Processing stories
There’s been a lot of chatter about this whole big thing called Artificial Intelligence (AI). This is a term as wide and nebulous as is it trendy, but I’m interested in one bit of it, Natural Language Processing (NLP), and what it can do for things like Bastion.
I’ve had some experience designing around NLP twice in the past that I can’t talk about. One time was for the government and the other time was for a massive tech company that might as well be the government. Both times it was about trying to extract meaning and connections from what otherwise is a bunch of otherwise flat text.
When dealing with a large story, made out of a lot of this text, the primary goal of Bastion is to make it liveable or otherwise manageable. Sure it already lets you choose how you want to approach and navigate the story world and that, but what if it also could take a bunch of text and make sense of it?
The idea of Bastion is that you can navigate and explore more than just plough through. It doesn’t do anything beyond providing a structure to browse and choose what to read. There is no notion of providing understanding automatically. Right now, with any writing, the author has to go through and assign things to all that text – this guy does this, or this takes place there. What NLP can do is analyse the text and do things to it and around it auto-magically. For instance, it could produce (reasonably) readable summaries, or even extract which is the main character in a section and even map them to the other main characters.
The possibilities are quite amazing and can take things beyond static books. You could build a chatbot that could act like it’s one of the characters you’re reading. Or the book-software could generate things you’re about to read based on what you’ve already read. As an author, you could even have it generate itself based on things that you have already written. What would be interesting to see is how you could use NLP to create plots and actions and then give them a brush of human and interestingness that we as humans are still most of the time better at than machines.
So where does it stop being your book if a machine is writing half of it for you? The dream is, of course, to chuck a bunch of half-assed notions and witticisms on post-it’s and have the machine produce Frantzen for you. Chances are we’re nowhere near this, and that’s probably a good thing. What we do want though is tools to help make us better writers and readers, and that’s something hopefully we’ll soon get around to doing with things like NLP.