In a groundbreaking collaboration, researchers from Harvard University and DeepMind, a Google subsidiary specializing in artificial intelligence (AI), have created a virtual rat with an artificial brain that can control complex movements. The goal of this project is to gain insight into how the brain manages movement.
Utilizing real data from rats recorded at high resolution, the researchers trained an artificial neural network to control the virtual body within a physics simulator called MuJoCo. This simulator accurately replicates the environmental forces that influence movement in real-life scenarios, such as gravity. Through their experiments, the researchers observed that the artificial brain in the virtual rat triggered the same neural control networks as those found in real rat brains when facing movement challenges.
According to Harvard neuroscientist Bence Ölveczky, who specializes in training real rodents to perform complex behaviors for the study of neural circuits, working closely with DeepMind researchers was invaluable. The team trained an artificial neural network to develop inverse dynamics models that mimic how the brain guides movement, such as reaching for a cup of coffee.
The next phase of their research involves providing the virtual rat with autonomy to perform tasks similar to those done by real rats. This advancement in virtual neuroscience opens up exciting possibilities for using AI-simulated animals to study neural circuits and disease mechanisms. Additionally, the platform created with DeepMind could have applications in designing better robotic control systems.
Overall, this collaboration between Harvard University and DeepMind’s AI systems holds promise for a wide range of applications beyond brain research, including identifying disease-causing DNA mutations and optimizing the design of materials for various technologies.
