Google Deepmind Develops Artificial Intelligence That Can Mimic Human Brain Cells
Whenever anyone needs to go from one place to another, their brain automatically calculates the best route based on previous experiences. And if there is traffic or they want to go another way, the brain simply computes the alternative route.
Our brains have an in-built navigation system that enables us to find our way. However, what is still largely unknown is the actual computation that takes place when we are planning any route from A to B.
But now, with the development of artificial intelligence, neuroscientists may be able to gain a deeper understanding of the ‘navigational circuitry’ of our brains.
A team from Google DeepMind has been able to develop an artificial intelligence that mimics that of the human brains navigational system as a virtual being navigates throughout a virtual environment.
Grid cells are what enable us to navigate and the scientists at Google DeepMind and University College London have been able to develop neurons that act in a similar manner.
These cells are known to not just function alone, with the help of “place cells” that locate a specific location and “head direction cells” the action when our head points in a specific direction.
In order to find out if this incredible evolution could be transferred into machines, a team at Google DeepMind began constructing a neural network. Neural networks are a series of interconnected workings that are inspired by the way the human brain works. More layers mean a deeper, and better understanding network much like the human brain.
They have developed a number of computer programs that have been able to dominate professional board game players, learning the game in as little as a day. However, until the latest advancement these studies have failed to contribute the neurological understanding.
The DeepMind team managed to train a neural network to navigate through a 2.2m x 2.2m environment with the addition of distance and track integration through what is called path integration.
The neural network also fired the grid units that are seen in the human brain and managed to successfully navigate through the virtual environment.
The study confirmed that grid cells are vital to any animal’s ability to navigate regardless of whether they have been to the location before.
Dharshan Kumaran, a co-author for the study, said that this gives “evidence that the function of grid cells may extend far beyond giving a localisation signal.”