‘Pong’ learned faster by brain cells in a dish than AI

This is according to an article from New Scientist (via IGN) and the team’s work at Cortical Labs. A network of neurons in a dish, or a “mini-brain”, are placed into a virtual game world environment and given a goal-oriented task. “We think it’s fair to call them cyborg brains,” said Brett Kagan, chief scientific officer at Cortical Labs, adding that this is the first time these neuron networks have been found to perform such a task.

You can see the software and brain cells in action in the above video, which explains that “pulses of electrical activity from micro-electrode interfaces simulate the game world. In these virtual worlds, each cyborg mini-brain acts as if it is the paddle that hits the ball.

“While the mini-brains are not as good at Pong as computer-based AIs,” the video adds, “they do learn faster”. It would take an AI around 90 minutes to learn how to play Pong, but these mini-brains take just five.

This work could lead to the creation of synthetic brains, and Cortical Labs wants studies like this one to push for live neurons to be used in traditional computing, which could help computers “solve problems in unfamiliar situations”.

If you thought that was delightfully strange, scientists have been teaching rats to play Doom to help research into prosthetics or human-computer interfaces, as the rats use custom VR rigs to guide the player through a level.

In other news, supply for the Nintendo Switch is likely to struggle in the early part of 2022, thanks to demand and shortages.