Written by a Robot: Will Algorithms Kill Journalism?

Israeli startup Articoolo claims its AI engine can create original content on any topic in just minutes – and we won't know the difference.

From the left: Articoolo founders Nir Haloani, Lilia Demidov and Doron Tal.
From the left: Articoolo founders Nir Haloani, Lilia Demidov and Doron Tal. Moti Milrod

Writing content can be a headache for website managers. Producing it takes time, skilled manpower and money. In the current digital reality, brands, companies and small to midsize business owners must have a website in order to maintain contact with clients and attract new ones. In addition, they need to have an email distribution list and profiles on social networks.

There are currently over one billion content websites across the globe. Research shows that every day, four million blog entries are written and 215 billion commercial emails are sent out. Even though content designers and writers aren’t that high on the salary scale, writing original material poses a not insignificant financial burden. In larger companies that require an abundance of original content to maintain their presence in the digital world, expenses can pile up.

A solution to these problems is offered by Israeli startup Articoolo. It makes writing a text a simple, low-cost process by using artificial intelligence. Articoolo has devised a technological tool that can produce an original, high-quality 500-word article on any topic in two minutes. The reader will not be able to distinguish whether the text was written by human being or a machine. “With regards to artificial intelligence, producing a coherent article is complicated,” explains Doron Tal, one of the startup’s founders. “Our solution enables a reformulation of pieces of content that are already out there in digital space, presenting it in a uniform manner, in language that is clear and correct.”

Articoolo’s first intention was to use artificial intelligence in the world of education, helping university students write their assignments. The company quickly realized that the world of marketing was in even greater need of automated writing. “Over the last decade, advertising has undergone a sea of change. Customers now look for value and are less impacted by slogans. Companies and brands have thus shifted their emphasis and investment to content,” says Tal.

The startup was founded two years ago by Tal and his partners, mathematician Dr. Nir Haloani and software developer Lilia Demidov, both of whom have extensive experience in compressing data and internet advertising. Last year Articoolo was one of eight startups to be accepted into the prestigious IBM accelerator program. This gave its founders access to one of the most powerful artificial intelligence systems in the world: Watson.

“Within a month and a half we raised initial capital to the tune of a million dollars, and for a year we worked on the basic algorithm. Our vision was to build a machine based on artificial intelligence which could write an article that people would think a human had written. It was important to us that the text would look original, not copied and pasted,” says Tal.

“In any market segment, if the article is copied from somewhere and is not unique, it has no significance. You could use Google to get a text. We went through a long process of quality assurance, detecting problems and errors, and each time we improved the algorithm. This is an endless process, we’re always improving it and we’re about to take out a patent on the product,” he adds.

Produced in India

What do you think of the quality of the text the machine produces?

Tal: “Six months ago we launched a pilot and reached a situation where 100% of our articles were unique. We’re satisfied with the result. We haven’t encountered any sentence which is illogical. Quality is a subjective matter and we’re half-way to achieving our goals.”

How does the system handle fake news? What happens if the algorithm uses a fabricated article?

“By definition, the algorithm searches for sources which are article-like. We usually refrain from using content from news websites or from blogs that deal with current events as sources for an article. This reduces the chances of hitting a fake news piece. Obviously, even with articles as a source, we could come across incorrect ones, but this is the same as a human writer looking for sources and finding something for which he can’t trace a source. Our chances of tripping up like that are lower since we have categorized different sources, reducing our chances of using unreliable ones.”

How many users does Articoolo have?

“We have 33 thousand users who have generated at least one article, 2,000 of whom are paying customers. Seven percent of people who register become paying clients. 60 percent of paying customers are also subscribers, or returning customers.

“Strategically, our objective is to approach big companies and organizations that require original content frequently and in large quantities. We have some pilot projects with companies dealing in native advertising and content, and we’re examining the capability of our tool to handle large capacities. Our prices depend on whether companies buy a subscription or wish to purchase a limited number of articles. Ten articles cost $19 and 100 cost $99. On the other hand, subscribers can generate 30 articles for $29 or 250 for $99.”

Who are your competitors? Are there other companies that generate content through artificial intelligence?

“We did some research and looked at other solutions in this area. None of them provided a complete solution as ours does. For example, there are software products called spinners which can take text or one word and replace them with synonyms or similar text. The problem is that they don’t analyze the text, which can lead to text that's incoherent. One needs the original article which the spin process uses. A startup called Narrative Science, for instance, writes abstracts but can also produce text, but this is based on fixed patterns. We don’t produce technical texts but provide products to companies that need original content in large amounts and at high frequency.

“We try to direct the system to produce texts in the format of articles, so that our main competitors are actually human writers. Research shows that 70 percent of internet content in English is written in India. There are multiple websites for writing articles, managed by people in India at relatively high cost. During our pilot we realized that there is high demand and that people are willing to pay. In our next stage we’ll be able to raise our prices since we hardly have any competition. When the main alternative is human writers, the price of the final product becomes significant.”

Written by an algorithm

Are you a threat to content writers? Do journalists have to start looking for a new career?

“We’re not going to replace journalists. This is an aid to many kinds of users. There still are journalists who go over the content, edit it and in our case translate it as well. We’ve concluded that journalists and professional writers prefer not to let an algorithm write instead of them. They are daunted by such an algorithm. On the other hand, they’d like some help in writing, especially during early stages in which they sit in front of a blank page and need some starting point. This algorithm doesn’t make journalists redundant. On the contrary, it underscores the importance of human writers, since the algorithm will never match a journalist, with his or her discernment and insight. It can make life easier for journalists without completely replacing them. It will take a long time for a machine to completely replace a human writer.”

Nevertheless, your product raises questions regarding the work of writers or journalists: Should they publish text written by a machine rather than by them?

“I don’t regard a journalist using Articoolo and its technological tools in order to get ideas or to formulate them as unethical. The whole world is moving towards artificial intelligence and there’s no escaping it.”

In recent months, several content websites have approached the company requesting to try its tools, and several partnerships have been forged, some of them on a permanent basis. One example is BITA, a Japanese technology blog which uses Articoolo’s algorithm to write articles relating mainly to technology news and surveys. This collaboration increased the company’s exposure in Japan and Japanese clients now make up 15 percent of all of their customers.

Articoolo’s tool raises many questions about automatically generated content with no human intervention. Other than threatening the jobs of content writers and journalists (although Tal denies this), content generated by a machine still requires the customer to check and verify the content in certain cases to ensure that no false information is published. Preparing the text then becomes a longer process than the customer intended. However, Articoolo strives to produce simple text, not academic treatises or news items. This is why the company is managing to meet the high market demand for quickly producing large quantities of relevant content which is coherent and relatively accurate.

Without a human touch

The text below, entitled “The Financial Implications of Climate Change,” was completely written by Articoolo’s artificial intelligence system. The system was given five topical keywords and limited to 250 words.

The algorithm analyzed every word and scanned the internet to choose the main source to write the article. The software subsequently located further related content and composed key sentences for the article using those sources. Finally, the raw text was rewritten and edited using the natural language processing engine in order to ascertain that it was legible and comprehensible. The whole process took no more than two minutes.

Text generated by Articoolo's AI