Artificial intelligence and the media: renew or die

Over the next two to three years, computer vision, natural language algorithms and generative content, and deep learning, coupled with increased computing power of machines, large amounts of existing data, and more ubiquitous accessibility to New technological tools will allow journalists to make richer and deeper information, or to verify and edit data. Many of the trends that follow, from machine reading comprehension to predictive machine vision and computational photography, will give journalists superpowers, if they have the training to use these emerging systems and tools.


The technological revolution, in which the unstoppable development of artificial intelligence (AI) is framed, will not leave any sector unscathed, and the media industry is no exception. Overcoming the traditional inertia of resistance to change that is often shown by agents whose sectors are being dismantled and redefined, the communication industry must pay attention to the technological trends that are emerging today and that, properly used, can reinvent the journalistic profession as we know it today.

Carl Benedikt Frey and Michael Osborne, researchers at the University of Oxford, predicted that the combination of robotics, automation, AI and machine learning would also affect journalists, among others. According to a survey by Oxford University and Yale University, artificial intelligence will overtake humans in many activities in the next ten years; for example, translating languages ​​(in 2024), driving a truck (in 2027), writing a best-selling book (by 2049), and working as a surgeon (in 2053).

Of all the signals that journalism can receive from technological evolution, without a doubt, one of the strongest is that of artificial intelligence, which is more than just a technological trend, since it is behind many current innovations: from interfaces Vocals from phones to search engines suggesting personalized content, from smart factories to autonomous cars. The 2019 Tech Trends For Journalism and Media report ,  recently published by the Future Today Institute,  which identifies up to 75 technological trends that in the short term will affect the media sector, gives special prominence to artificial intelligence and highlights its application in different specific fields.

They know how to answer what is happening, but they still do not answer why it is happening

Artificial intelligence brings concepts such as “automatic journalism” and “augmented journalism” to the journalistic profession. There are already experiences that make machines capable of turning raw information into a narrative. We are talking about primitive “electronic journalists”, capable of telling a story, but who at the moment only know how to answer the question “what is happening?”  and that they still cannot answer “why it is happening”, leaving that competition, for now, reserved for the human reporter. There is talk of “increased journalism” when it refers to the increased analytical capacity that artificial intelligence provides to the journalist.

Subjectivity when preparing a story is another of the limitations that a journalistic algorithm can face, since information often has to be assessed in relation to a context or comparing it with previously known standards. For example, how much is much or how much is little? When the news of a public investment can speak of a large sum or, conversely, of very little money? When the number of victims of an event is very low or, instead, a real tragedy? The editor’s subjectivity is what establishes the evaluation of the facts and that capacity, today, is not possessed by intelligent machines.

The first journalistic robot
Although it is not known for sure who was the first, analysts point to Dreamwriter as the first journalistic robot in the world. He wrote a “flawless” 916-word financial article in just 60 seconds. It was designed more than five years ago by the Chinese video game company Tencent, also the creator of one of the most successful video games today, League of Legends (LOL). As reported in some days by local media, the automated news of Dreamwriter caused panic in the newsrooms of the country. The article, titled “August Consumer Price Index,” was written in Chinese and did not contain a single error.

Although Dreamwriter may have been the first journalistic robot, the seed of the current Heliograf (used by the Washington Post ), Syllabs ( Le Monde ), Quakebot ( Los Angeles Times ), Soccerbot (Yonhap News) and Quill ( Forbes ), the first algorithm Writing, called Tale-Spin, was developed in 1977 at Yale University. He used “knowledge about problem solving, physical space, interpersonal relationships, character traits, bodily needs, and story structure” to write texts. Since then, a number of algorithms have been developed to power journalistic robots.

The Associated Press (AP) is one of the media that started using artificial intelligence in January 2015 to automatically generate information. In 2018, they themselves noted, they automatically generated more than 3,000 stories about corporate profits for US companies each quarter, representing “an increase of ten times more information than reporters and editors were able to write. previously”.

Andreas Graefe, an expert in computer journalism, says how “once developed, algorithms can not only create thousands of news stories for a particular topic, but they also do it faster, cheaper and potentially with fewer errors than any human journalist. ” A statement with which one hundred percent agree Helen Vogt, head of the Innovation area of ​​the Norwegian News Agency, or John Micklethwait, editor-in-chief of Bloomberg News.

While Vogt explains that “machines don’t make the same mistake twice as long as there is a human to help train them,” Micklethwait says that “done correctly, automated journalism has the potential to make all of our jobs more interesting.” Even Vogt is able to assert that “most news that is based on numbers will be written by algorithms in the coming years. There is no reason for it to be written by humans. ”

Something that has already been happening in the French newspaper Le Monde since 2015, thanks to the use of an intelligent tool called Syllabs, which produced 150,000 web pages in four hours five years ago during the 2015 French elections. France has 36,000 municipalities, and Syllabs, with the right data could produce highly localized websites for even villages of just 35 inhabitants. Claude de Loupy, CEO and founder of Syllabs, believes that “robots cannot do what journalists do, but they can do amazing things.” And he considers that all this means “a revolution for the media.”

Heliograf and ‘Washington Post’
In addition to Le Monde in France, there are already a few media outlets in the United States that use robots to write news and free up journalists’ time, so that they are more dedicated to investigation, analysis and interpretation and thus add more value to the stories. The Washington Post , for example, began using its artificial intelligence technology, Heliograf, to write around 300 short reports and alerts at the Rio Olympics in 2016. Since then, the Post has used Heliograf to cover news from Congress, the governor elections in Washington state and the DC high school soccer games.

The Washington Post is trying to figure out how to use Heliograf to help its journalists. During the last presidential election, he used Heliograf to alert the newsroom when the results were starting to turn in an unexpected direction, giving reporters time to fully cover the news. They also see potential for reporters to spot trends in financial data sets, for example, or even to update weather information in real time.

‘Le Monde’, the ‘Washington Post’ and the BBC are clear examples of AI use

The Post accounts for the stories and page views generated by Heliograf, but quantifying its impact on the amount of time it gives reporters to do other work and the value of that work is more difficult. It’s also difficult to quantify the advertising revenue and subscriptions that can be attributed to those stories reported by your journalist robot. To do this, Heliograf today has around five people dedicated to it, not including the editors who work on the different content.

The BBC is another clear example of AI use; in this case, machine learning, to attend to what the public wants to see. Thus, the BBC’s R&D team has designed a five-year initiative to use artificial intelligence and determine what its audience wants to see. To achieve this, the team is partnering with data scientists and experts from UK universities, as well as media and technology companies based in Europe.

The Data Science Research Association intends to create “a more personal BBC” that can entertain in new ways. Researchers will analyze user data and apply algorithms to obtain information about audience preferences. The details are vague for now, but the team says they plan to use machine learning on their own traditional, digital-streaming content to gain new insights.

Also the Chinese state news agency Xinhua, in collaboration with the Alibaba group, is building an artificial intelligence platform for automated news production, combining multiple data sources. Known as Media Brain, it is an open platform where media agencies can share their data resources, cloud computing, internet of things and artificial intelligence in the news production process.

The applications proposed for Media Brain cover all stages of such production, from searching for potential clients to news gathering, editing, distribution and analysis of comments. The platform’s capabilities include video and image face verification. Additionally, it can be used to track copyright infringement of all forms of media.

In all of this context, a 2017 study by the Reuters Institute on the use of automated journalism in European news agencies found that, although most news agencies are already using or actively exploring automation in news generation, only two of the surveyed organizations use algorithms to compare new information with historical data and provide interpretations, adding analytical value.

The most popular areas for automation: finance and sports reporting

Automation is not yet widely used for more complex reporting. The most popular areas for automation are finance and sports reporting. Currently, there are also limitations in the ability to generate natural language.

In conclusion, artificial intelligence seems to be here to stay in the newsrooms of the media, supporting journalists in their work and not replacing them. A large proportion of news consumption already occurs through algorithm-based recommendations on social media platforms. Algorithms are now also being used in content generation. If used as an aid, it could improve the creation and improve the distribution of news. Claude de Loupy himself, from Syllabs, is quite descriptive in this regard: “We will not have a robot winning a Pulitzer, but we will have a journalist who wins a Pulitzer using robots.”

AI Applications
The Future Today Institute report lists several fields or applications of artificial intelligence that may impact journalistic activity:

  • Machine learning in real time. Among other things, it can guide new web readers to the content they are looking for and also adapt and personalize content on the fly to suit the tastes of a specific user.
  • Automatic reading comprehension. The perfection in the search for information will come when the machines “understand” what they read, refining the searches, and do not limit themselves as now to return results based on ordering tags or keywords.
  • Understanding of natural language. Unstructured text floods the network in the form of postson social networks and blogs, emails or texts on websites. This is all the text that does not have metadata incorporated to be indexed and mapped. Natural language comprehension tools help extract concepts from these documents and establish relationships, which for the journalist can be of great help when searching quickly and investigating large volumes of documentation.
  • Generation of natural language. We are talking about algorithms that write stories from raw information, something that is already used by media such as Bloomberg and the Associated Press. Algorithms that generate voice, sound or video. They are programs that learn to associate sounds from the physical world through videos: what it sounds like to step on dry leaves or to hit a sofa with your hand, for example. In this way, the machine is trained to recognize how objects in the physical world interact, in order to, in the future, be able to automatically and autonomously generate sound for TV series, programs and videos.
  • Complete images. Another function that artificial intelligence can perform is to complete an image. Once it has been taught or fed with huge amounts of images, the program is capable of completing information that does not appear in the photograph. For example, you can refine photos by adding elements that have fallen outside the frame of the original photo.
  • Predictive automatic vision. The Massachusetts Institute of Technology (MIT) is working on algorithms that learn to predict how humans who appear in video will behave. You can, for example, guess if in a series like Desperate Housewives (it’s one of the ones they use in this experience) the protagonists are going to hug, yell or slap each other in the next scene. In the future, they could predict news consumer behavior.

LeoRobotIA is the first journalistic robot capable of writing texts, and not filling out a template, in Spanish and with great linguistic variability. For this, part of structured data that converts information into thousandths of a second.

This initiative is led by four Spanish journalists, experts in technological information, and a spin-off company [derivative] from the Polytechnic University of Madrid, Dail Software, experts in artificial intelligence, machine learning, and natural language processing in Spanish.

Leo is already working on several projects to automate the previews, the chronicles and the live shows of football matches for both the Spanish league and the leagues in the United States, Argentina, Colombia, Mexico …, all of them Spanish-speaking, but also, even from other international leagues.

Also LeoRobotIA is developing a project to make real – time comparisons of mobile devices for intelligent assistants, and another to make readable for users companies bills utilities [utilities].

LeoRobotIA is convinced that “the use of artificial intelligence in news automation not only produces large volumes of information, but also facilitates the work of investigative journalists and other forms of data-based journalism.” Nor does he believe that artificial intelligence will replace the journalist in the publishing world. Instead, “what is likely is that AI can offload some of the repetitive tasks that don’t require creativity or high-level decision making. By taking on these routine tasks, AI can increase the journalist’s ability. ”

Other journalistic robots
Google has provided the British Press Association (PA) news agency with almost a million dollars to create software that will collect, automate and write about 30,000 local stories a month. Dubbed Radar (Reporters and Data and Robots), the software “will automate local reports with large public databases from government agencies or local agencies.”

The Soccerbot of Yonhap News in South Korea writes stories related to the English Premier League. In its first season, 2016-17, it produced a total of 380 automatic experimental items, each within one or two seconds after the end of each match.

The Associated Press and Thomson Reuters are using machine learning algorithms to write stories, and The New York Times plans to automate its comment moderation.

Earthquakes LA is a tool developed by the Los Angeles Times to automatically generate earthquake news with templates.

The Norwegian News Agency is working on an algorithm to produce stories of the 20,000 football matches that are played in the country each year. The news is based on data from the Norwegian Football Federation.


Leave a Comment