New research indicates that fake faces created by artificial intelligence appear more credible to humans than actual people. Deep learning – an algorithmic learning process used to train computers – is utilized to produce a human who looks real, a technology known as ‘deepfakes’. In addition, they can be used to transmit messages that have never been expressed. For example, an altered video of Richard Nixon’s Apollo 11 address or a phoney Barack Obama attack on Donald Trump.
The account name was the only clear indication that this wasn’t the real deal when TikTok videos allegedly showed ‘Tom Cruise’ making a penny disappear and eating a lollipop appeared in 2021. The creator of ‘deeptomcruise’ is using ‘deepfake’ technology to present a machine-generated version of the famous actor doing magic tricks and having a solo dance-off on social media.
Researchers asked participants to determine whether StyleGAN2 faces derived from an algorithm were authentic or artificial. The participants had a success rate of 48%, which is somewhat lower than flipping a coin. Using the same data set in a second experiment, participants were trained on how to spot deep fakes, but their accuracy rate only increased to 59 percent.
The University of Oxford, Brown University, and the Royal Society conducted research that found that most individuals were unable to determine whether they were watching a deepfake video, even when they were warned that the content might be fake.