Google DeepMind, Harvard Develop AI-Powered Virtual Rat to Study Movement

Researchers from Google DeepMind and Harvard University have created a virtual rat powered by an AI "brain" that can direct the rodent's motions in strikingly similar ways to a real rat's brain.

In a study published in the journal Nature, the researchers showed that the patterns of neural activity in the AI system produced natural, coordinated movements that matched those observed when recording from an actual rat's brain cells during motion.

According to the researchers, these findings mark a significant step toward understanding the brain's mechanisms for controlling movement.

“These results demonstrate how physical simulation of biomechanically realistic virtual animals can help interpret the structure of neural activity across behavior and relate it to theoretical principles of motor control,” the findings read.

Understanding exactly how the brain works has baffled scientists for years. Seen as the ultimate computer, mimicking how the brain operates is as elusive a dream to some researchers as artificial general intelligence.

Google DeepMind has been working on this concept for some time, coining the term NeuroAI.

In its latest efforts with Harvard, Deepmind applied deep reinforcement learning to a virtual rodent. The rat was designed to be a biomechanically realistic representation of a real rat and was instructed to copy the whole-body movements of how a rat might move in a virtual environment called MuJoco.

Related:NeuroAI: Tapping the Human Brain to Fuel AI Development

The virtual rat was built using movement data recorded from real rats | Credit: Google DeepMind

Researchers then compared neural activity from a real rat’s brain with the virtual one and found it performed similar neural behaviors during movement.

“We found that neural activity in the sensorimotor striatum and motor cortex was better predicted by the virtual rodent’s network activity than by any features of the real rat’s movements, consistent with both regions implementing inverse dynamics,” the study read.

Harvard professor Bence Ölveczky stated that his team did not have the resources to run simulations and train neural networks like the ones found in the virtual rat without DeepMind’s help.

“From our experiments, we have a lot of ideas about how such tasks are solved, and how the learning algorithms that underlie the acquisition of skilled behaviors are implemented,” Ölveczky said. “We want to start using the virtual rats to test these ideas and help advance our understanding of how real brains generate complex behavior.”

Diego Aldarondo, a research scientist who worked on the project, said the techniques used in the virtual rat tests could be used to model the neural control of more complex animal behaviors.

Related:Google DeepMind's Six Levels of AGI

Matthew Botvinick, Google DeepMind’s senior director of research,  said the test provided insights into building embodied agents or AI systems that can translate intelligent thinking into physical actions.

“It seemed plausible that taking this same approach in a neuroscience context might be useful for providing insights in both behavior and brain function,” Botvinick said.