Australian AI scientists found out that when living brain cells in a dish are placed in what researchers call a "virtual game world," they can learn to play the old-school video game Pong, New Scientist reported.
"We think it’s fair to call them cyborg brains," Brett Kagan, chief scientific officer of Cortical Labs, is quoted in the report as saying.
According to the publication, many researchers from all over the world have been researching networks of neurons in dishes, often forming brain-like organoids. But the Cortical Labs findings are the first time that mini-brains have been discovered to undertake goal-directed tasks.
However, hundreds of thousands of human brain cells in a dish can not only learn to play Pong, but also increase their performance faster than artificial intelligence, according to the report.
"We often refer to them as living in the Matrix," Kagan said. "When they are in the game, they believe they are the paddle."
Furthermore, although these brain cells in a dish are not as good at Pong as AI or real people, they do learn faster, the scientists suggest in their study, published at bioRxiv, as earlier it was reported that a separate group of researchers taught a table tennis robot to play the game in 90 minutes.
"The amazing aspect is how quickly it learns, in 5 minutes in real time," said Kagan. "That’s really an amazing thing that biology can do."
The "DishBrains" the company is developing are made up of between 800,000 and 1 million live brain cells, about equivalent to a cockroach brain. Some contain embryonic brain cells from mice, while others contain stem cell-derived human brain cells. Microelectrode arrays that can both stimulate and read the activity of the cells are grown on top of the cells. The firing of electrodes on the left or right of one array tells the mini-brain, or the paddle, whether the ball is on its left or right, simulating a rudimentary version of Pong with no opponent. The closeness of the signals is indicated by their frequency.
The paddle moves left or right according to certain patterns of activity throughout the neurons. The computer responds to this activity, and the mini-brains learn how to operate the paddle thanks to feedback from the electrodes.
Improved machine learning and drug testing in order to observe how experimental medications influence the brain are named as two possible applications for the researchers' findings, according to their paper.