Google team develops AI bot that can learn on its own

Scientists from DeepMind, Google’s artificial intelligence division, have created a computer programme that uses basic reasoning to learn complex systems such as the London Underground and family trees.

The advancement could mark a major breakthrough in the development of AI, as the “differentiable neural computer” (DNC) can solve problems without any prior knowledge. Instead, the DNC learns to use its own memory to answer questions about complex data. In a study published in the journal Nature, the technology also demonstrated it can solve a block puzzle game using reinforcement learning, says a report published in different websites on Saturday.

What makes the DNC impressive is that it can learn to form and navigate complex data structures all on its own.

The researchers demonstrated how the programme can analyse a description of an arbitrary graph and answer questions about it. The authors of the study, Alexander Graves and Greg Wayne, also explained how it works in a blog post.

“When we described the stations and lines of the London Underground, 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?,” the researchers write.

The DNC could also plan routes when asked questions like, “How do you get from Moorgate to Piccadilly Circus?”

When confronted with a family tree, the programme was able to answer questions that require complex deductions. Even though the scientists only told it parent, child, and sibling relationships, the DNC could figure out questions such as “Who is Freya’s maternal great uncle?”

DeepMind was founded as an AI start-up in 2010 and Google bought the London-based company for £400m ($487.6m) in 2014.

Artificial intelligence has long stirred people’s existential fears, inspiring numerous books, films and TV show. Even high profile scientists such as Stephen Hawking have expressed concern over AI developments.