Game of Thrones author George R. R. Martin is a notoriously slow writer. Did you know the original Game of Thrones was first published in 1996? The latest book in the series, The Winds of Winter, has been due for years. Martin insists it’s still on the way, but there are no signs of that happening anytime soon.
Fortunately, artificial intelligence seems more than happy to step in and take things over.
Software engineer Zack Thoutt is currently working on a recurrent neural network that uses Martin’s past source material — the first five books in the series — to write what it believes will happen next. Spoiler: it writes a lot faster than George R. R. Martin.
Thoutt recently completed a course on Udacity which studied artificial intelligence and deep learning. The neural network, which should absolutely be called George A. I. Martin, was Thoutt’s project for the course.
So how does it work? Thoutt explained the process the neural network uses to the website Motherboard.
“With a vanilla neural network you take a set of input data, pass it through the network, and get a set of outputs,” said Thoutt. “In order to train these models you need to know what the model should ideally output, which is often called your labels or target variables. The neural network compares the data it outputs with the targets and updates the network learns to better mimic the targets.”
In order to prevent the network from repeating items over and over — which, frankly, would be pretty boring reading — Thoutt is implementing what is called a ‘long short-term memory’ recurrent neural network, which remember important points in stories so it doesn’t make the mistake of using them over again.
I don’t want to get into any real spoilers, but I will say this: the network has made some pretty on-point predictions — some that superfans of the show have been speculating about for quite some time. The entire project is truly impressive, and one can’t help but wonder: could artificial intelligence have a serious future in storytelling?
Think about the places A.I. could have a real narrative impact. Video games, for instance, are already capable of procedurally generating worlds and ‘quests’ with limited variables. The ability to weave a whole new story for a new world could produce something truly special: a game that truly never ends. And another exciting prospect for fans of authors who have long since passed: the works of these authors could live once more, using neural networks similar to Thoutt’s to learn their styles and tendencies and create new works. We could get a brand new Chronicles of Narnia entry from a sort of virtual C.S. Lewis, which raises another important question: could his estate sue?
I for one, welcome our new robot storytelling overlords. What about you? What do you think about Thoutt’s neural network authoring a new chapter in the Game of Thrones series? And how do you feel about A.I. potentially telling stories of their own in the future? Leave your thoughts below.