UMOD: Understanding and Monitoring Digital Wildfires

Fake news are common on the internet today, and in many cases they are quickly forgotten and have little effect. But sometimes, dangerous fake news can spread quickly and cause economic harm, destroy a person's reputation, or incite people to riots and violence. These incidents are called digital wildfires, and they can be a threat to society. Our goal is to understand these wildfires. We worked to find out why some people spread fake news. What motivates people that start them in the first place, and what kind of news might spread further and cause damages. Solving this requires that computer scientists and psychologists work together to better understand and prevent damage from such wildfires. In a study with several hundred participants, we found that people tend to vastly overestimate their ability to detect fake news. Messages that resonate with people's political beliefs can elicit strong emotional responses and are thus likely to be shared, no matter if they are true or false. Thus, malicious actors can exploit this effect by creating targeted messages for inciting people to action. The findings also indicate that education in politics and history, rather than cognitive abilities determines whether people can recognize fake news. We also study visual misinformation such as DeepFakes. We designed a model for the spread of news on the internet, which allows comparing the spread of true and false news articles over the internet, as well as misinformation on social media. We also developed methods based on artificial intelligence that can detect fake news and tested them on COVID-19 related misinformation.

The anticipated long-term impacts of the project are: - to encourage further development of technical misinformation detection - create long-term interdisciplinary collaboration - encourage the use of AI in computational social science - inform the public discussion on digital wildfires, misinformation, and its countermeasures. For more information, please see the results report.

In the recent years, digital wildfires, i.e. fast-spreading online misinformation have been identified as a considerable risk to developed societies, which raised the need for strategies to alleviate that risk. However, due to the speed with which online information spreads today, in combination with its immense volume, human monitoring of the Internet is completely infeasible, which gives rise to the need for an automated system. On the other hand, the requirements for such systems w.r.t. reliability, functionality, flexibility, and trustworthiness are immense. And while several approaches have been developed in the recent past, almost all of these attempts attack the problem purely from the technical side, generally using machine learning techniques. Our approach differs in that we study the problem from both sides, from the technical, but also form the human side by performing experiments and interviews aimed at understanding how people assess trustworthiness online, which content is likely to spread far, and why actors spread misinformation. On the technical side, the projects aims to design, implement, and deploy an IT system capable of analyzing large amounts of online news, focusing on Norwegian news sites and international sites that are cited frequently. The goal is to track where specific news items first appeared and how they spread, study the spread of misinformation, and, by using the knowledge gained from the experiments, enable the system to distinguish between misinformation and factual news. The overall objective is the prevention of digital wildfires via automated early warnings form the system, as well as enhanced preparedness for such events through intense study on the spread of such wildfires and the underlying reasons of the phenomenon.