Can Deepfakes Influence Elections? MIT Researchers Reveal Surprising Results

Since the emergence of deepfakes, many have expected that these technologies would eventually be used to influence elections. However, a new study by researchers at the Massachusetts Institute of Technology (MIT) suggests that this may be more complicated than it seems.

Study Overview

The researchers set out to determine whether political videos are more persuasive than their text-based counterparts. The answer turned out to be quite unexpected: they are not.

Specifically, two studies were conducted with about 7,600 participants from across the United States. In both studies, participants were divided into three groups. The first group watched randomly selected “politically persuasive” campaign videos or popular political videos on COVID-19 from YouTube. The second group did not watch the videos but read their text transcripts. The third group served as a control group.

Key Findings

Afterward, participants in each group filled out questionnaires evaluating the credibility of the message they had watched or read. They were asked whether they believed the person in the video or text actually said what was presented. Participants also rated how much they disagreed with the main viewpoint expressed in the message.

The researchers wanted to answer a twofold question: does seeing really mean believing, and if so, how much can video or text influence someone’s opinion?

It turned out that people are indeed more likely to believe what they see with their own eyes rather than what they read. In other words, seeing does mean believing. However, when it came to the persuasiveness of the message, there was virtually no difference between video and text.

“The fact that video is more believable does not mean it is more likely to change someone’s mind,” summarized Adam Berinsky, one of the study’s authors.

Limitations and Additional Insights

Of course, like all scientific research, the MIT study has some caveats. First, while 7,600 participants is a large sample, it may not capture the full range of perspectives among American voters. Second, as the researchers themselves note, the slight persuasive advantage of video over text may be even smaller outside of a research setting.

“It’s possible that when scrolling through your news feed, you’re more likely to notice a video than a text post. You might be more likely to watch it. But that doesn’t mean the video is more persuasive than text. It just means it’s likely to reach a wider audience,” noted David Rand, another author of the study.

Conclusion

In other words (at least according to the study’s authors), a deepfake video featuring a politician is unlikely to influence someone’s opinion more effectively than fake news about that same politician. The only real advantage video has over text is that people are more inclined to believe what they see than what they read. Additionally, video content may reach a broader audience.