In its December 2019 issue, The Economist published an interview with a piece of artificial intelligence (AI), called GPT-2. GPT-2 was trained to generate text and in the interview it answered questions about the state of the world in 2020. On the question ‘What is the future of AI?’ the bot replied: “It would be good if we used the technology more responsibly. In other words, we should treat it like a utility, like a tool. We should put as much effort into developing the technology as necessary, rather than worrying that it’s going to harm us and destroy our lives.”
GPT-2 does not really understand what it says, and although its answers in the interview are a bit vague and general, its performance is still pretty impressive. Earlier this year, OpenAI, the developer of GPT-2 already introduced its successor GPT-3, which is a lot more powerful. Automatic text generation is one of many examples that show the potential power of AI used by the media.
In June 2019 the European Science-Media Hub (ESMH) already published an article on AI in journalism, following a three-day summer school for young journalists and students in Strasbourg on that topic. What have been the most important developments and discussions since then?
“The most important development in the last year has been the growing adaptation and awareness of AI in newsrooms worldwide”, says Nicholas Diakopoulos, assistant professor in communication studies and computer science at Northwestern University (USA) “The type of jobs in journalism will evolve. Some jobs will look less like traditional reporting jobs and will involve more IT-skills.” – Read Nicholas Diakopoulos’ full interview
Diakopoulos gives some examples of recently developed journalistic AI-tools: The New York Times launched an R&D lab to experiment with AI in journalism. The Washington Post began experimenting with news discovery tools, partly in cooperation with Diakopoulos. On the academic side conferences like Machines + Media and the Computation + Journalism symposium have been established
Before the corona crisis broke out and the Olympic Games of 2020 were cancelled, The New York Times was experimenting with computer vision, also an AI-technology, to construct 3D-scenes of sports events. The idea is to use AI to augment the experience of tv-viewers in real-time, for example by providing insights into the live performance of athletes.
Together with a team of The Washington Post Diakopoulos developed a news discovery tool for the US 2020 presidential elections. Their goal was to use a data set of a few hundred million registered voters across the US to try and find interesting locations where a political journalist might go to in order to discover a demographic trend in the electorate, for example because there turns out to be a spike in the new registration of Hispanic voters in Texas.
Diakopoulos and his research lab are also developing a tool that is meant to help investigative journalists to dig into the use of algorithms by the US government. The tool can be set up to automatically give an alert every week: For example: “Last week I found eight new algorithms that I think matches your interest.” At the moment nine journalists are testing the tool.
Impact on society
In the fall of 2019 the London School of Economics published a global survey of journalism and AI. The LSE-report was based on interviews with 71 newsrooms spread over 32 countries. One of its conclusions was that AI will reshape journalism in an incremental way, but with structural effects in the long term. The report also notes that the use of journalistic AI-tools raises questions about the implications for society. What are the effects on democracy, on diversity of the journalistic reporting and on public values in the reporting?
Professor Natali Helberger of the University of Amsterdam (UvA) studies exactly these questions. She stresses the fact that historically speaking journalism and technology have always had a close relation: photography, telephone, radio, tv, computer, internet, smartphone — they all drove changes in journalism. “The introduction of each technology came along first with a major hype, then with a wave of concern and even dystopia, and finally with a constructive phase in which the technology was used for the benefit of journalism.” – Read Natali Helberger’s full interview
According to Helberger it is wrong to denigrate a particular technology per se. Instead, she says, we should investigate how we can use the technology to support a democratic society. “AI does have transformative power for journalism. It provides new ways of engaging with the audience and creates possibilities for people to find information more efficiently and get better informed. But with that power also come responsibilities, for example to protect fundamental rights and freedom.”
Because of these responsibilities, Helberger is worried about the lack of structural independent funding for R&D in journalism. “A lot of tech innovation in journalism is funded by the Google News Initiative. It’s really cool that they do it. But Google is a company, right? Media play a key role in our democracy and they should always be independent.”
Her own research focus is on the use of automatic news recommendation and which consequences they have on the diversity of the reporting. Helberger: “Diversity in ideas, opinions, cultures, ethnicities and religions is important in a democracy because it teaches us tolerance. Especially in our present polarised time the media have to make sure that they are inclusive, can serve everybody and not just a particular group.”
Together with the German public broadcaster ZDF, she and her university colleagues are now testing a diversity toolkit that they developed, among others in cooperation with data scientists from the commercial broadcaster RTL in the Netherlands. This is a generic, stand-alone toolkit that helps media professionals to understand and assess the diversity of their algorithmic recommendations. Depending on these insights and whether the news outlet wants for example more engaging content, more political content or more minority voices, the wheel of the recommendation tool can be turned.
“Every AI-tool in the media has to be optimised for the particular news outlet”, says Professor Natali Helberger, “because with a generic tool you have very little to say about the values which you as a journalistic medium find important. A machine cannot decide about which values are important. That’s fundamentally a human decision.”
• LSE-report ‘New powers, new responsibilities. A global survey of journalism and artificial intelligence’ (2019)
• LSE introductory course on Machine Learning for journalists
• LSE ‘Journalism AI Collab’ in order to develop prototype AI-powered solutions that can act as a model for other news creators
• ‘Automating the news’ by Nick Diakopoulos
• In May 2020 Natali Helberger received the ‘Digital Journalism article of the year award’ for her article “On the democratic role of news recommenders”
• In 2015 Natali Helberger received a 1,5 million euro ERC-grant for research entitled: “Profiling and Targeting News Readers: Implications for the Democratic Role of the Digital Media, User Rights and Public Information Policy”