How many brain cells does it take to play Pong? A neural network system called DishBrain knows. Neurons that learned to play are the proof. It is the first synthetic biological intelligence experiment of its kind.
Neurons can adjust their activity to perform a specific task. “We can interact with living biological neurons. We force them to modify their activity, leading to something that resembles intelligence.” Says neuroscientist Brett Kagan of the biotech company Cortical Labs.
Feedback
DishBrain combines neurons extracted from embryonic mice and human neurons grown from stem cells. A total of 800,000 brain cells are needed to play Pong. In one game of the game, microelectrodes were on both sides of the plate. They indicated whether the ball was to the left or right of the paddle. DishBrain is able to move the paddle to meet the ball. But to play well, the neurons need feedback.
The team developed a computer program to issue critiques through the electrodes whenever DishBrain failed. “That’s the beautiful and pioneering thing about this work. Giving the neurons sensations – the feedback – and the ability to act on their world.”
“We chose Pong because of its simplicity and familiarity. It was one of the first games used in machine learning. An unpredictable stimulus was applied to the cells. And the system as a whole reorganized its activity to best play the game and minimize the random response.”
Replicating brains
This has some intriguing possibilities, especially in artificial intelligence and computer science. Could something like the human brain be replicated? This goal might not be that far off. But there are other implications. For example, DishBrain could help chemists understand the effects of various drugs on the brain, at the cellular level. Even tailor drugs to a patient’s specific biology.
Neurons that learned to play games show us a fascinating path. Further experiments will expand the possibilities for responses in other settings. The study was published in Neuron.
