Researchers from Griffith University, the University of Queensland and Swinburne University of Technology have been studying political deepfakes in the lead-up to the Victorian election. Their preliminary findings, based on a meta-analysis of 19 published studies involving more than 24,000 people and an experiment with 411 Australians, reveal a troubling information landscape.
The meta-analysis found that deepfakes can appear just as persuasive as other political content, including authentic footage and conventional misinformation. They are also associated with lower trust in news and democratic institutions, potentially making people question both fake and real content. This uncertainty contributes to what researchers call the "liar's dividend," where genuine content is dismissed as inauthentic.
In the experiment, participants viewed either an authentic political video, a low-quality deepfake, or a high-quality deepfake. Approximately 75% of participants could correctly identify deepfakes, regardless of quality. However, one in four people were unable to do so. Notably, participants who consumed more news were better at detecting low-quality deepfakes but became less accurate at identifying authentic content, particularly with sophisticated deepfakes.
The Victorian election is still months away, yet deepfakes have reportedly been used in paid advertising on social media, including fabricated scenes of a machete attack at a petrol station and floodwater pouring down the steps of parliament. Previously described as a potential "sleeper issue" for Australian politics, deepfake political content is rapidly becoming a real electoral challenge.