By Niels Niemann
“Probably closer to 2030 to 2040.” This is what Elon Musk tweeted a few weeks ago, when the New Scientist announced on Twitter that artificial intelligence will beat people “at everything” by the year 2060. What the founder of Tesla and SpaceX did not know at that moment: The tweet to which he referred had been generated completely autonomously by an artificial intelligence.
The process of social media posts being written and published by an artificial intelligence is described by the key word automated content creation. 42% of marketeers already use it as stated in the Digital Trends 2018 study which was initiated by Adobe and conducted by Econsultancy. The reasons for letting an artificial intelligence write posts on social media are different, but there is one most marketeers agree on: 78% of marketeers expect automated content creation to simplify the maintenance and actualization of content, according to a study conducted by Forrester Consulting in the year 2016. Especially media houses and news sites, which publish several dozens of posts on social media per day, use tools for automated content creation.
Echobox. That is the name of the artificial intelligence which tricked Elon Musk on Twitter and encouraged him to answer to a tweet. It creates and posts content for major media companies like the Guardian and Le Monde and is able to understand, analyze and write texts. The man behind Echobox is an HSG alumnus called Antoine Amann, who formerly worked for the Financial Times. “Everything we build is designed to generate maximum traffic for publishers. We have different algorithms. One determines what content has to be posted at what time. Another can predict which topics and postings will most likely find viral distribution, and another can immediately share breaking news,” Amann stated in an interview with Forbes.
Elon Musk was not able to recognize a tweet posted by an artificial intelligence. But what about other readers? Can they differentiate between posts published by humans and posts published by an AI? In a non-representative experiment conducted with students, most of them were able to tell the difference. The posts written by an algorithm were mostly shorter and free of emotions. However, most of the students were still more or less impressed by the quality of the AI posts.
Even Amann himself is not completely convinced by automated content creation. “We advise publishers not to post too much automated,” Amann backpedals towards Forbes. “Linguistically, algorithms are not that good yet. While our goal is that someday it will be unrecognizable about what has been posted by a human and what has been posted by a machine, at the moment algorithms still make mistakes.” But do humans not also make mistakes? Mistakes are not really the problem of automated content creation.
“Creativity.” This was the most-frequently mentioned answer of students why AI will not replace humans in marketing. Amann himself is aware of that problem. “You’d have to teach a computer creativity, but no one has done it yet.” That is not completely correct. There were attempts to build a creative artificial intelligence. Attempts.
In 2016, Microsoft designed a Twitter bot called Tay which should emulate the behavior of people engaging with it. The experiment failed tragically. After less than 24 hours, Tay had adapted so much anti-feministic and anti-Semitic vocabulary, that Microsoft had to take it down. But was Tay at least creative? Not really. Most of her over 96’000 tweets were nearly copied one-to-one from other users.
The reason why so many companies are letting their posts being created automatic is not because they expect exceptionally well-written posts. Especially media companies have more than enough journalists who can write ten times better posts than an AI will ever be capable of. Interestingly enough, the before mentioned study from Econsultancy showed that only 2% of marketeers use artificial intelligence to create a better customer experience. What counts is not good text. No, the so-called “click-through rate” is the measurement of how good a post is.
The click-through rate describes the ratio between those who see content and those who click on it. “Publishers are dependent on advertising. And for advertising, the click-through rate is an important quantity. The higher it is, the better,” Amann states in an interview with Forbes. This is now the point where AI can flex its muscles. Through machine learning algorithms, the click-through rate of automated content is in most cases much higher than the average. Higher click-through rates mean higher advertisement revenues and these revenues bring home the bacon for journalists.
There is just one big problem. Services like Echobox are extremely dependent on social media platforms like Facebook and Twitter. Their algorithm decides which posts are seen by users and which get lost in the vastness of the Internet. For example, in June 2016, Facebook drastically decreased the visibility of media posts in favor of posts of friends. Appreciated by users, this modification of the algorithm was not positively received by media companies. Two months later, Facebook decided to reset this change, without justifying this decision.
Automated content creation is not really about good content. It is about influence and reach of media companies. What suffers is the creativity. Antoine Amann has an even worse vision of journalism in the future. “I believe in the future, with the mass of data we will have, a journalist no longer has to write an entire article alone. I think machines will write it and journalists will only check and correct it.” The peak of uncreative automated content creation.