With the advances of modern data storage technology, chips the size of your fingernail are capable of storing an entire library’s worth of knowledge, so one thing you might think computers do better than people is remember things. But according to Google Inc.’s DeepMind team, the artificial intelligence research group that developed AlphaGo, that is not entirely true.

In a new paper published in the journal Nature, DeepMind has outlined a process where it trained a neural network to have human-like memory, giving it not only the ability to store data, but also to recall that information and use it to solve novel problems.

“Neural networks excel at pattern recognition and quick, reactive decision-making, but we are only just beginning to build neural networks that can think slowly – that is, deliberate or reason using knowledge,” the DeepMind team wrote in a recent blog post. “For example, how could a neural network store memories for facts like the connections in a transport network and then logically reason about its pieces of knowledge to answer questions?”

DeepMind calls its new method differentiable neural computers, and the team demonstrated its capabilities using the London Underground, one of the largest public transit systems in the world.

“When we described the stations and lines of the London Underground,” the DeepMind team said, “we could ask a DNC to answer questions like, ‘Starting at Bond street, and taking the Central line in a direction one stop, the Circle line in a direction for four stops, and the Jubilee line in a direction for two stops, at what stop do you wind up?’ Or, the DNC could plan routes given questions like ‘How do you get from Moorgate to Piccadilly Circus?’”

The DNCs’ capabilities have some obvious value for a number of Google products and services, including the company’s self-driving car initiative. The highway system is several orders of magnitude larger and more complex than the London Underground, and an AI that could intuitively create routes between locations would be an incredibly powerful tool for an autonomous vehicle.

DeepMind also demonstrated the ability for the DNC to build a complete family tree based on information like “Natalie is the daughter of Alice” and “Ian is the husband of Jodie.” The system could then answer questions such as “Who is Freya’s maternal great uncle?” Obviously, there is already genealogy software that can accomplish a similar task, but what sets DeepMind’s DNCs apart is the fact that they are not programmed to do this. Rather, they develop the ability almost intuitively.

“The question of how human memory works is ancient and our understanding still developing,” the DeepMind team concluded. “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.”

You can watch a video showing the DNC’s family tree process below:

Image courtesy of Alphabet