First Words: The World of Automated Storytelling

automated storytelling

If you have recently read a financial report, sports game recap, or weather update, chances are that story was written by a computer. The process, commonly referred to as “automated storytelling,” involves running data through an algorithm which organizes that data into a readable story. The software is now widely accessible and relatively affordable thanks to companies like Narrative Science and Automated Insights.

The software is most suited for producing structured stories like financial news and job reports that maintain a pattern throughout, and news organizations like The Associated Press and The Washington Post are using automated storytelling to free up staff and increase their output. More recently, groups outside the media sphere have been working toward using automated storytelling to write more creative narrative works. But will computers really be writing entire novels in our lifetime? And, if so, should we let them?

Let us first address the logistics. How could a computer produce an original, meaningful, and well-written novel? To do so, consider how a media outlet might produce a hypothetical news headline: “Unemployment drops to 4.8 percent over the weekend.” You could pay a writer to come to work every day, read through dense economic reports, work out the necessary calculations, and write such a simple headline. Or, you could teach a computer to do it and use that writer to pursue more hard-hitting, investigative, and unstructured stories.

Francesco Marconi, manager of strategy and development at the AP, says the theory behind teaching a computer to write is fairly simple. “The way it works is that we create templates,” said Marconi. “So for this specific data point the machine is going to write this sentence, and it’s going to replace this value with that specific value on the table.”

In our hypothetical case, we would tell the computer to look for the latest unemployment data Monday morning and calculate the unemployment rate in that data set, then fill in the blanks in a prewritten sentence. Easy. But unlike earnings reports, good stories have the power to alter not only our understanding of the world, but also our neural chemistry. Researchers have discovered that good writing, especially writing full of metaphors and heavy descriptions, activate the same areas of our brain that are activated when we physically engage in the activities described so well in writing.

Think about your favorite novel — “The Great Gatsby,” “War and Peace,” “To Kill a Mockingbird” — whatever it may be. How might the same theories used to generate automated earnings reports be applied to a system that could allow a computer to write the next Great American Novel, complete with mind-altering language, structure, and storylines?

Charles Melcher, CEO and founder of the Future of StoryTelling, says the key remains in templates — there are templates for even the most complex, meaningful writing, just as there are templates for earnings reports, that can be used to output such works. “I’m a believer that a lot of what we call ‘creativity’ is actually seeing certain kinds of patterns,” said Melcher. “People are not creating in a vacuum, they’re iterating off stories that have come before, [so] how difficult would that be for a computer to digest 100 million stories, see certain patterns, create some new names, change a few things, or just combine pieces of many other stories to the point where it’s unrecognizable?”

Such seamless combinatory work is possible in part because every good story shares the same characteristics: compelling characters, detailed settings, strong narrative, and writing that elicits an emotional response. Furthermore, when you think about your favorite novel or story, you realize that there are certain parameters you as a reader accept before reading.

You know that a mystery novel will start with a murder, then cycle through a series of suspect characters, and offer a few twists and turns before revealing the killer. A romance novel will follow the typical “boy meets girl,” “boy loses girl,” “boy gets girl back” trope, et cetera.

Knowing these details, a programmer can then set the parameters (the template) for a story, given historical exemplars, and send the computer off to search through millions of data points (previous stories, for instance) and create a unique piece.

Fred Zimmerman, CEO of PageKicker, a startup that algorithmically replicates the entire publishing process, has already moved a step closer toward teaching computers to write unique, original, and long-form content, building on the theory that good stories maintain certain patterns and characteristics that can be replicated. “You give [PageKicker] a concept, it goes out and searches for related content, assembles what it finds, analyzes it, puts it into this formatted e-book, [then] distributes it to online bookstores and to online catalogs,” said Zimmerman. “It can [even] socialize and drive publicity for it, and it does all this in one command and about 30 to 60 seconds at the cost of pennies.”

In other words, the computer can reach into the depths of the internet and search through the appropriate literature before stitching it all together. It’s still not an original creation, but it is nearing what Melcher described with regard to a computer learning from other examples of certain types of stories to create its own algorithmic narrative.

But even if computers can one day write the next Great American Novel, the question remains — should they?

Melcher says one of the main advantages automated storytelling provides is the possibility of extreme customization. “Good writing is the kind of writing that people can relate to, they capture a time or a place or a community, so there are ways for computers to be able to do that beautifully, or potentially beautifully,” said Melcher. “You can create many more stories that are personalized for very specific purposes or people.”

“The simplified logistics and the potential for improvement are promising features associated with automated storytelling,” says Zimmerman. “Writing a book takes months and months of effort, it usually costs a few thousand dollars in fees, labor costs associated with the process, etc. When you finish, what you have is out of date, and it’s static. When a robot builds the book, it’s up to date, it’s inexpensive, it’s not a bottleneck for your time, and it has the benefit of software and machine learning that are always improving, so every time you use it or download it, it’s getting better.”

Of course, this technological feat doesn’t come without pitfalls. Computers can make mistakes just like humans, and, more interestingly, algorithms can hold biases (algorithms are written by humans, and it is easy for a programmer’s own biases to saturate computer code). “There’s a lot of work to be done in terms of the ethics around creating automated stories and, specifically, what’s the editorial oversight over the algorithms that are developed for automated storytelling,” said Marconi.

Given recent advances in automated storytelling, Melcher, Zimmerman, and Marconi all agree that while we are nowhere near a point where computers can write original novels, the industry is taking swift steps in that direction. One day, possibly in our lifetime, we will be reading stories crafted by computers, and asking ourselves: “Should we?”

This article was first published in the print edition of Brain World Magazine.

More From Brain World

Be the first to comment

Leave a Reply

Your email address will not be published.