Yes, Google's DeepMind is now able to teach itself based on the information it already has.
In a major step for artificial intelligence, Alphabet's hybrid system, called the Differential Neural Computer (DNC), uses data already stored by conventional computers and matches it with a highly intelligent AI and a neural network capable of rapid parsing.
“These models can learn from examples, like neural networks, but they can also store complex data, just like computers,” said DeepMind researchers Alexander Graves and Greg Wayne.
What does this mean?
Just like a brain, a neural network uses interconnected nodes to stimulate specific centers needed to complete a task. In this case, AI optimizes the nodes to find the fastest solution and deliver the desired result.
According to the researchers, over time, it will use the acquired data to become more efficient at finding the correct answer.
So instead of having to learn every possible outcome to discover a solution, from the information fed to it by programmers, DeepMind will be able to find answers from its previous experiences, from its internal memory and not from external programming.
Depending on how you look at it, the event could be a serious turning point for smarter AIs that may one day be able to think and learn just like humans do.
https://deepmind.com/blog/differentiable-neural-computers/
