Scientists at Monash University created the “DishBrain” a semi-biological computer chip with about 800,000 human and mouse brain cells grown in the laboratory on their electrodes. In five minutes he learned to play Pong.
The microelectrode array in the heart of DishBrain is able to read the activity of brain cells and stimulate them with electrical signals, so the research team created a version of Pong in which the brain cells received a moving electrical stimulus that represented which side of the “screen” the ball was on and how far away from the racket. They allowed brain cells to act on the paddle, moving it left and right.
Then they created a very basic reward system, using the fact that small groups of brain cells tend to try to minimize the unpredictability of their environment. If the racket hit the ball, the cells received a pleasant and predictable stimulus. But if it failed, the cells would receive four seconds of totally unpredictable stimulation.
This is the first time that lab-grown brain cells have been used in this way. giving them not just a way to perceive the world but to act on it, and the results were impressive.
So impressive that the research, carried out in collaboration with Cortical Labs, a start-up in Melbourne, Australia, received a grant of US$407,000 from the Australian National Intelligence and Security Discovery Research Grants program.
These programmable chips, which fuse biological computing with artificial intelligence, “could outperform existing pure silicon hardware in the future”says project manager, associate professor Adeel Razi.
In other words, DishBrain’s advanced learning capabilities could underpin a new generation of machine learning, especially when incorporated into autonomous vehicles, drones and robots. I could give them, says Razi, “a new kind of artificial intelligence capable of learning throughout its life.”
The technology promises machines that can keep learning new skills without compromising old ones, that can adapt well to change and map old knowledge to new situations, continually optimizing the use of computing power, memory and energy.