A recurrent neural network has taken up the task of writing the next book in George RR Martin’s A Song of Ice and Fire series ahead of season 8 of Game of Thrones.
Game of Thrones has solidified its claim as one of the most-watched television series in history – racking up an estimated 16.5 million viewers during season seven’s ultimate episode, The Wolf and the Dragon – and all enthusiastic viewers have left before the next season of the show debuts is the hope that George RR Martin produces the next book in the series’ genesis saga, A Song of Ice and Fire.
Unfortunately for the impatient among us, Martin has taken his time to craft The Winds of Winter – the next entry in his epic saga. Martin is at present still writing the novel, meaning that we may not see the work for some time to come.
However, one artificial intelligence – created by software engineer Zack Thoutt – has taken up the challenge of writing the novel itself.
The artificial intelligence – more accurately called a recurrent neural network (RNN) – has digested all content with the first five books of the series and is now endeavoring to write the sixth on its own, all the while mirroring Martin’s writing style and introducing new plot elements of its own creation.
Speaking to Motherboard, Thoutt revealed that he offered the network a prime word with which to begin each chapter, and then was able to instruct it to write a pre-determined amount of words. The network was then left to its own devices, which allowed it to generate some interesting fan fiction; including stories where a new character called Greenbeard appears and Daenerys Targaryen is poisoned by spymaster Varys.
Thoutt offered that though Martin’s work is substantial in volume, the network would fare far better in its efforts if it had a great deal more material with which to work and if Martin’s prose were more simple.
If you have trouble recalling who’s who in Game of Thrones, you’re not alone; Thoutt offers that the network struggles to maintain consistency and remember which characters have died. “Martin is obviously very descriptive in his writing, so those extra adjectives and the fictional locations and titles are just more complications for the network”, Thoutt offered.
Another Song of Ice and Fire project has been led by Inverse, wherein interested readers and viewers have been asked to submit theories on how the final season of Game of Thrones will pan out. Once that data has been acquired, the project is set to form a predictive model of what might happen in season eight.
What are your thoughts? Be sure to let us know your opinion in the comments below!
Read the fiction created by Thoutt’s recurrent neural network here