-chip systems might look like in practical use.

By Impact Lab