Researchers From Cortical Labs Develop DishBrain: A Neural Network With Biological Neurons

Over the last decade, neural networks have become a trendy topic, ranging from image recognition to text generation and even video gameplay applications. On the other hand, these artificial neural networks are just mounds of math inside a computer. While they are capable of tremendous things, the technology has yet to demonstrate the ability to develop actual intelligence.

Researchers at Cortical Labs, Australia propose that integrating neurons into digital systems to tap on their inherent intelligence could enable performance that would be impossible to achieve with silicon alone and provide insight into the biological origins of intelligence.

The overall goal of the research is to use biological neurons’ processing capability to build “synthetic biological intelligence.” The general premise is that real neurons are significantly more complicated and capable than software-based neural networks. Thus, rather than fiddling around with human-created intelligence models, it makes more sense to use biological neurons to develop a viable intelligence from scratch.

The researchers looked at neural networks grown from mouse and human cells. That is how DishBrain was born, a system that demonstrates natural intelligence by utilizing the neurons’ intrinsic adaptive computation in a structured context. DishBrain was pushed to the limits in a synthetic game setting similar to Pong. A sequence of electrodes in the biological neural network (BNN) was triggered based on the game state, delivering sensory input to the cells. Other electrodes then controlled the up and down motion of the paddle in the game.

The neural network was then taught to operate the game intelligently using several feedback techniques. The fundamental concept was founded on the Free Energy Principle, which states that biological systems should act to maintain a world condition that corresponds to their internal models. As a result, the “Stimulus” condition feedback loop was created to provide unpredictable random input when the paddle missed the ball and predictable feedback when the paddle hit the ball perfectly. This method was then compared to a quiet mode, in which the paddle’s stimulus was entirely shut off when it hit the ball, and a no-feedback manner, in which no extra stimulation was delivered relative to the game state.

https://www.biorxiv.org/content/10.1101/2021.12.02.471005v2.full

If the findings are correct, the researchers were able to develop a whole brain in a vat and train it to control a video game. This finding adds to the growing data that neuronal cells may be maintained and interacted with successfully. Furthermore, it gives a foundation for more excellent knowledge of how our brains function, both conceptually and physically.

This study demonstrates that organic neurons can be taught to accomplish tasks intelligently in conjunction with computer interfaces. Humanity is only now learning how to communicate with genuine biological brains, and we may master that skill before attempting to build our own from scratch!

Paper: https://www.biorxiv.org/content/biorxiv/early/2021/12/03/2021.12.02.471005.1.full.pdf

Reference: https://hackaday.com/2022/03/01/researchers-build-neural-networks-with-actual-neurons/