Google DeepMind artificial intelligence better at navigating mazes than puny humans

Hampton Court maze would be too easy for DeepMind-powered AI

DeepMind, the London-based artificial intelligence outfit owned by Google, claims to have devised code enabling robots to out-think puny humans when it comes to navigating mazes.

DeepMind's latest AI, the company says, can figure out the route through a maze and spot shortcuts far quicker than any human. That's according to the research written up in the journal Nature.

The AI did this through the use of artificial takes on 'grid cells', which help handle navigation in the brains of mammals by working out the position of an animal or human in an environment by effectively figuring out the distance between two points - a vector - through the use of a series of meshed hexagonal grids.

The discovery of a pseudo internal GPS for mammals in 2005 means it's a relatively recent addition to the knowledge of biology and neuroscience.

But DeepMind trained a neural network on data taken from observing how simulated rats navigate 2D enclosures. From there, the AI system was found to have electrical patterns similar to the hexagonal ones in mammal grid cells, indicating that the AI had created its own take on the biological GPS found in humans and animals.

With the ability to tap into pseudo grid cells, the AI was able to trounce humans at navigating a maze.

"The performance of agents endowed with grid-like representations surpassed that of an expert human and comparison agents, with the metric quantities necessary for vector-based navigation derived from grid-like units within the network," explained DeepMind's researchers.

"Furthermore, grid-like representations enabled agents to conduct shortcut behaviours reminiscent of those performed by mammals. Our findings show that emergent grid-like representations furnish agents with a Euclidean spatial metric and associated vector operations, providing a foundation for proficient navigation."

When the grid cells were removed from the AI it didn't perform as well, which indicates said cells are a key part of effective navigation even in previously unseen or visited environments.

DeepMind and its University College London partners reckon the AI tech could be used to better understand the way human brains work, though it hasn't actually shown anything new for the neuroscience world to get its teeth into.

It's also worth noting that even the world's neuroscience top brass still don't really know how the brain works.

So the chance that Google's AI nerds working on tech that's essentially a less complex artificial take on the neurons and synapses of mammallian brains will crack open massive insights into brain function is somewhat doubtful.