Skip to content

Harvard and Google create the first virtual rat to decipher how the brain controls movement

Researchers from Harvard University and DeepMind, the research company artificial intelligence (AI) from Google, have created a virtual rat, with an artificial brain capable of controlling complex and coordinated movements, with the aim of helping to decipher the mechanisms by which the brain controls movement.

Through this realistic digital model of the rat, described this Tuesday in Nature magazine, science takes a step further by developing an AI system that not only thinks intelligently, but also translates that thought into physical actions, in this case in the movement of the limbs.

LOOK: Former OpenAI employee warns about the dangers of AI smarter than humans

The researchers used real rat data recorded in high resolution to train an artificial neural network (the brain of the virtual rat) in order to be able to control the virtual body in a physics simulator called MuJoco.

In this simulator, gravity and other forces comparable to those of the normal environment where the movement occurs are present.

In their experiments, the researchers found that “the artificial brain activated the same neural control networks when faced with movement as the real brains of real rats”emphasizes Harvard neuroscientist Bence Ölveczky, an expert in training real rodents to learn complex behaviors in order to study their neural circuits.

How a cup of coffee moves the brain

“The collaboration has been great: DeepMind had developed a system to train biomechanical agents to move in complex environments that has been very good for us, since we did not have the resources to run simulations like those,” says Ölveczk in a statement.

The Harvard researcher’s team worked closely with Google DeepMind researchers to train an artificial neural network to develop so-called inverse dynamics models, which scientists believe the brain uses to guide movement.

The authors cite the example of how when picking up a cup of coffee the human brain quickly calculates the trajectory that the arm should follow, and translates that calculation into motor commands to execute the movement.

“Similarly, based on data from real rats, the neural network received a reference trajectory of the desired movement and learned to produce the forces necessary to generate it. “This allowed the virtual rat to imitate a wide range of behaviors, even those for which it had not been explicitly trained,” explains Ölveczk.

Key models to solve diseases

These simulations open an unexplored field of virtual neuroscience in which AI-simulated animals, trained to behave like real ones, can provide comfortable and transparent models to study neural circuits and see how they are compromised in diseases.

LOOK: Nvidia, Microsoft and OpenAI, investigated in the US for possible antitrust practices regarding AI

The researchers advance that the next step will be to provide the virtual animal with autonomy to solve tasks similar to those of real rats.

“We want to start using virtual rats to test these ideas and help advance our understanding of how real brains generate complex behaviors,” Ölveczky continues.

Although their laboratory is focused on studying the functioning of the brain, the platform created with DeepMind could be used, among others, to design better robotic control systems, they emphasize.

Google DeepMind’s artificial intelligence systems are helping to develop key tools to identify, among others, changes in human DNA that can cause diseases; or to discover materials that can be used to make better solar cells, batteries, or computer chips.

Source: Elcomercio

Share this article:
globalhappenings news.jpg
most popular