The Australian startup building computers out of human brains
Artificial intelligence systems are designed to mimic the way humans think. So why not recruit brain cells to help?
Three years after its founding, Cortical Labs went viral in 2022. It had been developing a new type of processor that combines silicon with human neurons, literally brain on a chip. It demonstrated the system by teaching its “DishBrain” processor – brain cells in a sensor-equipped petri dish – to play the video game Pong.
It was the world debut of bio-computing, which uses brain neurons to expand what traditional processors are capable of. Cortical Labs’ founder Hon Weng Chong saw it as a new, more efficient way to create artificial intelligence models. Instead of using huge amounts of data to train AI models, like OpenAI did for GPT-4, biocomputers train neurons to think for themselves by giving them positive or negative feedback.
Because neurons communicate using infinitesimal electric shocks, bypassing traditional transistors’ need for high voltage, the processors’ power consumption is a fraction of existing CPUs and especially the GPUs that are used to train AI models. It’s that last bit that intrigued Amazon chief technology officer Werner Vogel, who visited Cortical Labs’ Melbourne office last year.
"We spent a lot of time proving this soup of neurons can do something, that you can coax them into doing something that is a little bit more useful," Chong said to Capital Brief. "After that [the Pong demo] we did a hard pivot to actually start commercialising the technology."
