MORE THAN A fifth of people who unknowingly watched deepfake videos of Irish politicians being interviewed as part of a study told researchers that they “remembered” the interviews actually taking place.

However, the study by researchers at University College Cork also found that participants were almost as likely to form a false memory of a fake story that was presented in a text format as one which was shown to them as a deepfake video. 

Deepfakes are digitally manipulated images, video and audio that are designed to create fake material featuring the likeness of an individual, often to misrepresent their views or speech.

They have become easier to make in recent years and are increasingly used for malicious purposes, including the spread of misinformation but also for use in scams and as part of the production of non-consensual pornography.

As part of the UCC study, more than 2,000 people were shown a series of news stories and asked to tell researchers where they had seen those stories before, and how they affected their political views.

The participants were also unknowingly shown a deepfake video of an Irish politician – one of Taoiseach Simon Harris, Tánaiste Micheál Martin or Sinn Féin leader Mary Lou McDonald – mixed in with the real stories.

Each of the deepfakes featured Harris, Martin or McDonald saying something politically damaging about the topic of housing.

People were also shown the same stories, including the fakes, in text form in order to gauge whether deepfakes were more effective at spreading misinformation.

The study found that those who watched the deepfake videos formed a false memory that they “remembered” the story about 23% of the time, while those who saw the stories in text format had a false memory of “remembering” them about 22% of the time.

Those who saw the deepfake videos also reported a slightly reduced intention to vote for the featured politician (a 4% decrease). However, when they saw the same story in text format, they reported a much larger reduction in voting intention (a 13% decrease).

However, the study also found that an overwhelming number of people were able to identify deepfakes and fake news when alerted to the fact that they were contained with the study.

Participants were a slightly more likely to detect a fake story when it was presented in text format (83%) compared to those who detected a deepfake (78%). 

Dr Gillian Murphy, who led the study, said the findings show how deepfakes are not necessarily more problematic than text-based misinformation.

“Recent evidence pushes back against the gut instinct that most of us share – that “seeing is believing” and this technologically sophisticated form of misinformation will be more powerful than what has come before,” she said.

“To be clear, these findings do not mean that deepfakes are not potentially harmful, merely that we do not yet have evidence that they are drastically more potent than simple written misinformation.”

Interested in taking part? Find out more here.