Scientists have recently established a direct connection between artificial intelligence and the nervous systems of tiny worms, enabling the AI to guide these creatures toward a food source.
This breakthrough showcases a fascinating collaboration between brain and AI. The researchers used a technique called deep-reinforcement learning, commonly employed in training AI to master complex games like Go.
This method involves an artificial neural network, which is a software system inspired by the structure of biological brains. The network analyzes sequences of actions and outcomes, allowing the AI “agent” to learn strategies for interacting with its environment to achieve specific goals.
In a study published in Nature Machine Intelligence, researchers trained an AI agent to guide one-millimeter-long Caenorhabditis elegans worms toward patches of Escherichia coli in a four-centimeter dish.
A nearby camera tracked the location and orientation of each worm’s head and body, and the agent received this data three times per second, covering the previous 15 frames to provide a sense of both past and present.
The agent also had control over a light aimed at the dish, which could be turned on or off. The worms were genetically engineered to have certain neurons that would activate or deactivate in response to the light, sometimes inducing movement.
The researchers experimented with six different genetic lines, each varying in the number of neurons sensitive to light—from just one neuron to all 302 neurons in the worms. The light stimulation had different effects depending on the genetic line, such as causing the worm to turn or preventing it from doing so.
Initially, the scientists collected training data by randomly flashing lights at the worms for five hours. This data was then used to train the AI agent to recognize patterns before allowing it to control the light itself.
The AI agent successfully learned to direct the worms toward the target faster in five of the six genetic lines, including the one where all neurons responded to light.
Remarkably, the AI and the worms worked together; if the AI guided a worm straight toward a target but encountered small obstacles, the worm would navigate around them.
T. Thang Vo-Doan, an engineer at the University of Queensland in Australia who has conducted independent research on cyborg insects, commended the simplicity of the setup.
He noted that reinforcement learning is highly adaptable, and AI based on this method can solve complex tasks. Chenguang Li, a biophysicist at Harvard University and the lead author of the study, emphasized the potential of this method to tackle more challenging problems.
Her team is now investigating whether this approach could enhance deep-brain stimulation used to treat Parkinson’s disease in humans by optimizing the voltage and timing of the stimulation.
Li envisions a future where reinforcement learning, combined with neural implants, might even grant us new skills by merging artificial and biological neural networks.
Discover:
