By Stuart Roberts on September 28, 2016

Automated content: Can algorithms write your content for you?

"She was still dreaming. She had to be. . . . The plush four-poster brass bed and the veiled canopy were definitely out of some Arabian Nights dream. Maybe she was an exotic princess being made desperate love to by a handsome sheik. Carol willed the dream to continue and then her eyes absently focused on the gigantic mirror directly overhead. . . . " 

On May 1 1993, US company the Carol Publishing Group released a steamy, somewhat unremarkable novel about "women . . . men . . . fame, fortune and temptation" called Just This Once. The work was a passion project of Scott French, a devoted, some might say obsessed, fan of best-selling novelist, ‘Valley of the Dolls’ writer Jacqueline Susann. In total, French spent eight years and $40,000 on the book, an effort which his even own publisher said was "not a great literary work."

French had been published before, although he mostly focused on books about tech: The Big Brother Game in 1975, Spygame: Winning Through Super Technology in 1988 and High-Tech Harassment: How to Get Even with Anybody, Anytime in 1990.

Which begs the question: why was a surveillance and tech writer spending time and money on a two-bit potboiler filled with crude sex scenes and cumbersome romance?

The clue lies in the co-author credit on ‘Just This Once’s sleeve: a Macintosh IIcx computer called ‘Hal’. Hal’s artificial intelligence learnt and mimicked Susann’s style to perfection.

The book may have only sold 70,000 copies, but for a generation of computer programmers, Just This Once was a technical, if somewhat trashy, marvel.

In his quest, French started learning about AI by attending seminars where he rubbed shoulders with the likes of RAND Corp. and the CIA. They were interested in national surveillance and espionage, he was more concerned with anti-heroes and happily-ever-afters.

Macintosh iicx

Big mac: French's co-author for Just This Once was groundbreaking (Image credit: Wikimedia commons)

"I'd say [Hal] did almost 100% of the plot, 100% of the theme and style,” French said later when talking about his unusual writing partner. “Often, it came up with three adjectives in a row and I had to put a verb in there. But I didn't change its basic storyline or themes. I didn't feel I had the right to do that because I would have violated Miss Susann's style."

The tech was so rudimentary in the early 90s that the project became a costly labour of love. French spent years scanning portions of two Susann books, Valley of the Dolls and Once Is Not Enough, into Hal’s databanks, and deconstructing the writer’s style into 100 different parameters which Hal then turned into the final prose.

Since then, things have changed.

Artificial intelligence is in vogue, the technology is becoming more affordable, and in 2016, automated content is on the cusp of the mainstream. Creatives, marketing agencies, newspapers and more are watching on with one part interest, one part fear.

Some are doing more than watching on.

The now

At his recent Cannes Lions talk, WIRED editor Kevin Kelly explained: “the next 10,000 startups= Take X, add AI’.

We often think about the automation revolution in terms of process and logistics: of course a car can be automated with enough sensors; of course robots can serve food or wash windows. These are mundane jobs that humans don’t want—why not hand them to the robots? But creativity? That’s the preserve of artists.

Let’s take a look at Take Emily Howell’s work, for instance. Her beautiful classical piano compositions send shivers down your spine.


Emily is a bot: a computer program built by David Cope during the 1990s, and able to create stunning pieces that, to the untrained ear, sound humanlike. A purist would argue the emotion is lacking compared to, say, Leif Ove Andsnes or Alfred Brendel (thank you, Google), but it is technically stunning: the result of a computer learning to be creative. The tech will only improve.

So what of words? When Scott French’s Just This Once was released, Barbara Keenan, publisher of Affaire de Coeur, a monthly magazine that reviewed romance literature, said pointedly, "I don't think a book written by a computer will be popular."

Skip forward 23 years, and a robotic author so impressed judges in one of Japan’s most prestigious writing competitions, The Hoshi Shinichi Literary Award, that it become the first AI project to make it through the first round of judging. The news caused the LA Times to ask: Is the future award-winning novelist a writing robot?

Maybe, but there’s still a way to go. Science-fiction novelist Satoshi Hase, said of the project: "I was surprised at the work because it was a well-structured novel. But there are still some problems [to overcome] to win the prize, such as character descriptions.” The message? AI is good at some things, and not so great at others.

Still, out of 1,450 applicants for the prize, 11 involved AI programs. And you can expect that number to rise dramatically in the coming years.

So far, so experimental. Much like French’s novel, these cases are expensive research projects, completely out of the reach, and budget, of most marketers and content agencies.

But there are publications already publishing automated content created by algorithms on a daily basis. And to find out more, we have to turn our gaze away from the writing desk, and towards the newsdesk.

And it’s a story that we can trace back to the 19th century.

AI in the news

By the time of his death on March 16,1899, Canadian-born Joseph Medill had been a lawyer, helped found America’s Republican party, been a confident and friend to Abraham Lincoln, and overseen the office of Chicago mayor following the Great Chicago Fire of 1871. It’s testament to his talent as a newsman then that his New York Times obituary describes him first as ‘editor of The Chicago Tribune’.

For all of his remarkable achievements, it was as an editor that he made and maintained his name. Medill was a real coup for the Tribune—a Chicago-based press upstart: he had run a series of successful newspapers including The Leader, in Ohio.

One of the paper’s co-owners, J.D Webster, travelled to Cleveland especially to persuade him to join. As a former lawyer, Medill proclaimed he had “caught the smell of printer’s inoculation which possesses its victim until death.”

His last words were “What is the news?'"

The more pertinent question in the current journalistic climate might be ‘Where is the news?’. A recent Pew Research study entitled State of the News Media 2016 highlights a dramatic drop in newsroom staff; a workforce 20,000 positions smaller than it was 20 years prior. This change is no more apparent than at the publication Medill helped to create.

Joseph Medill Chicago Tribune

Stuck in the Medill: The newsman's legacy lives on through his school of journalism and his paper (Image: Public Domain)

His paper, The Chicago Tribune, was the founding company of the Tribune Publishing Group which, following a series of acquisitions, staff culls, bankruptcies and buyouts, relaunched in June 2016 as Tronc (a hideous portmanteau of ‘Tribune online content’).

Tronc’s now infamous employee video (now sadly, if sensibly, made private) has been much derided by pretty much all who have crossed its path. The video is stuffed with meaningless management babble, but its message is clear—content optimisation, machine learning and, tellingly, artificial intelligence are key areas of growth for a business which has increasingly culled news staff.

This isn’t an article about dwindling press reporters and newsroom staff, but this landscape goes to explain the motivations of newspapers and magazines: automated content is quick, cheap and largely effective. For an industry where costs need to be cut at every corner, AI represents an obvious answer to a serious problem. Get rid of the writers and bring in the bots.

And it also explains why so much resource is pouring into the research and development of increasingly sophisticated robo-writers.

The legacy of Medill

Joseph Medill was an editor and journalist in the purest sense, and he created a publishing legacy both familial—three of his grandchildren went on to run newspapers—and professional, through the Medill School of Journalism (MSJ). The MSJ was established at Northwestern University, Illinois; set up in his honour in 1921 to inspire generations to follow his passion of “telling the truth fearlessly”.

Only it’s not the Medill School of Journalism any more. Rather, the less pithy Medill School of Journalism, Media, Integrated Marketing Communications, where courses in news reporting stand on an equal footing with “classes in brand strategy, digital and interactive marketing and marketing analytics”, and professors teach journalism side-by-side with computer science.

This blurring of the boundaries is no more apparent than in Medill graduate and former professor, Kristian Hammond. Hammond wore two substantial hats at Northwestern—professor of electrical engineering and computer science, and director of Northwestern’s Center for Innovation in Technology, Media and Journalism—and he was instrumental in establishing the Northwestern Knight News Innovation Laboratory a.k.a The Knight Lab in 2011. The Knight Lab’s aim was to “bring journalists and computer scientists together to accelerate local media innovation.”

Fisk Hall Northwestern University

School of thought: The McCormick Journalism Center at the Medill School of Journalism, Northwestern University (image via Wikimedia Commons)

By the time The Knight Lab was established, Hammond, along with his fellow Medill grads Larry Birnbaum and Stuart Frankel, had already started on a research project (then known as StatsMonkey). The idea was to develop AI which could automatically generate news stories based on data points.

At the time, StatsMonkey focused around little league baseball so they fed data like batting averages, run rates and scores into the program, the AI did the rest. The results were too interesting, and effective, to be left in the lab, and this research project soon became a bonafide business venture.

The three partners called it Narrative Science, and the resulting product, Quill, now writes content and reports for over hundreds of businesses and publications worldwide.

If you’ve ever read Forbes’ earning reports, for instance, you’re reading Quill copy. The LA Times uses them to report on earthquakes in the area.

And Hammond’s job title now? Chief Scientist.

Narrative Science is not the only name in the game, however. Automated Insights—backed by the Associated Press (AP) and Samsung amongst others—developed Wordsmith, which operates in much the same way as Quill.


When Automated Insights was acquired by Vista Equity Partners for an estimated $80 million in February 2015, it was heavily focused on producing quarterly earnings reports for AP, but the plan was to expand to business intelligence, media, finance and marketing — as well as personal fitness and sports. Automated Insights now produces more than 3,000 articles a quarter for the likes of Microsoft, Yahoo and of course the Associated Press itself who made history when they hired Justin Myers as the world’s first ‘automation editor’.

On its site, it claims to have produced 1.5 billion ‘Naratives’ across 50 industries in the last 12 months making it the largest producer of personalised content in the world.

There’s a distinct parallel here between each of those platforms and publications and the way they’re utilising AI. Each is taking large chunks of data, and churning out content based on the figures. The content is functional but, much like Just This Once 20-odd years ago, unremarkable.

But that’s not to say that tech doesn’t have a creative hand in our writing at all. And in fact, the largest influence currently goes beyond the words, and into our daily editorial decisions.

Data-driven storytelling

Joseph Medill would have used old-fashioned news sense, gut feel and intuition to drive his news reportage. Editorial now, however, is as much about technology as it is human input, both in marketing and journalism. Real-time analytics allow sites to monitor the performance of individual articles, A/B test headlines, and much much more.

This isn’t without its critics, of course. The ‘giving the public what they want’ argument is as old as newspapers themselves: do we take the Steve Jobs stance that “people don’t know what they want until you show it to them” or do we take the stance that the public are telling us what they want with their clicks?

The slightly dull answer probably lies somewhere in between, but the fact is, we’ve never seen a time where analysis of consumer engagement is conducted at such a granular level. We know more about what readers read than ever before, but we also know how they interact with articles, what they look at next and what their responses mean. And all this knowledge, powered by technology, increasingly directs the creative.

Automated content

In an interview with Fast Company, Dao Nguyen, Buzzfeed’s publisher, said: "Having technology, data science, and being able to know how to manage, optimise and coordinate your publishing is the thing that gives you a competitive advantage."

Audience data already leads editorial in most of the biggest global publishers. If AI can analyse what people want to see, it’s not much of a stretch to imagine a bot making the requisite copy changes itself—tweaked headlines, snappier intro, improved keywords—based on audience reaction, too.

This scenario might seem like a way off, but bots are becoming better writers year on year. And studies have shown that readers often struggle to tell the difference between AI and human content.

How does it measure up?

Consider these two paras:

“Things looked bleak for the Angels when they trailed by two runs in the ninth inning, but Los Angeles recovered thanks to a key single from Vladimir Guerrero to pull out a 7-6 victory over the Boston Red Sox at Fenway Park on Sunday.”


“The University of Michigan baseball team used a four-run fifth inning to salvage the final game in its three-game weekend series with Iowa, winning 7-5 on Saturday afternoon (April 24) at the Wilpon Baseball Complex, home of historic Ray Fisher Stadium.”

You might have immediately guessed that the second excerpt was written by a bot, and you’d have been right. But don’t assume everyone thinks the same. The two paragraphs are part of a 2014 study by Swedish professor Christer Clerwall entitled "Enter the Robot Journalist" which pitted algorithmically generated articles against articles written by humans. In his findings, people largely couldn’t decide which was which.

That’s not to say people couldn’t detect differences: participants found the journalists’ work more pleasant to read and clear. On the flip side, they found the bot’s effort to be more trustworthy, accurate and objective. Short story, they preferred the journos work, but that wasn’t enough to convince them that it was human written.

More recently, The Guardian article compared a piece on Leicester City’s title-winning football season by their football correspondent, Jacob Steinberg with a piece by Wordsmith.

“Having jumped 13 places in a year, leaping from 14th to 1st, Leicester City are easily the most improved side in the league and Jamie Vardy’s role in their staggering rise cannot be overstated. The second top scorer in the league with 24 goals, Vardy has scored 35.29% of Leicester’s 68 goals. Only Harry Kane and Odion Ighalo were a bigger source of goals for their team, with Kane scoring 25 of third-placed Tottenham Hotspur’s 69 goals (36.23%) and Ighalo scoring 15 of 13th-placed Watford’s 40 goals (37.5%).


“That underlines Leicester’s overall effectiveness. Although they conceded as many goals as second-placed Arsenal, and one more than Tottenham, they have been more consistent. They were first at Christmas, while Arsenal were second and Tottenham were fourth. “It’s a magical season,” Claudio Ranieri, Leicester’s manager, says, justifiably so, given that a summer expenditure of £26.7m on transfers made them the eighth lowest spenders.”

And here’s the same story as reported by Wordsmith:

“It was a season for the ages for Leicester City as they lifted the Premier League Trophy and were crowned champions of England. Leicester City featured one of the league’s most skillful attacks, netting 68 goals. Jamie Vardy led the way with an incredible 24 goals. In addition to their offensive prowess, Leicester City possessed one of the strongest defenses in England.


“Shipping only 36 goals all season, their defense was able to frustrate even the most potent of attacks. Hoping to finish in the top ten after a 14th place finish last season, Leicester City splashed out 26.70 million in the summer transfer period. Leicester City sat in first place at Christmas after an incredible start to the season, and they continued to impress the second half of the season. After taking a few moments to reflect on the season, the Leicester City manager weighed in with, “It’s a magical season.”

Setting aside our own preconceived bias, it’s difficult to pick out too many signals that one piece is written by a robot and one by a professional sports journalist. AI is learning quickly.

Look at the (near) future

The key question for creatives is: will a bot steal my job? The answer is, unlikely. In the foreseeable future, at least. A study by Oxford University in 2013 analysed over 700 job roles and assessed the likelihood of the bots taking over, and writer/author was relatively low on the list—525th to be exact. We’re still a way off being able to give a computer a blank sheet of paper and telling them to come up with something.

As derided as it has been, the near-to-mid future of automated content is likely to follow the model laid out by Tronc. It’s not unreasonable to suppose that we will work side by side with robo-researchers and writers.

AI can research swathes of data far quicker than a human. They’ll compile relevant information and present it in relevant ways, allowing the writer to come in, analyse the automated bot report and add insight, context and flair to the piece.

And this is not so different to the way national papers currently work. Local press agencies source stories and send copy to the nationals. The copy is often rewritten by in-house staff to meet its own house style.

Automated content writing robotWrite-bot: Will algorithms ever master the nuance of language? (Image credit: Mirko Tobias Schaefer via Wikimedia Commons)

To use a musical analogy; DJs using vinyl had to beatmatch one record to mix into another. This process was time-consuming and imperfect. With digital decks, the two tunes are automatically beatmatched to the same tempo. Does this make DJing any less of a skill? A purist may argue yes. But what the digital assistance did was free up the DJ’s time to be creative with other interesting elements like effects, looping, phasing, filtering and whatever else.

The donkey work (matching tempo) is done by computer; the flourishes and art—the creative—is added by the human.

If you use Google Analytics, you can see this in action for free with Quill Engage. Quill uses Narrative Science tech to present dense analytics in a readable format, and the resultant reports provide context in an accessible way. It’s down to the marketing manager or editor to provide the insight.

Look at the (far) future

Narrative Science’s Chief Scientist, Kris Hammond, however, sees the future of automated content slightly differently. For one, he predicted that a robo-writer will win a Pulitzer prize in the next five years (although it’s worth noting he delivered this bold prediction to a Guardian journalist with something of a wry smile). More interesting, and less hubristic, are his views on personalisation.

“...we would rather use the technology to tell [a] story a million different ways that nails what is important about it to a particular group or even an individual. A natural disaster story might result, instead, in a report sent out to 10,000 companies showing how their supply chain is impacted, or, say, directly to people who the data suggests have relatives out there. That will not come very soon, but that is what news will become.”

The question Narrative Science is trying to answer is not one of creativity, but one of relevance to the audience.

In the very same interview, Kris Hammond estimates that content written by algorithm will make up 90% of journalistic reporting by 2030. This suggests that long-term, automated content is going to have to get vastly more interesting as well as more personalised, or our collective standards will have to drop considerably.

So how does a bot tell an interesting story? Well, AI relies on two things: smart algorithms and data points. In an increasingly connected world, the pot is ever growing. Thanks to the internet of things, data will be available from cars, CCTV, social media, the internet, live video, people’s homes and much more.

Add to this vast improvements in other technology like facial recognition is it’s not beyond the realms of possibility that an AI will be able to pull in all these different data sets and create a story from scratch with all the nuances of a seasoned writer. Just much more quickly.

This is all supposition, of course. But with the speed of technological advancement the realities could be much more incredible and impressive.

John Medill argued that independent press represent truth and progress to advance the intellectual, social, and moral welfare of the people, and nowhere is progress represented more than at the very institution that was created in his honour.

The imperative for scientists and technologists like Kris Hammond is to keep studying, keep improving on the technology, and keep pushing the boundaries of what’s possible. Where that takes us eventually is up for grabs, but it’s going to be an interesting read, whoever writes it.



[8929 _i="0" _address="" /][8929 _i="1" _address="" /][1 _i="2" _address="" /][8929 _i="3" _address="" /][8929 _i="4" _address="" /][8929 _i="5" _address="" /][2 _i="6" _address="" /][8929 _i="7" _address="" /][8929 _i="8" _address="" /][8929 _i="9" _address="" /][4 _i="10" _address="" /][8929 _i="11" _address="" /][CDATA 1="(function($)" 2="{window.fnames" 3="=" 4="new" 5="Array();" 6="window.ftypes" 7="=" 8="new" 9="Array();fnames%910" _i="12" _address="" /][0 _i="13" _address="" /][1 _i="14" _address="" /][1 _i="15" _address="" /][2 _i="16" _address="" /][2 _i="17" _address="" /][3 _i="18" _address="" /][3 _i="19" _address="" /][4 _i="20" _address="" /][4 _i="21" _address="" /]

Published by Stuart Roberts September 28, 2016