Wobby and fact-checking with AI

The challenge: fact-checking during elections

According to research, Flemish citizens ask for fact-checks of politicians' statements in light of the upcoming elections in 2024. When it comes to the spread of fake news, it is not just politicians' words that need scrutiny — it is everything that circulates online.

Fake news is a problem that keeps spreading. And AI plays a dual part in this challenge. On the one hand, AI can contribute to the spread of disinformation with its ability to manipulate images or mimic audio. On the other hand, it can assist in swiftly sifting through data to verify statements and claims or to uncover manipulated images. When used appropriately, the technology can be really useful for the fact-checking team of VRT NWS.

Luc van Bakel, chief editor at the fact-checking team of VRT NWS: "We've already seen it happen abroad. Cloning votes, deep fakes, or memes, you name it. All of it aimed at influencing public opinion during elections. In Belgium, we haven't encountered it to that extent yet, but it's best to be as prepared as possible. We use AI to detect fake news. Think about translation tools or facial recognition tools like Google Lens. We also use tools that automatically search for fake news, allowing us to debunk these stories more quickly."

The solution: Wobby, an AI-data platform for journalists

To find solutions for fighting fake news, VRT Sandbox didn't have to look far within the organisation. Wobby.ai, an AI data platform, was launched in January 2023 by Amra Dorjbayar, who spent 7 years at VRT NWS as an investigative journalist, factchecker, and ultimately a data journalist.

Amra Dorjbayar, founder of Wobby.ai: "Wobby emerged from a very specific need within the newsroom. It is often difficult for journalists to quickly find the right data, especially when it comes to daily news and breaking news. The news cycle moves so quickly. By the time a journalist finds, analyses, and incorporates the right data into an article, the news has already moved on. That caused some frustration, so I started looking for a solution. Moreover, journalists are often presented with selective, pre-digested data, which means there is a high chance they miss crucial insights."

The idea behind Wobby is to make data journalism accessible to journalists who don't have, or don't need extensive technical skills. With the Wobby platform, a journalist can easily query massive, complex datasets. Wobby can then search the data, perform analyses, and automatically generate various types of charts. It is designed for journalists who may lack technical data skills but have the right critical mindset.

The case

To validate the Wobby platform within the VRT NWS production environment, two iterations were carried out between May 2023 and October 2023.

The first test of the Wobby platform was conducted with the investigative journalists from the Pano research team during in-person onboarding workshops. Afterwards, the journalists were given time to use the platform. Feedback from users indicated that the primary feature of the platform - uploading their datasets and using an AI chat functionality - was less essential. The biggest challenge for journalists is finding and quickly searching through relevant datasets.

With this feedback, Wobby pivoted to the current version, where journalists can use Wobby to instantly find and access official datasets.

The second iteration involved testing the updated Wobby platform with a real-life use case for VRT NWS. A PISA study by the Flemish government showed that reading skills among Flemish youth are on the decline. VRT NWS wanted to compare these findings with data from Flemish libraries to see if fewer books are being borrowed compared to previous years. Using Wobby.ai, they quickly found the relevant data and drew insights.

Ellen Debackere, VRT NWS journalist: "A key advantage of Wobby is that it provides AI-generated insights right from the start. This allows you to immediately see whether certain trends are emerging. As a result, we can quickly identify compelling journalistic stories. You can then use simple text instructions to explore the datasets further. In this case, we clearly saw that the number of borrowed books in Flemish libraries was indeed decreasing. However, to explain this behavior, you need to look beyond just Wobby."

The results

Now that the VRT Sandbox case has ended, Wobby will be added to the fact-checking tool arsenal used by the VRT NWS check team during the 2024 elections. Here, AI will only act as a copilot - a critical mindset is crucial during election times.

"When politicians make claims, it's vital for us not to assume they're true immediately, but to fact-check them," says Luc van Bakel. "Wobby is interesting to us because it compiles vast amounts of data and allows you to ask questions to the system. This lets you gain insights more quickly and makes large volumes of data more searchable. However, the ultimate responsibility still lies with the journalist. AI doesn't think, it just generates, and there are still regular errors. Think of ChatGPT's hallucinations, where the program cites fake articles as sources."

Amra Dorjbayar: "In developing Wobby, we apply some design principles to ensure human responsibility. When a journalist creates an article, the ultimate responsibility for publishing still lies with that journalist. Context and creativity remain crucial, and some aspects will always be left to humans."