Artificial Intelligence Will Turn Journalists Into "Centaurs," Not Replace Them
Many fear that artificial intelligence (AI) will one day replace huge swaths of the workforce. Last year The World Economic Forum predicted that AI and automation technologies will displace 5 million jobs by 2020, with healthcare, energy, and financial industries experiencing the greatest job losses. Some pundits predict that AI will also replace journalists. The theory is that as machines gain a more nuanced understanding of language, they will be able to produce the type of content that human journalists currently produce. The Washington Post, for example, uses an AI bot called Heliograph to produce news stories. Heliograph has covered the Olympics in Rio as well as the 2016 election season. Is it merely a matter of time before machines replace human content creators?
Stephen Masiclat, director of the New Media Management program at The S.I. Newhouse School of Public Communications, doesn’t think so. He envisions a future where artificial intelligence augments human intelligence, not replaces it. “Human intelligence can be amplified,” says Masiclat. “Machines can help journalists cut through the noise. That’s the promise of AI.”
Masiclat, who has spent decades studying the impact of AI on media, says that chess world champion Gary Kasparov hit on the promise of AI when he invented of a new form of chess that incorporates humans and machines. In 1997 Kasparov became the first world champion to lose a game of chess to a machine, losing to IBM’s Deep Blue Computer. A year after this experience Kasparov developed a new style of chess that he called Centaur Chess. In Centaur Chess, teams of people and AI compete to find the most inspired chess moves. Kasparov purposely called it “Centaur” and not “Cyborg” because computers don’t replace human chess players, but rather combine with human talents to create something entirely new (analogous to the half horse-half, half-human Centaur.)
The Limitations of AI
Although AI can produce news stories when fed the appropriate data, machines cannot ask the right questions to discern the truth of a story. Endgadget reported recently that voice assistant Google Home is feeding consumers fake news when they ask questions like “Is Barack Obama leading a coup d’état?” The AI understands the key components of the question, “Barack Obama” and “coup,” and it can discover patterns within Google search activity. Users who search for these terms frequently end up on sites like “Secrets of the Fed,” but Google Home cannot question the reliability of these sources. As a result, it is quick to inform users, “not only could Obama be in bed with the Communist Chinese, but Obama may in fact be planning a Communist coup d’etat at the end of his term in 2016!”
AI cannot understand the emotional elements of a story either. While Heliograph can report on the results of the women’s balance beam event at the Olympics, it cannot capture the exuberance of the crowd when a gymnast landed a particularly difficult dismount. It cannot make the reader feel the crushing disappointment of a gymnast who missed her chance at the bronze medal by a fraction of a point.
The ability to verify sources by asking the right question and bring empathy to a story remains the purview of human journalists, says Masiclat.
Centaur Journalism: How Humans Can Work With AI
Much of Masiclat’s research explores how computers can recognize words and understand connections between them. Masiclat is especially interested in how humans can refine word patterns that AI machines discover. For example, after a machine identifies a particular cluster of related words, humans can then view that cluster and improve it, relating words the machine did not and enhancing its understanding. “We’re using human input as an additional vector term [a numerical representation of information] to make the AI smarter, and that’s the thing that’s going to happen — all of us are going to participate in making AI smarter over time,” says Masiclat.
Editors at The New York Times are working to make AI smarter by semantically tagging people, places, and organizations in their articles. The goal is to help machines understand the relationships between articles that mention the same organization, place, or person. Eventually, the AI will be able to provide helpful services to editors, like identify quotes from a person across all of The New York Times’ articles. That will streamline editors’ ability to research and fact check a story.
With a refined understanding of words, machines should be able to work with humans to develop content that would maximize reader engagement. However, this still requires that journalists generate structured articles (or structured data) that machines can understand. AI can recognize different parts of a structured article and understand how those parts relate to one another. Coupling this understanding with first-party data about reader preferences, an AI engine could personalize the journalist’s article in countless ways and increase reader engagement, says Masiclat.
For example, journalists will write stories that machines remix and personalize based on reader behavior, spawning infinite versions of a single news item and delivering the most relevant versions to the right readers. If a technology professional is interested in an in-depth report on virtual reality, but doesn’t have time to read about it, a machine could provide that reader with a cliff notes version. AI could recognize the VR topics that interest the reader most and move those sections toward the top of the abridged version. Personalization has long been an aspiration for publishers; AI makes it possible for publishers to achieve that goal on massive scale.
One day, Masiclat hopes that AI machines can recognize different writing styles and reformat an article in the style a reader prefers. For example, one journalist might frequently begin articles with a historical precedent, while another might begin articles with a reference to a book. Eventually bots will be able to understand these two different styles and produce two articles from the original document, and deliver those article styles to the readers who will engage with them most.
Masiclat shared his expertise on artificial intelligence and machine learning at FUSE: The Convergence of Media & Technology. Watch Masiclat’s full presentation, which explores disruptive technologies like augmented reality, virtual reality, and blockchain, here.