Artificial intelligences are becoming better storytellers by the day. Last week, a novella written by an AI program nearly won a Japanese literary contest. “The Day a Computer Writes a Novel” (Konypyuta ga shosetsu wo kaku hi) is a surprisingly human tale of an AI that recognizes its writing skills and abandons its programmed task of aiding humanity in order to satisfy an artistic urge. The Japanese News reports (in an article that appears to be taken down at the time of this article update, September 2017) that this meta-novella and 10 other AI-authored submissions faced competition from over 1,400 man-penned manuscripts for the Hoshi Shinichi Literary Award.
Like a literary Turing test, judges were blind as to whether manuscripts were written by humans or programs. They critiqued the submissions based strictly on literary merit. Although “The Day a Computer Write a Novel” successfully passed the first of four rounds, judge and science fiction writer Satoshi Hase relayed at a press conference that the program lacked fully developed characters, and thus failed to advance beyond the second round.
Nonetheless, this marks the first time an AI program progressed beyond the initial phase of the Hoshi Shinchi Literary Award, a competition recognized for accepting submissions from both human and machine authors. In the wake of AI painter, The Painting Fool, and AlphaGo’s unexpected defeat of Go master Lee Sedol, it might seem AI are invading our most precious artistic and leisurely pleasures.
But is an AI author such a bad thing?
There are plenty of texts that even professional writers find no joy in writing. Emails can be rote and laborious. Financial reports are as drab as they are redundant. The occasional blog post can even be an effort in endurance. While AI has conquered centuries old boardgames and competed in literary competitions, it’s also tackling writing tasks that we’d rather not do anyway.
(For a deeper dive on this topic please see our full article on automated journalism.)
Managing Your Email Overload
Two weeks ago, Google rolled out an updated version of its automatic email answering system, Smart Reply for Inbox for the web. The machine learning algorithm offers three brief suggestions, such as “Thank you,” “Sounds good,” and “OK,” which users can select to send to edit. Senior Research Scientist on the Google Brain Team Greg Corrado told Wired, “The first decision is made by an artificial neural network very much like the ones we use for spam classification and separating promotional emails from personal ones.”
This neural network enables the system to determine whether or not the email is appropriate for a quick reply. As such, the consumer version can’t yet offer answers to binary questions with answers more complicated than yes/no, but it’s still designed to learn and develop with users. Indeed, Corrado claims the “research grade” version of Smart Reply successfully responds to binary questions much of the time, though it also often answers things in hysterical ways that aren’t quite conducive to professional correspondences.
Anyone, from students to business professionals, would find value in quick reply emails that become more refined with continued use. Smart Reply supposedly learns from user input and contextual cues to adopt more natural language over time. Even though the system isn’t particularly sophisticated at present, user engagement should help it develop. So, if you currently use Google’s Inbox and notice these options in the reply box of your next email thread, it might not hurt to give it a try. But don’t turn to Smart Reply as a practical option for truly putting a dent in your email inbox (yet).
Blogging and Reporting
Some of the articles you currently read are actually written by AI. Like the Hoshi Shinchi Literary Award judges, most of us probably can’t tell the difference. For example, organizations like the Associated Press, Yahoo, and Comcast use Wordsmith software offered by Automated Insights (or it’s main competitor, Narrative Science), which automatically crafts blog posts, business reports, and sports summaries from spreadsheets. Wordsmith inputs data into user-designed sentences like its playing a game of ad libs. You can read our full description of the software’s capabilities here and watch the company’s explainer video below.
Wordsmith seems ideal for statistical analyses and writing tasks in which numbers are the most important part of the content, though Automated Insights insists Wordsmith can also tackle important marketing narratives as well. But how does the software hold up against it’s human counterparts? National Public Radio (NPR) ran a quick study that faced off on of its White House correspondents against the Wordsmith system. Though Wordsmith wrote its article at an incredible speed, NPR listeners voted for the human correspondent-written article in terms of readability.
Does your business need an AI author? That depends on what content you need.
Automated responses like Google’s Smart Reply perform well in moderation. The more complex the email reply, the less Smart Reply sounds like you. When it comes to blog posts and reports, the speed and efficiency at which AI can construct texts – with the help of spreadsheets and some human guidance – makes them ideal for content that requires a more systematic approach. But AI haven’t yet surpassed humans in crafting engaging narratives. Even the ‘greatest’ AI novelist has only just managed to infiltrate the literary word.
Image credit: Casey Gutteridge/SWNS.com