I want you to do me a favor and watch these two trailers for the same film:
I know which one I like better. The first trailer captures the mood, while the second tries to give us the plot. As someone who now avoids movie trailers, because I’d rather watch the 120 minute version of the film rather than the 3 minute version, the first trailer is exactly what I’ve been begging Hollywood to give me for well over a decade.
Only, Hollywood didn’t give me the first trailer. IBM’s artificial intelligence Watson did.
You can read about the process here. And yes, a person was involved in creating the final cut, but it’s easy to imagine this final step being automated in the near future.
Earlier this week, I met up with the legendary science fiction author and translator Ken Liu (you should totally read THE GRACE OF KINGS). Ken and I discussed the inevitable future in which machines and humans will one day write together, and the future beyond that when humans won’t be needed at all. I’ve blogged about this several times in the past, and Ken has written about the same topic in his coverage of NaNoGenMo, an annual attempt to create an entire novel using machine-learning bots. It will happen, the question is when, and what will the transition to there look like?
In another conversation this week, I was asked how long it would be before a machine wrote a novel that was better than the human counterpart. I believe we’re a long way off, but it’s striking to me that of the two trailers above, it’s the Watson-generated one that makes me want to see this film, and the human-created one that makes me feel like I’ve already seen the film. This is a unique circumstance of course, because the ephemeral and surreal leave me wanting more, while the chronological and story-driven teaser leaves me only mildly interested. Ephemeral and surreal might be what current AIs do best, as evident in this film, written entirely by an AI.
There are parallels here with children. I’m often shocked and moved by the things young kids say. Without a filter, and with a limited vocabulary, they often cut right to the heart of an issue with startling precision, or they dance around the unknown with stirring poetry. They see things differently, and they communicate differently, and many conversations are enlivened and enriched by their participation.
This is already true of artificial intelligences and machine learning artists.
We may mock them at first, the way we laugh when a baby babbles, but soon they are dropping insights, then painting striking imagery, then crafting wholly new stories, and then winning awards hidden behind the guise of a human author. I wrote about this last possibility in THE PLAGIARIST, where a man steals art from computer-generated worlds and passes it off as his own. I believe this will happen one day, and we will have to figure out how to cope with it. Who owns the art? The discoverer? The machine? The original programmer? What about when these bots are open-sourced, crowd-sourced, or self-replicating?
There are four primary stages or phases that I can see us going through before a wafer of silicon is awarded a Pulitzer. They are: Random Generator, Filter, Centaur, and Turing
We’re already in the first stage, Random Generator, because machines are helping us write novels now. Not everyone takes advantage of these tools, but I know of many authors who will fire up a random name generator when they’ve run out of ideas for secondary characters. The computer makes a number of suggestions, and the writer picks the one they like best.
This stage is going to increase in complexity, with plot points and locations offered up at random, as well as dialog choices. It’ll be a Mad-Lib style grab-bag, and the machines will learn from the users what choices win out (the way Google learns which search results are most helpful). In this way, they will only get better. We use writing prompts from professors and craft books already; the computers will be even better at this.
The second stage, Filter, will have a lot in common with an area in which we’re already creating art with the help of machines, a stage that Ken pointed out to me. Most of us have taken a photograph, swiped through a series of filters, and chosen the resulting pic that we like best. The Prisma App takes this to another level, creating works of art in various styles based on the photographs we feed it. We’ve been doing this both in photography and music via filters and effects for a while now. Soon, we will do it with the written word.
Imagine, for instance, taking a novel written in the first person, pressing a button, and reading through the same draft in third person. Or imagine a manuscript written in present tense, and quickly moving it all into past tense. This is in the realm of possibility today, if an engineer were so inclined. With a quick swipe, the flavor of the text is changed, and it’s up to the writer (or reader) to decide which style they like best prior to publication. Or better yet, just as readers can today change the size of the font in their Kindles, imagine being able to choose to read The Hunger Games in third person past tense, if you prefer this style.
There will be an uproar, of course, among some creatives who think the manner in which a work is presented should only be up to them, but that cat got out of the bag years ago, wrote some slash-fic, made a version of its favorite novel without any of the cuss words in it, read the last chapter first to make sure everyone important survives, and left a dead canary on the stoop. Artists are going to have to become comfortable with readers owning the right to read in whatever style they like. Person and tense will be suggestions by the author, not holy writ.
Imagine further being able to change the genders of characters, or their race, or sexual orientation. Now we cross into very interesting waters, because some of these choices are underrepresented due to commercial choices made by publishers. Will we lose some of the power of fiction if readers are able to choose to read only from the point of view of characters like them? What will this new power do for the empathy engendered by fiction? If used properly, it will be a boon. If used improperly, it will be a setback. Me? I’m for having more choices and options and seeing what the reader decides.
All of this will be moot as we get into the third and fourth phases of AI literature. The third phase will entail novels written by Centaurs, or the partnership between human and machine. Competitive chess is already in the Centaur stage, with these mixed teams beating any of the top players or computers when they don’t partner. One day, I will sit down with my writing program, give the barest of prompts, and the computer will spit out rough prose for me to revise. Or I will write, ask for a prompt, and get a suggestion in return. We will co-author the work. Neither of us could have written the final piece alone. This is some wildness to consider, but I reckon it’ll happen during my lifetime (I’d say we’re 15 – 30 years away).
Fourth stage is the Turing stage, the holy grail for readers, the end of writing as a profession, and I give it anywhere from 50 – 200 years. Entire novels will be written from scratch by machines, a million novels spurting out in the blink of an eye, and they will be tailored to individual readers, win major awards, and be as sublime and moving as anything we’ve ever read before. We balk at the idea now, but just as manually driving a car will seem insane one day (unless on a closed track by daredevils with death wishes), a handwritten novel will also seem bizarre. Why do with long division what a calculator on our phone can do for us?
Sure, people will still write, but very few will read these works. And the process will happen so gradually that hardly anyone will understand what has happened. We already read sports stories and financial articles written entirely by computers, and we either don’t notice or don’t care even when we’re told. Soon, these will be news articles, and pop culture articles. Things driven by fact. Then we will read reviews of books and film written by machines like Watson, who can analyze a work and tell us if it’ll be a bestseller (this is already a thing). From giving a simple grade between F and A, the machines will gain a language and an opinion to describe why a work is good or bad. We will get used to these developments in stages. The end result is inevitable. As a reader, I can’t wait.