In June 2020, MSN News (a news distribution service operated by Microsoft) announced the layoffs of 77 contract employees responsible for collecting, selecting, and editing articles. The reason, astonishingly, was that AI-equipped computers would take over these tasks in place of the human journalists. AI has become capable of performing such work in journalism.

MSN News homepage (Photo: Yow Shuning)

In contemporary society, AI is appearing in more and more situations. It is used to display recommendations when searching in online shopping or subscription services, in autonomous driving technologies in cars, for diagnoses in medical settings, and by the police to predict locations and times where crime is likely to occur—AI is utilized across a wide range of fields. So, in today’s increasingly widespread use of AI within our information-driven, global society, to what extent will AI come to replace journalism, which is entrusted with conveying the events and phenomena occurring around the world every day? Will it remain merely an auxiliary tool that supports journalism conducted by humans? With international reporting as a backdrop, I would like to analyze its possibilities and potential impacts based on current data and the initiatives currently underway.

What is AI?

What does AI actually mean? AI is an abbreviation for Artificial Intelligence, called jinkō chinō in Japanese. This artificial intelligence attempts to reproduce, using computer programs, part of the activities in which humans think and then act. So what specific functions does AI have? It is considered to have five main ones.

Test drive of a vehicle equipped with AI-based autonomous driving technology (Photo: Dllu / Wikimedia [CC BY-SA 4.0])

The first is the learning function. At the most basic level, AI learns by repeating trial and error. As it repeats failures, when a successful outcome occurs, it stores the circumstances and then applies that memory the next time a similar situation arises. For example, this is used in predictive text on keyboards. When typing on a computer, automatic conversion selects among multiple characters or words, and the system learns those that are frequently entered and those that fit the context and becomes able to convert accordingly.

The second is logical reasoning ability. This function imitates the way humans think when trying to solve problems, deriving results through inference. AI learns universal, general premises and applies them to individual tasks. For instance, analyzing viewing histories on video streaming services: if a viewer watches multiple movies in the same genre, AI can estimate that they like that genre and conclude it should recommend similar titles.

The third is problem-solving ability. This is the ability to present solutions to reach predetermined goals. For example, in medical settings, in place of a physician, AI using deep learning (Note 1) can identify mutations in DNA that cause disease and predict the onset of illness.

The fourth is recognition ability—the ability to recognize and distinguish real-world objects and artifacts. Although there are many challenges because objects can look different depending on angle and lighting, this is used in visual sensors for self-driving cars and in robotic vacuum cleaners.

The fifth is language ability. This goes beyond simply understanding grammar; it is the capability to understand and respond to language as used in actual writing and conversation. AI assistants that, through voice recognition, search for information or play audio are examples.

By using these functions, AI has been assisting human activities. So how can such technologies be applied in reporting?

The current state of AI utilization opportunities in reporting

AI is already being introduced by news organizations in various situations. The first is information gathering. Mainly from social media, other news articles, or databases published by international organizations, governments, and research institutions, vast amounts of information are automatically collected and analyzed by web crawlers (Note 2). This allows the discovery of global phenomena, events, and trends and can help generate story ideas. It also enables the periodic collection and updating of new data in fields that have long been of concern.

Robots as journalists? Showcasing progress of AI in reporting (Photo: Deutsche Welle/Flickr [CC BY-NC 2.0])

The second is involvement in news production. At the most basic level, AI can detect and correct typos and grammatical errors, but it is increasingly used for content as well. For example, tools are being developed to fact-check the substance of politicians’ statements. It can also write simple news articles by following templates or by learning from the writing styles of past articles. For instance, when an earthquake occurs in a given country, it can instantly and automatically create an article based on fixed information such as the epicenter, intensity, and magnitude. Furthermore, it can automatically update articles that have already been published—for example, automatically refreshing election-result stories with new data as votes are counted. It can also select photos and videos that fit the article and perform editing tailored to the page or medium.

The third is news distribution. By learning the tendencies of individual readers and viewers, it distributes or recommends articles accordingly. It also matches sponsors with reader and viewer tendencies to display ads suited to the audience.

So how far has AI adoption actually progressed? A survey conducted in 2019 of 71 news organizations across 32 different countries found that 37% reported having an AI implementation strategy, suggesting further progress in tasks where AI is expected to permeate newsrooms.

A Russian newsroom (Photo: Jürg Vollmer/Wikimedia [CC BY 3.0])

Benefits of introducing AI in reporting

Let’s consider the benefits for both the producers and the audience when AI is introduced into news organizations.

First, regarding the benefits on the producer side when AI is introduced: an obvious one is reduced working time. Having AI handle information gathering and text editing may sometimes replace journalists’ work. However, journalists whose time is freed up can secure more time for on-the-ground investigations such as fieldwork and street interviews, enabling them to write more differentiated articles based on information obtainable only in the field.

Another potential benefit is the ability to expand the scope of information gathering. For example, it becomes possible to access social media, other media outlets, and databases around the world. Languages that previously posed translation difficulties become far less of a barrier thanks to automatic translation, allowing information to be gathered regardless of language. As a result, it will become possible to report on countries that usually receive little attention.

Furthermore, in terms of analysis, it can discover connections between events, phenomena, and fields that seem unrelated at first glance by comparing big data. Comparing big data refers to work that compares issues spanning multiple fields—such as the economy, health care, and environmental problems—across multiple countries. This can lead to finding story ideas and to gaining understanding of the context and side effects of issues.

A street interview by a human journalist (Photo: Ezarate/ Wikimedia [CC BY-SA 4.0])

Let’s take reporting on the mass demonstrations and social movements known as the “Arab Spring” as a case study of these AI benefits. This series of events escalated from small incidents to the point of toppling multiple governments. If AI had been introduced into reporting on this chain of events, what would have been possible? First, if AI were used to constantly monitor information on social media and in media outlets around the world, it could detect a surge in words related to “demonstrations” and quickly sense anomalies on the ground. Analysis of social media and other sources could identify which countries, cities, and regions were seeing demonstrations spread, and how quickly. Through automatic translation and analysis of the content of messages, it could also understand in detail what participants were demanding and what actions they were taking. Furthermore, by comparing this with databases that publish information on prices, unemployment rates, levels of inequality, oil prices, and more, it would be possible to correctly understand and report the background and phenomena of issues spanning multiple countries.

AI could also be used when whistleblowers leak massive amounts of data to news organizations. Examples include the “Panama Papers” and “Paradise Papers,” which were confidential documents on financial transactions using tax havens, as well as the WikiLeaks organization and the “U.S. diplomatic cables WikiLeaks leak,” in which some 250,000 U.S. diplomatic documents were leaked to WikiLeaks and other media outlets. In such cases, AI can be used to analyze who is connected to whom and to what events within the data, in conjunction with other relevant databases. This can lead to many discoveries and new stories.

While the above are ways traditional news organizations such as newspapers and TV stations can utilize AI, new media leveraging AI technology are also emerging. One example is a website called Liveuamap (Note 3). This site automatically collects and analyzes information from social media and local media to detect where and what is happening in the world, including which armed groups control which areas in armed conflicts. Instead of writing articles like traditional news organizations, it is a new medium that represents the situation on a map. The site was developed not only with AI but through the cooperation of software developers and specialist journalists. These maps, which feature the latest conflict situations made possible by AI, are used by major news organizations as reliable data.

The situation in Libya as seen on Liveuamap (screenshot as of June 11, 2020)

Now the benefits for readers and viewers. First, if news organizations use AI functions such as information gathering and translation, readers and viewers can obtain information from a broader range of locations around the world. This helps in understanding the current flow of world events. Second, thanks to AI’s learning functions, it can infer what news readers and viewers want and deliver information tailored to their preferences. In other words, readers and viewers can more easily obtain information related to their interests and save the effort of searching for articles, enabling deeper knowledge of their areas of interest.

Drawbacks of introducing AI in reporting

Next, let’s consider the drawbacks that may arise when AI is introduced into news organizations.

On the producer side, technical and financial issues are easy to imagine. Technically, there is a burden associated with implementing and managing AI. Introducing AI requires large upfront investment, and whether to hire experts to handle AI technologies or train them in-house, either path imposes costs. Also, introducing AI may reduce personnel costs—or eliminate jobs—leading to the possibility of journalists being laid off, as in the MSN News example at the beginning.

Drawbacks may also arise for readers. AI’s learning functions make it easier to display articles related to readers’ and viewers’ interests based on their browsing history. While this is convenient, as noted above, it also has a dangerous side: by limiting content to readers’ and viewers’ interests, they may lose diverse perspectives. Because content of low interest is less likely to be displayed, a reader interested in sports, for instance, will have fewer chances to encounter or access pieces on international issues. Conversely, only those with an interest in international issues will end up viewing such content. As globalization rapidly advances and problems increasingly intertwine the global and the local, there is a risk that international issues will be less widely recognized. The result is a narrower view of the world.

There is also the danger that AI’s editing capabilities could generate fake news. With AI-driven deepfake (Note 4) technology, videos and images can be created that make it appear as if politicians and celebrities actually said something, producing false news that even ordinary readers and viewers have difficulty discerning. As a result, more news organizations could end up disseminating misinformation. Although AI technologies to distinguish fake news are also being used, once fake news is distributed, it affects the credibility of news organizations.

Data management with AI (Photo: Medialab Katowice/Flickr [CC BY 2.0])

Furthermore, because news organizations collect and use information about readers’ and viewers’ online behavior, ethical issues such as privacy arise. One example is the Cambridge Analytica scandal. Cambridge Analytica was a consulting firm known for using big data to help its political clients—politicians and parties—gain advantages in elections. By using large amounts of information such as browsing histories and behavior obtained from Facebook and analyzing individuals’ political views and personalities with AI, it profiled individuals and then delivered pinpointed political ads tailored to those characteristics into their Facebook news feeds. In doing so, it allegedly manipulated opinion to raise a candidate’s image and gain electoral advantage, with a particular focus on capturing votes from unaffiliated swing voters. Cambridge Analytica is said to have conducted the above activities in elections in countries such as the United States, the United Kingdom, Brazil, Kenya, and Ukraine among others. Thus, there is also the danger of information manipulation using AI.

Limits of AI use in international reporting

At present, AI is useful for lightening journalists’ workload in areas such as collecting information from around the world, identifying relationships among information, automatically detecting trends, editing, and distributing news. However, some reports suggest that it can still only replace 15% of a reporter’s work and 9% of editing work.

In what situations does AI assistance not work? First, determining what is newsworthy—judging, with a comprehensive understanding of society and the world, what to take up, from what angle, and in what depth—requires humans who can consider the social context at the time. Second, skillfully eliciting information from news actors (perpetrators, victims, decision-makers, etc.) and judging whether it is true. The ability to consider not only the surface meaning of words but also the meaning hidden behind them is uniquely human. Third, understanding the context and significance of events and phenomena—not only historical, cultural, and political contexts, but also the personalities and emotions of the people involved. Humans can respond to a variety of information and situations without relying solely on data. Fourth, creating and delivering expressions and presentations that foster reader and viewer understanding and leave an impression. It is people who can convey events and phenomena in ways that resonate emotionally with audiences.

A data journalism training session (Photo: Deutsche Welle/Flickr [CC BY-NC 2.0])

Let’s look again at the “Arab Spring” example mentioned above. In such large-scale anti-government demonstrations, using AI could make it possible to grasp the occurrence and spread of protests and to perform a certain degree of analysis and prediction by combining data from various other fields. However, deciding where to start covering as news, how each country’s government perceives those facts and plans to react, and what emotions the people involved have—these are aspects that only humans can judge and analyze. It is also difficult for AI to determine how to communicate such events to the world in ways that move people’s hearts.

Conclusion

AI is developing ever further, and its role in international reporting will likely grow in the future. We may also see situations where it changes the form of journalism. However, viewed comprehensively, AI will not replace journalists but rather become a force that augments them. Through the coexistence of efficient AI-driven tasks and the knowledge gained through journalists’ activities, the quantity and quality of the information we can obtain about the world may increase.

Note 1: Deep learning. A type of machine learning that trains computers to learn tasks that humans perform naturally.

Note 2: Also called bots, spiders, or robots. Application software that automatically collects and stores text, images, videos, etc. published on the Internet.

Note 3: An abbreviation for Live Universal Awareness Map; in Japanese it means “world awareness map.”

Note 4: A technology for synthesizing images of people using AI.

Writer: Kaito Seo

Follow @GNV_tweets