Sentient neurons learn to play a video game
The DishBrain chip was tested to determine its abilities, including self-learning.
Neurons cultured in petri dish are able to learn to play Pong, a classic video game, exhibiting “intelligent and sentient behavior”, according to Australian neuroscientists.< /p>
This study, published in the journal Neuron(in English), opens the way to a new type of research that could one day use neurons to process information, a kind of biological machines that would support digital computers, according to the Australian researcher who led research, Brett Kagan, of Cortical Labs.
Machines are incapable of learning things very quickly. For a machine learning algorithm [an artificial intelligence technique] to learn something, it needs thousands of data samples, he explained.
While a dog can learn a trick in two or three tries, he notes.
Neurons are the foundation of intelligence in all animals, from insects to humans.
For their experiment, aimed at discovering whether it was possible to harness the inherent intelligence of neurons, Brett Kagan and his colleagues took neurons from embryonic mouse brains as well as neurons from adult human stem cells. /p>
They then cultured these neurons around arrays of microelectrodes capable of detecting their activity and stimulating them. The experiments involved clusters of about 800,000 neurons, the size of a bumblebee's brain.
In a sort of simplified version of the tennis game Pong, a signal was then sent from the right or left to indicate the location of a ball and the cluster of neurons, dubbed by the researchers DishBrain (or brain in a box, in French), responded with another signal to move the racket.
A signal was sent from the right or left to indicate the location of a ball and the cluster of neurons responded with another signal to move the racket.
One of the main obstacles was figuring out how to get neurons to learn a task.
Previous work had suggested giving them a dose of the happiness hormone, dopamine, with each correct action , but this proved difficult to accomplish with the necessary speed.
Instead, Dr. Kagan's team relied on the so-called free energy principle theory introduced more than 10 years ago by Karl Friston, the lead author of the study published Wednesday, and according to which the cells instinctively seek to minimize the unpredictability of their environment.
When the neurons managed to hit the ball using the racket, they therefore received predictable information indicating their success. But when they missed, the received signal was random, unpredictable.
The only way for neurons to keep their world controllable and predictable was to be more successful at hitting the ball, detailed Brett Kagan.
His team considers DishBrain to be sentient – what they define as being able to perceive sensory information and respond to it dynamically – but don't go so far as to talk about consciousness, which implies being aware of one's own existence.
This canned brain even tried his hand at a game that replaces Google's search engine when there's no internet connection, which involves running a dinosaur through obstacles, and early results have been encouraging, Brett said. Kagan.
Scientists next want to test how drugs and alcohol affect the intelligence of these neurons, but Dr. Kagan is mostly excited about the possibility. to develop biological computers.
It's a rigorous and interesting neuroscience [experiment], says Tara Spires-Jones of the University of Edinburgh, who hasn't participated in the study.
Don't worry, even if these boxes of neurons can modify their responses when they are stimulated, they are not gifted boxes of intelligence of science fiction type, these are simple (although interesting and important from a scientific point of view) answers from circuit, she reassures.