Scientists teach lab neurons to play ‘Pong’

The cultured neurons came from mouse brains and human stem cells. They were also taught to play Google’s ‘Dinosaur’.

Brain cells grown in a lab have learned to play the classic video game Pong and may be capable of “intelligent and conscious behavior,” Australian neuroscientists have concluded.

Brett Kagan, who led the study published Wednesday in the journal Neuron , said their findings open the door to a new kind of research on biological information processors, complementing normal digital computers.

“What machines can’t do is learn things very quickly,” he said, explaining that if you need a machine learning algorithm to learn something, that requires thousands of data samples.

But a dog, for example, “can learn a trick in two or three tries,” said Kagan, scientific director of Melbourne-based Cortical Labs.

Neurons are the basis of intelligence in all animals, from insects to humans.

For their experiment, which sought to discover whether it was possible to exploit the inherent intelligence of neurons , Kagan and his colleagues took neurons from the embryonic brains of mice, as well as neurons from adult human stem cells.

They then grew these neurons around microelectrode arrays capable of detecting their activity and stimulating them. The experiments involved a group of around 800,000 neurons , roughly the size of a bumblebee brain.

In a kind of simplified version of the video game Pong , a signal was sent from the right or left to indicate the location of a ball and the group of neurons , dubbed by the researchers ‘ DishBrain ‘ (brain in a box), responded with another signal. to move the racket.

“Principle of free energy”

One of the main obstacles was how to make the neurons learn a task.

Previous work had suggested giving them a dose of dopamine, the “happiness hormone”, with each correct action, but this has proven difficult to achieve with the necessary speed.

Instead, Kagan’s team relied on the so-called “free energy principle” theory put forth more than 10 years ago by Karl Friston, lead author of the study published Wednesday, that cells are programmed to minimize unpredictability of your environment.

When the neurons succeeded in hitting the ball with the racket, they received predictable electrical signals indicating their success. But when they failed, the received signal was random and unpredictable.

“The only thing the neurons could do was enhance their attempt to hit the ball to keep their world controllable and predictable,” Kagan said.

His team considers ‘DishBrain’ to be “sentient,” which they define as the ability to perceive sensory input and respond to it dynamically, but they don’t go so far as to call it “conscious,” which implies awareness of one’s own existence.

The ‘DishBrain’ was also tested with another task: the game that appears in Google Chrome when an internet connection is not found, which consists of making a dinosaur run around some obstacles. And the preliminary results were encouraging, Kagan said.

Scientists now want to know how drugs and alcohol affect the intelligence of these neurons , but Kagan is more excited about the possibility of developing biological computers.

“This is interesting and rigorous neuroscience,” said Tara Spiers-Jones of the University of Edinburgh, who was not involved in the study.

“Don’t worry, although these boxes of neurons can modify their responses when stimulated, they are not science fiction-like boxes of intelligence, they are simple (yet interesting and important) circuit responses from a scientific point of view,” he said. ( AFP )