Google DeepMind's Artificial Intelligence Just Got A Lot Smarter

It can now learn from its memories.

Google DeepMind’s artificial intelligence (AI) was already pretty smart, but it just got a whole lot smarter.

When its Alpha Go programme beat world-class Go players earlier this year, it did so by studying millions of Go moves and predicting its rivals’ responses.

But its memory was still pretty limited in comparison with a normal computer’s.

Google’s researchers have now made it even smarter by marrying the brain-like processor known as a neural network with an external memory.

“These models [...] can learn from examples like neural networks, but they can also store complex data like computers,” DeepMind researchers explained.

Kim Hong-Ji / Reuters

The new system, a Differential Neural Computer (DNC), will be able to draw on everything it’s ever learned, helping it tackle even more complicated tasks.

DeepMind researchers used a couple of fairly simple examples to explain how the DNC is able to navigate around its own memory.

In a short video, researchers show the DNC registering the different relationships in a family tree.

The DNC can deduce, for example, that Freya’s maternal great uncle is Fergus by connecting the dots between different relations.

In a DeepMind blog, researchers Alexander Graves and Greg Wayne said that the DNC provides a new way for scientists to think about the brain:

“The question of how human memory works is ancient and our understanding still developing.

“We hope that DNCs provide both a new tool for computer science and a new metaphor for cognitive science and neuroscience: here is a learning machine that, without prior programming, can organise information into connected facts and use those facts to solve problems.”

Close

What's Hot