The "Can it run Doom?" challenge just crossed its next frontier.
Melbourne, Australia-based startup Cortical Labs has successfully trained clusters of living human brain cells to navigate the 1993 classic video game.
“In 2021, we got Pong working on the neurons. That was sort of like a first kind of test to see if we can have some kind of cool game that speaks to the people. But then the number one response we got is, can it run Doom?” Alon Loeffler, an application scientist at Cortical Labs, told Decrypt.
Inside the company’s CL1 device are roughly 200,000 living human neurons cultured on a multi-electrode array, allowing researchers to stimulate the cells electrically and interpret their responses in real time.
Can it run Doom?
For decades, Doom has served as an unofficial benchmark for engineers testing new systems.
Since Texas-based video game developer id Software released the game’s source code publicly in 1997, developers have ported it to a wide range of unexpected platforms.
Loeffler said the team initially relied on improvised low-level computing code to make the systems functional, but eventually decided to build a platform designed from the ground up to allow researchers to interact with neurons through high-level controls using simple Python commands.
Once Coritcal Labs built the platform, development accelerated.
“It took the collaborator, Sean, who wrote Doom, the code for us a few days instead of 18 months,” Loeffler said.
Teaching Neurons to Play
The neurons learn through feedback signals, receiving small rewards when they correctly aim at an enemy and larger rewards when they successfully shoot and eliminate a target, reinforcing the behaviors associated with those signals over time.
Cortical Labs researchers then used artificial intelligence to refine how game information was encoded into electrical signals sent to the neurons.
“The cells are actually learning the input,” Loeffler said. “But then the AI is trying to improve that input to try and get the cells to do what we want them to do.”
While the cells showed steady improvement while playing Doom, Loeffler emphasized that the neurons reacted to inputs rather than truly understanding the game.
“The system doesn’t actually know it’s playing Doom,” he said. “It’s getting electrical signals and then spitting out responses.”
Loeffler said working with living neurons requires a different approach than traditional programming.
“It’s a completely different shift in mentality,” Loeffler said. “You can’t just have this normal kind of computing system that you’re programming. It needs to be done in a whole new attitude and a whole new way of looking at things.”
Gaming, he said, serves as a public-facing demonstration while researchers explore practical uses.
Despite using human-derived neurons, Loeffler said the system does not resemble human cognition.
“Just because they’re human cells doesn’t mean it’s a human on that dish,” he said. “There are no pain receptors. There are no structures that could allow for higher-order functionality.”
Despite this, researchers, he said, see evidence of neural adaptability outside the brain.
“We still see adaptability to the environment. We still see learning,” Loeffler said. “Which shows you the kind of inherent abilities of neurons to adapt.”
