Child Reading Facebook AI Test
Could we soon be reading bedtime stories to robots? iStock

Facebook has released a conference paper outlining what it dubs The Children's Book Test (CBT), a benchmark designed to test its own neural network's ability to understand semantic connections within written text.

The test questions whether an AI network can accurately predict words or phrases that have been purposely removed based only on the context of the rest of the story. In other words, Facebook wants to know if its artificial intelligence can go beyond just reading and instead actually start to understand what it is reading.

With a cavalcade of timeless children's literature to choose from, the researchers settled on a list of more than 100 books, with works from the likes of Charles Dickens, Lewis Carroll, Robert Louis Stevenson and Rudyard Kipling. The novels used for the CBT were all obtained for free from the Project Gutenberg online library.

Facebook neural network test
Two examples of how the AI network examines a word gap (blue) in a children's story and replaces it with the correct word (red) based on the contents of the passage. Facebook / arXiv

While the scientists from Mark Zuckerberg's company have presumably carefully considered the stories used in the testing process, it does make you wonder whether an AI creation will ask why it does not grow up after reading JM Barrie's Peter Pan, for example.

This newly revealed research follows on from Facebook's mission to teach its neural network how to differentiate between objects in photographs and other images.

It would appear storytelling could be key for the future of artificial intelligence, as while Facebook is teaching its networks how to learn for the greater good of machines everywhere, other researchers are using interactive fiction to save the human race from oblivion at the hands of a robot uprising.