Computers have become an indispensable part of today's world, making tasks such as stocking shelves and operating cars much easier. Still, it looks like the devices won't be superseding authors anytime soon: It turns out that computers are not so great at creating original poems, short stories, or other creative works.
Scientists, poets, and others determined this in May 2016 after taking part in a Dartmouth College competition during which nouns—including wave, tourist, and floor—were given to computers that scientists had programmed to produce sonnets using artificial intelligence algorithms. The computers' task wasn't an easy one: A sonnet is a 14-line poem that adheres to a certain rhyme scheme, a specified rhythm, and a set structure.
On the night of the competition, a three-judge panel, which included Pulitzer Prize-winning author Louis Menand, was asked to read ten sonnets—six that were written by human contestants and four that were created by computers using two different algorithms—to see if they could identify the author as a human or a computer. In every instance, the judges were able to find the sonnet produced by a computer program because it didn't have the flow or narrative of a good poem. Some also had "idiosyncrasies of syntax and diction [or] uses of language that were just a little off," Menand said.
The competition included a short story segment and one involving computer and human disc jockeys (DJs). Computer algorithms were not much better at writing short stories than they were at creating poetry, failing to fool all but one judge, who was duped by a single story. But the sets of music, which were played from behind a black curtain, were more of a challenge. Dancers were asked to determine whether humans or computers had produced the various sets, and they struggled to do so accurately. Two algorithm entries managed to confound about 40 percent of the dancers.
Competition cofounder Dan Rockmore, a Dartmouth professor, said he was surprised at the computers' poor showing regarding the sonnets—but not that surprised, given that the competition was set up so that the judges knew they were looking for computer-generated poems.
"The judges were hunting for machines…not looking at a [greeting] card and reading the poem inside," Rockmore said.
Michael Casey, a music professor at Dartmouth who also helped organize the competition, said the results demonstrated the challenges faced by machines when they try to mimic the arts. It's hard to find an algorithm that can replicate the nuance of a story or the form and precision in a poem, Casey pointed out, though he said that he wasn't giving up on the idea that one day a computerized William Shakespeare could emerge.
"By doing this once, we may be able to encourage whoever is out there working on this kind of thing to take part and maybe we will get better algorithms," Casey said.
However, Sherry Turkle, a professor at the Massachusetts Institute of Technology who did not participate in the contest, disputes the whole idea of trying to get a machine to create art, such as poetry.
"Poetry needs to come from the experience of human meaning. That is what gives it life," she said.
Rockmore and Casey pointed out that Internet and social media algorithms already influence people's choices by feeding them lists of books they would likely want to read or music they might want to hear based on their earlier preferences, so it could just be a matter of time before computers generate the actual content, with the programmer of the algorithm replacing the writer as the artist.
"What if you could write beautiful stories that made people happy at the snap of your fingers? That would be a wonderful thing," Rockmore said. "It wouldn't mean humans weren't writing great things, too. They are both different art forms."
Vocabulary
algorithm (noun):a set of step-by-step instructions used for solving a problem
idiosyncrasy (noun):a peculiar characteristic
nuance (noun):a subtle characteristic of something
supersede (verb):to replace in power, authority, effectiveness, or acceptance