Google’s DeepMind AI just taught itself to navigate like humans
Approximate Reading Time: 2 minutes
Google’s sister company DeepMind recently created an AI which spontaneously taught itself the machine learning equivalent of gut-based navigation. This is quite similar to the crazy labyrinth puzzles we did as children.
In a recently published paper, DeepMind demonstrated how a neural network, on trying to solve a navigational problem, developed a method of spatial awareness. Interestingly, researchers found that this development imitates the creation of “Grid Cells” in mammals.
The phenomenon of Grid Cells was discovered in 2005, which was found to occur within the brains of mammals in order to aid navigation. In simpler words, it is our ability to generally understand where we are based on how far we’ve traveled and in what direction, and this judgement is ruled by these specialty cells, which form in hexagon-shaped patterns that the brain sort of overlays into space, causing neurons to fire when we move through it.
However, what’s more interesting about this is that while the phenomenon was discovered, scientists are still unsure how this actually works.
Apparently, DeepMind’s AI was going through these theories of Grid Cell, when researchers noticed it developed a system of its own that imitates human Grid Cells.
DeepMind’s blog reads:
“As a first step, we trained a recurrent network to perform the task of localising itself in a virtual environment, using predominantly movement-related velocity signals. This ability is commonly used by mammals when moving through unfamiliar places or in situations where it is not easy to spot familiar landmarks (e.g. when navigating in the dark).
We found that grid-like representations (hereafter grid units) spontaneously emerged within the network – providing a striking convergence with the neural activity patterns observed in foraging mammals, and consistent with the notion that grid cells provide an efficient code for space.”
Then to find out if the AI actually picked up on phenomenon, DeepMind used reinforcement learning to reward the AI for successfully traversing virtual game environments using vector-based navigation.
While as the AI formed its version of Grid Cells, the result was immediately seen to get worse, however, allowing the AI to develop its own version of Grid Cells gave it superhuman navigational abilities.