An artificial intelligence system developed by Google DeepMind, in collaboration with researchers, is helping to shed light on how mammals such as humans navigate their environment, according to a report from Science News.  The research, published in the journal Nature earlier this month, used AI to simulate specialized neurons called grid cells, which are crucial in helping mammals orient themselves in a space.

In the study, the virtual grid cells helped the AI plan more efficient routes through mazes, suggesting that the grid cells are also important to navigation and route planning.

In 2014, Norwegian researchers first discovered that rats use these grid cells to track their location in a space, by projecting an imaginary hexagonal framework onto their environment. The research won the team the 2014 Nobel Prize in physiology or medicine, and researchers theorized that the cells may also help mammals plan routes between locations in addition to helping them gain their bearings in a space.

Now, along with Google’s AI research firm DeepMind, University College London neuroscientist Caswell Barry has investigated that idea by training AI-simulated grid cells to make their way through a virtual maze.

Two AI systems, one with the virtual grid cells and one without, were incentivized with reward signals if they could navigate the maze. Both systems were taught a path through without shortcuts. When doors were opened that offered new shortcuts through the maze, the AI with grid cells took advantage of the more direct route, while the other AI did not, using the original, longer route instead. The AI also outperformed a human expert at completing the mazes.

University of Texas computational neuroscientist Ingmar Kanitscheider, who was not involved in the study, called the research “a big step forward” in our grasp of human navigational neural circuitry.

Not only does the research support the theory that grid cells help mammals plan routes through the environment, but also shows how AI can act as “a very powerful tool” in experimenting with neuroscience theories, according to Barry. The research team suggests that such artificial neural networks, simulating different areas of the brain, could replace animal testing for many of these experiments.

However, since such AI systems are built to learn on their own, it’s impossible to find out why a certain decision was made, according to experts like Francesco Savelli, a neuroscientist from Johns Hopkins University who provided commentary on the research. While researchers are sure that the simulated grid cells aided navigation, it’s not yet clear how they do so.

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.