Cortical Labs has demonstrated that around 200,000 living human neurons grown on microchips can play the notoriously complex video game Doom, which was thought to require advanced artificial intelligence. The company’s CL1 device and accompanying API allowed independent researchers to convert the game’s visual information into electrical stimulation of neuronal cultures and interpret the neuron’s responses as in-game commands. “We have solved the interface problem,” explained Dr. Aron Lerffler of the Cortical Institute, indicating that a functional link between the digital and biological worlds has been established. Although neuronal “players” are currently performing at a novice level and exhibiting trial-and-error learning, this demonstration proves the possibility of training and shaping the behavior of living brain cells to accomplish increasingly complex tasks.
CL1 system: 200,000 neurons on a multi-electrode array
Cortical Labs CL1 houses approximately 200,000 live human neurons cultured on a multi-electrode array and represents an important step toward functional biocomputing. Launched last year, the system’s capabilities were recently demonstrated through an unexpected application: playing the video game Doom. Cortical Labs has prioritized user accessibility and developed an API that aims to be as friendly, effective, and fun as possible, according to a company statement. This focus prompted a collaboration with independent researcher Sean, who was able to code a working version of Doom running on CL1 using the API within a week. This achievement builds on previous work in which the same system demonstrated the ability to play Pong, a milestone that proved that adaptive real-time goal-directed learning is achievable.
Doom’s complexity posed considerable challenges. “Doom is chaotic. It’s 3D, there are enemies, you have to explore the environment, and it’s difficult,” explained Dr. Aron Lerffler. The research team overcame this problem by converting the game’s digital information into electrical signals that neurons could interpret, stimulating the neuroculture’s sensory areas based on in-game events. The neuron’s firing patterns act as motor commands in-game, allowing the “doom guy” to move and shoot. Currently, Neuron’s performance is similar to that of a novice player, but the system clearly learns, seeks out enemies, and improves with feedback. “We have solved the interface problem,” Lerfler claims, and established how these cells interact and behave in real time. Cortical Labs is now issuing an open invitation to the research community to explore the platform, focusing on improving learning algorithms and encoding methods, and asking what to teach neurons next.
Cortical Labs API translates fate into neural stimulation
The development of functional brain-computer interfaces has progressed beyond simple tasks. Cortical Labs is currently demonstrating the ability of living neurons to engage with complex digital environments. The company’s CL1 device houses approximately 200,000 human neurons cultured on a multielectrode array and forms the basis of a new neurocomputing system. This achievement follows previous work in which the system was successfully used to play Pong, but came with significant challenges. Doom is a notoriously demanding first-person shooter and served as a test case for CL1’s capabilities. “Doom is chaos.” To overcome this, Cortical Labs has developed an API that allows users to convert digital information into electrical stimulation patterns that neuroculture can understand. Independent researcher Sean leveraged this API to map a game’s video feed to specific electrode stimuli, effectively creating sensory input to neurons. Sean implemented this method within a week using the Cortical Labs cloud platform. Although Neuron’s “players” are far from esports champions and play like beginners who have never seen a computer, the system does learn, explore enemies, and perform basic actions.
Real-time learning demonstrates the flexibility of biocomputation
Cortical Labs is pushing the boundaries of neural computing with the CL1 system, a device that houses approximately 200,000 live human neurons on a multi-electrode array. Dr. Aron Lelfler explained that the company’s recent focus has been on improving user accessibility, culminating in the development of the Cortical Labs API for effective and enjoyable interactions. The transition from Pong to Doom came with major challenges. “Doom is chaos.” To overcome this, the team focused on translating the game’s digital world into electrical signals, the biological language of neurons. The CL API allows users to interact with the neuroculture using Python commands to map the game’s video feed to patterns of electrical stimulation. For example, “When a demon appears on the left side of the screen, specific electrodes stimulate the sensory areas of the neural culture on the left side.” The resulting system allows neurons to respond to stimuli and issue commands in-game by interpreting their responses.
