Science has built a better mouse, controlled by an artificial intelligence (AI) brain.
Developed by researchers at Harvard and Google's DeepMind AI lab, this virtual mouse can accurately mimic the movements of real mice, a significant achievement that promises to advance our understanding of how the brain controls complex and coordinated movements.
It's a feat that even the most sophisticated robots today struggle to replicate, and the team believes their discovery will lead to vastly improved efficiency in future robots.
Professor Bens Olvechki of Harvard's Department of Organismic and Evolutionary Biology led the effort and used high-resolution data from real mice to train an artificial neural network. His lab is dedicated to mechanistic exploration of how the brain moves limbs.
“We found that neural activity … is better predicted by virtual rat network activity than by any real rat activity features, consistent with the application of inverse activations from both regions,” the researchers wrote.
According to the study published in the journal Nature, the virtual mouse simulation was created using MuJoCo's physics simulator, which incorporates realistic forces such as gravity to mimic real-world conditions. The artificial neural network that drives the movements of the virtual mouse was trained on inverse dynamic models, allowing it to predict neural activity in real mice with high accuracy.
The report notes that the results will help scientists interpret neural activity across different behaviors and link it to motor control principles. The principles obtained by studying virtual mice have implications for creating improved robotic control systems.
“These results show how biomechanically realistic virtual animal physical simulation can interpret neural activity underlying behavior and relate it to theoretical principles of motor control,” the study reads.
Image: Google Deepmind
“We've learned a lot from the challenge of building embedded agents,” Matthew Botvinnik from Google's DeepMind told the Harvard Gazette. . “It seems plausible that taking this same approach in a neuroscience context could be useful in providing insights into both behavior and brain function.”
The team used cutting-edge techniques such as AI, deep reinforcement learning, and 3D motion tracking to enable the virtual mouse to replicate a wide range of natural behaviors, including those that were not explicitly trained. The paper says the approach could form a new field of “virtual neuroscience,” providing an accessible platform for studying the neural bases of natural behavior to better understand how the brain controls movement.
This area of research is critical to the development of advanced prosthetics and brain-machine interfaces such as Neuralink or the technology being developed by Precision Neuroscience.
Insights from this work could lead to new treatments for movement disorders by creating neural circuits. In addition, the study found, the virtual mouse provides a clear model for studying neural circuits and the effects of diseases on these circuits.
Moving forward, the researchers plan for virtual mouse autonomy by increasing awareness of the skills of brain algorithms to solve tasks faced by real mice.
Olvecki was unable to comment further to Decrypt.
Edited by Ryan Ozawa.
Generally intelligent newspaper
A weekly AI journey narrated by a generative AI model.
