Computers do a lot of work for us— control components in our cars, help us check out at the grocery store, and count our money at the ATM. But can computers create “human-quality” music and literature?
A new series of contests at Dartmouth College is seeking algorithms that can create works of art that pass as creations of human musicians and authors. Dan Rockmore, the director of the Neukom Institute for Computational Science at Dartmouth and creator of these contests, spoke with NHPR's Peter Biello.
The contests are called DigiLit, PoetiX, and Algorhythms, and starting next school year they’ll put computer-generated works of music and literature to the test. What will those tests be?
DigiLit is the challenge to write an algorithm that will create human-like short stories like you may have found in the output in a class in creative writing. PoetiX is the challenge to write an algorithm that can create a sonnet. And Algorhythms is the challenge for a computer to act like a dance floor DJ of all things, so it mixes in a beautiful and energetic way a playlist that an audience responds to.
How will they be evaluated?
For the literary stuff, we’ll have panels of judges and the judges will read the machine-generated work intermingled with human generated work. It all sounds so antiseptic, doesn’t it? And they’ll just try to find the one or guess the one that was done by a machine. In the case of the dance music, that’s a little more complicated. The way in which we envision it is that the actual creator of the music will be hidden from the dancers and then they’ll vote thumbs up or thumbs down (so to speak) as to whether their set was machine-generated or human-generated.
Computers have been generating stories for a little while now, and they’ve been used to create news stories, for example.
Yes, it’s true. We’ve seen in the news lately that machines can generate things like journal articles or business reports. The truth is, I mean, machines have been doing that for a long time, certainly from the point of view of business reports. An Excel spreadsheet is essentially a business report and you can interpolate it with some text, so machines have been providing us with information in a stripped-down narrative form for awhile. What I would hope is a little more interesting and more novel about our competition is this—well, in the short story competition, the Digilit competition, the machine—the algorithm must create a narrative. The algorithm is meant to respond to a prompt, a noun-phrase prompt, like hat or cat or dog or wedding—
So it may come up with something like, “A man lost his hat and goes on a journey to find it.”
Yes. There’s a bonus to having a story attached to it. Similarly, in the sonnet competition in PoetiX—again, it’s to respond to a prompt and it has to adhere to an interesting and fertile constraint. Both kinds of strictures—strictures as well as freedoms I suppose—make these competitions a little bit more challenging.
And what would that say to you if the judges end up convinced that the computer-generated work was actually created by humans?
On the one hand, it might mean that what we’re reading today might feel more machine-like in some ways. On the other hand, it would mean, I suppose, that there are very creative people out there who can actually integrate literary ideas and computational ideas in such a way that they are creating, by their own way, interesting narratives and interesting poems. I don’t think we’ve reached any particular apocalypse when that happens, but what we might really find, should that happen, is that we’ve come to a new place in human creativity and machine processing.
There’s some prize money as well.
There is. Now you’re going to ask me what it is.
I take it the money is not the important part for you.
You know, it’s not. The money’s a nice carrot for a lot of people. If any of the computers pass the challenge, so to speak, of creating something that we believe is human-generated, those teams win $5,000. Then, if none of them succeed, then there are a range of prizes for just doing great work as judged by the panels. My guess is that most people won’t be doing it for the money. It’s good work, it’s nice to see good work rewarded, and the money, to be honest, does give the competition a little more visibility. What my collaborators and I hope for is that this inspires all sorts of interesting new work. There are all kinds of challenges in computer science, and potentially interesting or novel sort of twist is to have this humanities kind of integration with it.