The Book that Writes Itself
We need a German word for “thinking you had an original idea and then realizing many other people not only already had that idea but are well on their way toward implementation.” I hope someone can get on this. I bet someone already has.
A while back, I blogged about the possibility that one day my job will be taken over by machines. I think it’s important for all of us to consider this possibility, whatever it is that we do, and however outlandish the idea seems by current technology standards. How else will we see it coming? There’s a reason no one does. They all think their status is wholly unique until about three weeks after it isn’t.
Will machines ever write novels? That is, will novels ever write themselves? I believe if humans can stick around for another thousand years, it is inevitable. I’m also open to the chance (though skeptical) that some unforeseen advance in computing power or technology makes this possible in fifty years. Perhaps an actual quantum computer is constructed. Maybe in 50 years, a program like Watson gets more refined and has access to enough data and processing power that an emergent quality arises from what previously seemed wholly mechanical. That is, consciousness might flip on like a switch.
But how could this ever happen? How could computers ever learn to be creative? One of the answers to that question might be that creativity is more mechanical than we give it credit for being. This would be the Joseph Campbell school of thinking, where every protagonist is an archetype and every journey a hero’s journey. Or look at the work of Vladimir Propp, who detailed the 31 steps he thought every folk tale could be reduced down to.
Game theory and evolutionary psychology hint at the way that the human mind is both universal and largely predictable. Cultural relativists have no good answer for why we can read Sir Gawain and the Green Knight today and have it resonate with us across such vast time and in so many translated languages. The vast majority of fictional stories that work do so by satisfying our expectations. My romance friends will tell you that it isn’t a romance without an “HEA” or “Happily Ever After.”
This doesn’t diminish any genre by saying it’s predictable; it simply says something incredibly poignant about who we are. Imagine a murder mystery without clues or where the gumshoe never solves the case. Can you do it? Sure, but at some point the avant garde becomes a self-mastabatory exercise in trumping tradition just for the sake of obstinance. Mystery readers crack a book expecting the hero to nab the villain. The suspense is in how that happens. Just as the suspense in romance is how torn lovers will make their impossible union work.
In the formulaic there is the possibility of formula. In pondering how this might work, I drew on my expertise as a young writer who employed an advanced creativity enhancer known as Mad Libs. By supplying a few words into an existing framework, I created the unique with the help of some scaffolding. Perhaps the same could be done with fiction.
In some ways, we are already comfortable allowing computers to randomly assemble elements of our story. They’re called random name generators. It is often helpful to have a computer throw options in our faces rather than having to make up names ex nihilo, especially when we need a lot of names or a name that isn’t important but might bog down our writing flow. Here’s a random name generator that allows you to pick the gender and how common or rare the name is. There are many out there.
What if rather than names, we had plot elements thrown at us by a random story generator? If this feels like a violation of our creative sensibilities, then what are story prompts? What do we call it when a random newspaper article inspires us to write a piece of fiction? I can imagine a tool that works something like this (with the thing in brackets coming from a massive bucket of like-type options).
[Protag] desires [Thing]
[Obstacle] is encountered.
[Protag] meets [Friend]
Duo meets [Enemy]
[Item] is discovered. [Obstacle] overcome. [Enemy] defeated.
Could this tool write a story from beginning to end? No way. Could it create a prompt to get someone started? Absolutely. Just like a random name generator can give us a boost.
Thinking this was an original idea, I then discovered that a group of coders have made considerable progress on just such a tool. Using Propp’s 31 criteria, they have assembled a random-plot generator known as the “Bard.” It is extremely crude, but the idea is to create plot prompts and ideas for low-intensity, high-volume situations. Like video games, where a group might need to come up with thousands of side quests. Just as we might not subject our protagonist to randomness, a video game studio might not yet use a tool like this for the main plot, but if these programs improve, then side quests would certainly qualify.
I put many hours into a video game back in the mid 90s known as Daggerfall that had randomly generated quests. You met random NPCs, who wanted a random item, which was guarded by randomly generated enemies in a randomly-generated dungeon (no two were ever the same). It was a much-touted advance at the time. It was buggy, but we loved it. A program was telling us stories. Many who played the game probably had no idea this was even taking place.
Sports lovers are in the same position today. They have probably read a synopsis of a game that was written by a computer program. The Big 10 Network has been using the technology since 2010. These programs are getting better, and computers are getting faster and faster, while their repository of data and examples grows exponentially. This is like neurons piling up. At some point, the level of complexity creates a new sort of output.
This is how computers obviated the need for legal discovery teams by learning to “read” millions of legal documents to prepare for trial. This is not a simple search function. These programs use something like an understanding of language to draw inferences and find precedents that teams of people would miss. And at a fraction of the cost and time.
Let me step back a moment and say here that humans will never stop writing stories and telling stories. It’s in our DNA. But that doesn’t mean the consumption of stories from other people is in our DNA. If we can’t tell the two apart, how do we stick to our principles as consumers? We won’t be able to.
In one of his dozen championship games with Deep Blue, Kasparov said he saw something that felt like true creativity in his opponent. He even intimated that the IBM team was cheating at one point by employing a grand master to take over the machine. He may have been one of the first people to gaze into the eyes of infant AI, feeling that sense of wonder at a small hand closing around his finger in response to being touched. Ken Jennings may have been the next to feel this. Make no mistake, what Watson did in mastering Jeopardy wasn’t just astounding, by some informed people it was considered beyond the realm of possibility right up until the day it happened. Also keep in mind that chess was once considered one of the highest expressions of our creativity. Everyone seems to forget this. Once the future arrived, we rewrote the past to make it seem inevitable.
Now imagine a tomorrow where people-written books and computer-written books sit on the same shelf. Both sets have authors’ names on them. All variably please some readers and not others. The prices are the same, so we can’t distinguish that way. But the computers can write billions of stories a year, and humans can only muster a million. What happens then? The odds are against us. And if you think we’ll be able to “badge” human stories, you might want to hear that master-level chess players have been caught employing computers during tournaments. Similar cheating will take place in this future. Like the sports fan, you’ll read what you think was human generated, but some of it will be from a computer. Who can guess which of my character names, if any, came from a machine?
There are many “first steps” in this process. In the world of coding, which has much in common with writing fiction if you subscribe to the Campbell and Propp view of storytelling, DARPA is working on a system that will automate programming from very few inputs. Automatic novel generation will happen when random plot generators are married to just this sort of text expander. “Fist fight” results in a gripping blow-by-blow action scene among the story’s characters. “Love scene” gives us a tender moment of appropriate length and reading age. Is it in the realm of science fiction today? Absolutely. But someone is already working on it.