Scientists develop ‘cyborg worm’ with AI-driven brain

Scientists develop ‘cyborg worm’ with AI-driven brain

Matthew Hutson

Scientists have demonstrated an intriguing brain-AI collaboration by letting artificial intelligence directly manipulate the nervous system of a millimeter-long worm to guide the creature to a tasty target. They trained the AI ​​with a technique called deep reinforcement learning, the same one used to help AI players master games like Go. Artificial neural networks are software modeled after biological brains that analyze sequences of actions and their outcomes to extract strategies for the AI ​​”agent” to interact with its environment and achieve its goals.

In a study published in 2011, Nature Machine IntelligenceThe researchers found that the 1-millimeter-long Nematodes Earthworms heading for tasty spots E. coli The 4-cm dish was fitted with a camera that recorded the position and orientation of the worm’s head and body. The agent received 15 frames of information three times per second, giving it an understanding of what was and is happening at that moment. The agent could also turn on and off a light pointed at the dish. The worms were optogenetically engineered so that certain neurons could be activated or deactivated in response to light, inducing movement.

The team tested six genetic strains in which the worms had a range of light-sensitive neurons, from 1 to 302. Each strain had a different effect, for example causing the worms to change direction or preventing them from changing direction. The scientists first randomly exposed the worms to light for five hours to collect training data, fed that data into an AI agent to find patterns, and then let the agent move freely.

Supporting science journalism

If you enjoyed this article, please support our award-winning journalism. Subscribe. By purchasing a subscription, you help ensure a future of influential stories about the discoveries and ideas shaping the world today.

In five of the six lines, including one in which all neurons responded to light, the agent learned to guide the worm to the target faster than if the worm was left alone or the light was flashing randomly. What’s more, the agent and the worm cooperated: even if the agent guided the worm straight to the target, if there was a small obstacle in its path, the worm would avoid it.

T. Tang Voddoin, an engineer at the University of Queensland in Australia who is working independently on cyborg insects, praised the study’s simple setup. Reinforcement learning is flexible, and AI based on it can figure out how to perform complex tasks. “It’s easy to imagine how it could be extended to more difficult problems,” said Harvard biophysicist Chengguan Li, lead author of the paper. Her team is now investigating whether the technique could improve electrical deep brain stimulation for the treatment of Parkinson’s disease in humans by adjusting the voltage and timing. One day, reinforcement learning and implants may give us new skills, Li said, or a fusion of artificial and real neural nets.