Researchers from Harvard University, in collaboration with Google DeepMind, have harnessed artificial intelligence to create an artificial "brain" for a virtual mouse, enabling precise control of its movements in complex environments. This groundbreaking innovation has been published in the journal Nature.
The research team utilized real mouse data to construct a biologically accurate 3D model of a mouse. DeepMind's deep reinforcement learning algorithms trained an artificial neural network (ANN) brain for this model, allowing it to generate a variety of complex movement trajectories and forces accurately through an inverse dynamics model.
Source: Google DeepMind
This virtual brain not only mimics trained actions but also autonomously generates new behaviors it has never been trained for, with simulation results that are "beyond real." Project leader Ölveczky noted that DeepMind's technology provides strong support for complex simulations, which was key to the breakthrough achieved in this collaboration.
This research opens up a new avenue for exploring the neural circuits responsible for complex animal behaviors. By analyzing the operation of the artificial intelligence brain, scientists can gain insights into the principles of real biological brains. This innovative simulation method also holds promise for designing advanced robot control systems.
More excitingly, this could herald a new field of "virtual neuroscience." In the future, artificial intelligence simulations of biology might become transparent experimental models for studying normal and diseased brains, offering unprecedented insights and potentially leading to new strategies for treating neurological diseases.
Building on current achievements, the next step for researchers will be to endow the virtual mouse with more autonomy, simulating its process of learning new skills, and further exploring the mysteries of how the brain acquires complex behavioral abilities. With continuous improvements to this method, revolutionary advancements in neuroscience and artificial intelligence research are bound to emerge.