Impressive start
La Nación’s experiments with AI began with an investigation into renewable energy in Argentina. In 2016, then-Argentine President Mauricio Macri launched a program to develop the country’s clean energy sources, relying largely on socialized investment and international cooperation.
Inspired by an initiative to map every solar panel in the US, Florencia Coelho, a leading communications specialist at La Nación, proposed a project to map the progress of the program four years after it was launched.
La Nación’s data team began the project in collaboration with Mathias Felipe, a guest partner from the University of Navarra in Spain. The team used machine learning (ML) and computer vision, and worked with a third-party lab specializing in geospatial analytics and AI.
La Nación newsroom is pioneering the learning and application of AI in news production. Photo: La Nación
La Nación’s algorithm was trained to identify the shapes of solar farms in Argentina. Computer vision trains computers to analyze and understand images. 10,999 images were used to train the algorithm before a total of 7 million images were processed and 2,780,400 square kilometers of land were analyzed. The result is this data-rich series that shows how Argentina’s clean energy program at the time was falling short of its stated goals.
Be patient and cooperative
In fact, La Nación’s AI project faced many challenges. Access to satellite imagery was expensive. Solar farms looked a lot like agricultural farms. They didn’t have enough images of solar farms in Argentina in 2019 to train the AI model, so they had to get images from Chile.
Coelho shared another challenge and how to solve it: “La Nación’s infrastructure and editorial skills were not enough to carry out this project. We did not have the hardware or computing power needed for the project, so that’s why we collaborated.”
From this initial collaboration, La Nación’s data team learned the benefits of collaboration. They also learned that if they didn’t understand AI technology enough, they could end up targeting the wrong people. So they set up an AI lab made up of journalists, data analysts, and more to help La Nación accelerate its adoption of the technology.
The lab’s first project was to analyze the lyrics of Trap music, a genre of Hip Hop that originated in the American South in the late 1990s. It took them seven months to complete. The team used machine learning, natural language processing (NLP) models, Spotify’s API, and more to process 692 songs and learn about the themes, trends, and messages of this increasingly popular genre in Argentina.
La Nación used a variety of AI technologies to analyze satellite imagery for a project to map solar farms in Argentina. Photo: La Nación
However, the AI the journalists used had to deal with a number of linguistic issues, including new words appearing in trap songs, as well as many other characteristics of the genre. La Nación’s tests also showed that the AI was almost exclusively built for the English language. “Every natural language processing model is designed for the English language,” Bouret said. “It was very difficult for us to find libraries and routines that would help us solve the Spanish problem.”
Another big challenge for newsrooms looking to implement AI projects is time. “There are projects that take five to seven months — they’re long-term projects. Newsrooms are very difficult to understand because they’re always in a hurry. You have to be patient,” said Bouret, adding: “Investigative journalists can spend a year investigating corruption or an event. We’re the same, we’re investigative journalists about technology.”
Therefore, this expert said that collaboration, whether with third-party AI experts, universities or scientists , will help newsrooms speed up the process and reduce costs in embracing new technology.
“We all have to learn”
Collaboration between news organizations can also help advance AI adoption and uncover more resources, La Nación’s experience shows. La Nación’s AI team has been working on a gender gap tracker hosted by the London School of Economics ’ JournalismAI Collab.
They’ve applied it to specific projects at La Nación, like how to better understand gender bias in journalism. This work will help the newsroom’s business department assess how the performance of articles is affected by gender, such as how articles written by men or women affect readers, as well as other issues.
As an extension of its gender-tracking project, La Nación also participated in an open-source AI project that brings together news organizations around the world to work on detecting gender in faces. By sharing about 50 portraits of Argentine and Latin American people with the training team, they helped the AI model detect more diverse faces in terms of skin color and ethnicity.
That said, whether it’s between tech companies or between newsrooms, collaboration on AI projects is vital for news organizations. “AI skills are hard to learn, so it’s better to learn from each other, even from rival publications,” Coelho advises.
Finally, the expert urged: “We are actually competing with Google and Facebook to win back the attention of readers. It is unfortunate that it took us a decade to understand this. Therefore, our journalism needs to accelerate the process of learning and sharing thoroughly, and at the same time working together, even across countries. We will all have to learn, because AI is too big for one person.”
Bui Huy
Source
Comment (0)