Researchers train lab-grown brain cells to play Pong

The startup Cortical Labs unveiled the DishBrain system of synthetic biological intelligence, capable in real time of leveraging the adaptive computations inherent to neurons to perform target tasks such as playing Pong.

Researchers employed embryonic mouse material and induced pluripotent stem cells of humans. They integrated biological neural networks (BNN) into a silicon chip designed to read and write information via electrophysiological stimulation. Then the engineers placed DishBrain in a simulated environment — Pong.

“We chose this arcade for its simplicity and fame, and also because it is the first game used in machine learning,” said Brett Kagan, Chief Scientific Officer of Cortical Labs.

The team developed software capable of reading the electrical signals from the cells as they fired and applying them to the game, namely to control the paddle that keeps the ball in play. The program also provides feedback and sends the BNN information about changes in the simulated environment.

As a result, neurons in DishBrain were able to adjust their activity in real time during five minutes of play in the real world and improve performance, demonstrating learning. For comparison, systems that were stimulated but without feedback did not show improvement.

The developers stated that this deliberate change in behavior over time demonstrates the ability of cells even in a Petri dish to exhibit a kind of rudimentary intelligence under the right conditions.

According to Cortical Labs, the technology could eventually be used to test treatments for neurodegenerative diseases, including dementia and epilepsy.

“We needed a system to test the possibility of interacting with neurons to demonstrate intelligence and purposeful activity. Now we need to study these achievements and continue developing the technology to launch new applications,” said Kagan.

The team has already begun testing the system in other game environments. In the future they hope to make the technology more accessible and reliable.

Earlier in September, DeepMind developed AI agents capable of playing virtual football.

In July, OpenAI engineers trained an algorithm to mine resources, craft diamond tools, hunt animals and cook food in Minecraft.

In June 2021, Google Research researchers created AI that works like the human brain.

Subscribe to ForkLog news on Telegram: ForkLog AI — all the news from the world of AI!