Australia develops chip with human brain

Experts from Cortical Labs, Australia have successfully studied how to integrate biological neurons with traditional information processing chips.

In a recent announcement, Cortical Labs - a company based in Melbourne, Australia introduced the technology of integrating biological neurons with information processing chips. 'DishBrain', as the name given to the group of brain cells grown in a laboratory, has been 'taught' by Cortical Labs to play the Pong game.

"DishBrain is a breakthrough technology that helps us understand the intelligent properties of neurons," Hon Weng Chong, co-founder of Cortical Labs, told Nikkei Asia.

Cortical Labs hopes that the application of human brain cells will make high-power chips much more energy-efficient than previous silicon chipschips.

DishBrain 'training' process

To create DishBrain, the team at Cortical Labs cultured approximately 1 million brain cells - the equivalent of a cockroach's brain. They then transferred the group of cells into a microelectrode array made up of more than 26,000 sensors.

When connected to a computer, the sensors transmit electrical signals to neurons to simulate the game Pong - in this game, the player controls the support rod to hit the ball against the wall. The team will send an electrical signal to the array of microelectrodes to indicate the position of the ball. The "brain" then activates the neurons to move the rod into the correct position.

Cortical Labs has put its DishBrain chip through extensive testing.

DishBrain will make its own decisions and transmit this information to the computer through sensors. The software then gives feedback on DishBrain's actions. "These actions are repeated, forming a cognitive cycle," Kagan describes.

The team performed more than 4,000 test sessions. The results show that DishBrain only takes 5 minutes to learn how to play Pong, much faster than artificial intelligence. The more neurons played, the better they got, with performance improving by about 67% over time.

The potential of this technology

Cortical's chip will be a new breakthrough in creating smarter and more efficient artificial intelligence systems. In addition, this technology could also help scientists better understand the mysteries of the brain and develop drugs to treat neurological diseases.

The technology of applying neurons to chips is increasingly interested, because it helps create "a much more efficient way of doing computation", said Karl Friston, a neurophysiologist at the University of California, San Francisco. University College London in the UK said.

Cortical Labs hopes that using human brain cells will make high-power chips more energy-efficient.

The application of live neurons will be very useful in situations where fast processing and computation are required, for example in self-driving cars. In addition, the potential of this technology for research work is also great.

According to Mark Kotter, a neurosurgeon at the University of Cambridge, DishBrain is the key that opens the door to research into neurological or psychiatric diseases. Cultured brain cells can provide important insights into Parkinson's, Alzheimer's and other diseases far more effectively than animal studies.

Mr. Kotter continued: 'I am very excited about DishBrain because this technology will help build a strong foundation for research into errors in information processing. However, he offered a word of warning. "I think the hardest thing is turning a concept into an actual scalable product. It's an issue that shouldn't be taken lightly."

Difficulties

Other experts question the feasibility of using live neurons to power the chip.

Andre van Schaik, a neuroengineer at Western Sydney University, said: 'This technology has a huge disadvantage that is that the habitat of cultured brain cells needs to be carefully maintained, including avoiding bacterial growth, chemical imbalances and temperature changes. This condition requires machines and equipment that are much larger than chips and consume more energy."

Artificial neurons, on the other hand, require no such maintenance. This explains why scientists in this field often pursue that path.

DishBrain learns to play the game Pong.

IBM's TrueNorth chips contain 1 million digital neurons with 256 million connections. Intel is also developing its own chip, called Loihi. This chip can quickly detect sensations such as gestures, sounds or even smells.

Dr. Takuya Isomura, a researcher at the Riken Institute in Japan, also questioned how complex such chips really are. "In my opinion, the complexity of the tasks the chip can perform is quite limited," he said. That's partly because neurons have their own complex connections that are hard to replicate outside of the brain.

Another major drawback of the DishBrain prototype is that its neurons are only aware of two-dimensional space. "One of the advantages of biological brains is the ability to build networks in 3D," says Mr. Van Schaik. "But putting biological neurons on a 2D chip seems to eliminate that advantage."

Cortical Labs acknowledges that fact and is working on developing neurons capable of 3D recognition.

In addition, current AI technologies require large amounts of energy. A 2019 study estimated that training an existing language processing system would generate about 635 kg of carbon emissions – the equivalent of a round-trip flight between Tokyo and Ho Chi Minh City.