Monday, December 27, 2021

Disembodied Human Brain Cells Are Learning Pong Faster Than AI

Researchers have placed networks of neurons into dishes, placed them into what amounts to a simple version of The Matrix, and found they can learn Pong faster than an AI.

As reported by New Scientist, researchers at Cortical Labs explained that they've grown groups of human neurons into organoid mini-brains (a process also being used to attach Neanderthal brain cells to crab robots) and placed on micro-electrode interfaces. Those interfaces pulse with electricity, used to effectively convince the mini-brains that they are the paddles inside a game of single-player Pong.

“We often refer to them as living in the Matrix,” explained chief scientific officer Brett Kagan. Within that virtual game world, the neurons can begin to move those paddles, using them to stop Pong's relentless bouncing ball from passing by them.

What's impressive is how quickly they can learn how to do this. Cortical Labs says that an AI would generally take around 90 minutes to learn what to do in this situation, where the mini-brains seem to learn in just 5. However, an AI would become much, much better at the task after learning it. You can see a game in action on YouTube.

Cortical Labs hopes to use studies like these to combine live neurons into traditional computing, allowing computing to "solve problems in unfamiliar situations", which could be used to solve autonomous robotics problems.

Joe Skrebels is IGN's Executive Editor of News. Follow him on Twitter. Have a tip for us? Want to discuss a possible story? Please send an email to newstips@ign.com.



source https://www.ign.com/articles/pong-human-brain-cells-faster-learn-ai

No comments:

Post a Comment