Communications of the ACM
Google DeepMind's AI Learns Human Navigation Skills
An artificial intelligence developed by Googles DeepMind trounced experts at a maze game after it learned to find its way around like a human.
Credit: Jason Hawkes/PA
Google's DeepMind unit has developed an algorithm that outperforms people in solving a virtual maze, after noting that it spontaneously generated electrical activity similar to that of "grid cells" governing human navigational skills.
The scientists first built a deep neural network and taught it navigation fundamentals, inputting the types of signals that encode speed and direction in the brains of foraging rats. Feedback caused the network to improve its predictions of its location as it navigated a virtual environment.
The team observed that 25% of the artificial neurons in one network layer had begun firing like organic grid cells. They then assembled a more refined network and applied it to the maze game.
Tests revealed the algorithm not only employed grid cells for position tracking, but also to formulate the direction and distance to its objective so it could follow the most direct pathway.
From The Guardian
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA
No entries found