Scientists from Harvard University and Google's DeepMind AI Lab have collaborated to create a virtual AI mouse, which is not only a technological breakthrough but may also open up a new field called "Virtual Neuroscience".
The significance of this research lies in its potential to not only help us understand how the brain controls complex body movements but also to have a profound impact on neuroscience and robotics.
Virtual Rat: AI Brain for Agile Movements
Imitating the Evolutionary Miracle: The flexibility in human and animal movement is the result of long-term evolution. Scientists have been trying to crack the secret of brain control over movement by studying animals, particularly rats with intelligence equivalent to an 8-year-old child and exquisite body control abilities.
No Sacrifice Needed: Unlike previous studies, this research did not sacrifice any real lab rats. Scientists created a virtual AI mouse that can mimic all the actions of real rodents, even displaying some novel behaviors that were not explicitly trained.
New Field of Virtual Neuroscience Born
Nature Publishes Research: This pioneering research was published in the journal "Nature," demonstrating that the activation states in a virtual control network can accurately predict neural activity in the real rat brain.
Artificial Neural Network Training: The research team used high-resolution data recorded from real rats to train an artificial neural network, serving as the "brain" of the virtual mouse, controlling its body in the MuJoCo physical simulator.
Challenges and Learning in Embodied Agents
Embodied Agents: Matthew Botvinick from Google DeepMind stated that the team has learned a lot from building embodied agents, which need to translate thinking into action in complex environments.
Inverse Dynamics Model: Graduate student Diego Aldarondo, in collaboration with DeepMind researchers, trained an artificial neural network to achieve an inverse dynamics model, similar to the way the brain guides movement.
Profound Impact on Neuroscience and Robotics
Virtual Neuroscience: Researchers believe these simulations can open up a new field of "Virtual Neuroscience," providing convenient and transparent models for studying neural circuits.
Designing Improved Robotics Control Systems: This platform may be used to design improved robotics control systems, making robot movements more fluid.
Comparison between Artificial and Biological Neural Networks
Neuroscience Wind Tunnel: This platform allows researchers to test the biological realism of different neural networks and understand their ability to deal with complex challenges.
Exploring the Neural Basis of Behavior: This approach may be a very effective method for exploring the neural basis of behavior.
The collaboration between Harvard University and Google's DeepMind AI Lab has provided the necessary resources and opportunities for training networks for the virtual rat research. The collaboration aims to advance our understanding of how the real brain produces complex behaviors.
This research not only deepens our understanding of the cross-disciplinary collaboration between AI and neuroscience but also provides a new perspective for observing and understanding how the brain works. As virtual neuroscience develops, we may be able to design more agile and intelligent robots that can act freely in complex environments.
Link to Paper: https://www.nature.com/articles/s41586-024-07633-4
