Living cells have been used to create a biological computer that can control the movements of a robotic dog. Its creators say using cells to replicate the digital neural networks of AI models like ChatGPT – which are themselves modelled on biological brains – could create more efficient and adaptable computers.

The start-up Cortical Labs has created and commercialised a biological computer called CL1, which can be trained to navigate towards a target in a simulated 2D environment. In later tests, this training data was used to control the movements of a real robotic dog.

Ferrante Neri at the Education University of Hong Kong, who wasn’t involved in the research, says that a paper published by the company that is yet to be peer-reviewed shows experiments where the neural network improved as time passed, even as it was fed the same digital data – suggesting it was being successfully trained to do the task.

“That is promising evidence of learning from limited experience under these experimental conditions,” says Neri. “Establishing an advantage on this task is an early step towards handling the complexity of real-world navigation.”

Cortical Labs has already used living cells to perform computations: in 2021, it showed that its technology could play the computer game .

But the company’s latest experiments show there may also be applications in robotics in the physical world. Cortical Labs says that a biological computer could be less energy-intensive to run than a digital alternative, which would be an advantage in robotics where battery life is a key limitation.

Brett Kagan at Cortical Labs says the cells used in CL1 computers are created from stem cells that are generated from blood donated by people who have given their consent. “There is no evidence that these systems experience pain or suffering, but we recognise that this possibility requires careful monitoring and are addressing it with independent groups.”

“Despite the best advances in [digital] robotics, having them avoid running into obstacles remains a challenge. And that’s because there is still work to be done in areas such as higher order control, navigation and pathfinding,” says Kagan. “Biology and AI have different strengths; biology has evolved to adapt and respond to the world in real time.”