Google’s Prototype AI Tool Can Write a Short Story. But Is It Any Good?

A conversational AI created by Google is writing fiction using input from published authors, the tech giant announced Wednesday at a Google AI event in New York. Thirteen writers used Google’s LaMDA, or Language Model for Dialogue Applications, to build Wordcraft, a model writing editor they tested by writing sample short stories.

Google introduced LaMDA last year, saying it’s designed to be “a dialog engine able to engage users in conversation.” The Wordcraft project explores fictional text produced by the engine. Unlike existing text tools such as WordTune or Grammarly, Wordcraft is intended to help craft fiction, not just improve spelling and delivery.

Stephen King needn’t be shaking in his boots just yet, though. Here’s a sample from Nebula-nominated fantasy writer Eugenia Triantafyllou’s submission, Worm-Mothers: “All the birds in the sky were prey to the Worm-Mothers — they hungered and delighted in the birds they swallowed whole like grapes. Because the Worm-Mothers had no teeth to speak of, no eyes either. Their face was only a wide gummy mouth and a short black horn at the top of their egg-shaped heads.”

You can read more of the stories from the Wordcraft Writers’ Workshop here.

At the event, Google Research senior researcher Douglas Eck said he believed Wordcraft would “transform how people express themselves creatively,” Indian Express reported, with Google confirming his comments. 

Eck said that crafting prose is “not easy” and that LaMDA’s solo efforts are not quite Pulitzer-worthy.
“One clear finding is that using LaMDA to write full stories is a dead end,” he said.  “It’s the writers who are doing the work.”

Google will continue to work to move the needle on what artificial intelligence can do in the arts, Eck said, adding that he viewed AI “as a spice — an addition to what you’re trying to do.”

According to Google, LaMDA doesn’t truly understand language, meaning or context, but after trawling through reams of data on the internet it can “produce deceptively humanlike speech” and predict the next most likely word in a sentence, The Washington Post reported in July.

Earlier that month, Google fired engineer Blake Lemoine, who claimed LaMDA was sentient. The company had suspended Lemoine in June for violating a confidentiality policy after he gave several media interviews regarding LaMDA.