The Deepfake Journalists
Look at the image above. The picture to the left represents Oliver Taylor, a 24-year-old student of Political Science. Here is a screenshot of his old profile on Quora:
Looking at his picture and profile text, nothing immediately suggests that he is a fake persona with a synthetic, AI-generated photo. His blog posts on anti-Semitism and Jewish affairs have previously been featured in prominent newspapers such as Jerusalem Post and the Times of Israel.
In a blog post from late 2018, Oliver Taylor accused the academic, Palestine-supporting activist couple, Mazen Masr and Ryvka Barnard, of being “known terrorist sympathizers”. Masr and Barnard were confused by the harsh accusations seemingly coming out of nowhere from the unfamiliar British university student. The couple also found that something was off about the young man’s face but couldn’t quite put their finger on it. They contacted Reuters to call attention to the odd situation.
Reuters consulted the deepfake detection company Cyabra who disclosed Oliver Taylor’s “real” – or rather fake – identity. The picture to the right is a handout by Reuters, a heat map produced by one of Cyabra’s algorithms to highlight areas of suspected computer manipulation. Subtle details in the image to the left such as the undulating tiles in the background, the asymmetric collars on Taylor’s shirt, his misshapen teeth, and strange flyaway hair also give Taylor’s fake persona away under the scrutiny of experts. The real persons behind Oliver Taylor’s profile were never discovered.
To take another strange example of how “deepfake journalism” tries to influence politics and individual people’s destiny: a security analyst named Martin Aspen from a Swiss intelligence firm authored a 64-page document in 2020 that linked Joe Biden’s son, Hunter Biden, to shady business dealings in China. The document was leaked, spread on social media, and picked up by right-wing channels on the internet and by close associates of Donald Trump. Although disinformation researchers found that the report was forged and the author was a completely fictitious person, it gained a considerable amount of attention up to the 2020 presidential election. See Martin Aspen’s profile photo on Twitter below.
I bet that very few people would be able to spot that this person is not real. So-called “GAN-generated photos” (GAN stands for Generative Adversarial Networks) are practically indistinguishable from photos of real people. And they are not even that difficult to generate. Check out https://thispersondoesnotexist.com/ where you can browse and use photos of “deepfake people” for free, and learn more about the underlying technology as well.
The GAN-generated photos ads an extra degree of creditability to the fake journalist, blogger, or social media user, while the photos are also “untraceable”. Compared to simply grabbing a real person’s photo from the web, there is no original photo or person to be found, which is exactly why the GAN-generated photos are dangerous. As the technology matures, there is indeed a chance that even experts will be unable to detect that a certain image of a person is a deepfake.
Spread of Fake News on Social Media
Fake or misleading news and the use of pen names is obviously not a new phenomenon in journalism. The difference between now and before the digital revolution is that new stories can spread quicker than wildfires online. When news was primarily consumed in printed form, the scope and reach were much narrower. But thanks to social media networks, search engines, and the borderless nature of the internet, a piece of information can be obtained by millions of people simultaneously.
The speed at which news can travel to large groups of people on a global scale is a blessing as well as a threat depending on the truth value of a particular story. If a story is factual, it can only be considered a positive thing that the story can reach virtually anyone at once, but if the story is false, more people will correspondingly be misinformed with potentially damaging consequences.
The far majority of young Europeans and Americans rely on social media as their primary or only news outlet. As a result, younger generations are more exposed to misinformation, false conspiracy theories, and other fake news that might serve other people’s or organization’s hidden objectives. The fear is that fake online news will increasingly blur the line between true and false, make young people disoriented, susceptible to manipulation, and generally shape their perception of the world in a negative manner.
A study suggests that fake news on Twitter spread faster, farther, deeper, and more broadly than factual news. Key findings from the study include that:
- On average, false rumors reached 1500 people 6 times faster than true rumors.
- False rumors were 70% more likely to get retweeted than true rumors.
- The largest rumor cascades reached around 50.000 users when the rumor was false but only around 2.000 users when it was true.
- Bots appear to have accelerated the spread of true and false rumors at roughly equal rates.
The simple reason why social media users are more drawn to false rumors than true rumors is deeply rooted within human psychology. From a primal perspective, humans are naturally more drawn to a story that shocks us, provokes us, instills fear or disgust in us, than we are to an objectively true, ordinary, boring story. Although a part of us may suspect that the former story is false and the latter is truthful, we would rather engage with the false one, since it provokes strong feelings in us. In this way, the algorithms of Twitter and Facebook tap into our basic human psychology and exploit it.
A classic example of how fake news prevails over factual news on social media is the 2016 American Presidential election. In November 2016, 159 million visits to fake news websites were recorded. In the final three months of the US presidential campaign, the top-performing fake election news stories on Facebook generated more engagement than the top stories from major news outlets such as the New York Times, Washington Post, Huffington Post, NBC News, and others. It is believed that fake news also had a non-trivial influence on the UK Brexit referendum from 2016.
During the 2020 American Presidential election, agents from the Russian government came prepared with a fake news site called PeaceData that published more than 500 articles in English and about 200 in Arabic that were shared on Facebook, Twitter, and LinkedIn. The articles were shared by fake profiles with GAN-generated photos to steer left-winged voters towards the Trump Administration. Luckily, Facebook and Twitter reacted promptly. Facebook removed 13 accounts and two pages that together had gained 14,000 followers, while Twitter said that it also suspended accounts and blocked out any content from the PeaceData website.
Finally, in August 2020, the nonprofit organization Avaaz released a report regarding the state of global health misinformation, including anti-vax propaganda, on Facebook throughout the preceding year.  The report estimated that content on Facebook pages and groups that shared fake news about the global health situation received around four billion views. In April 2019, content from the top ten most viewed fake news sites on global health received four times as many views as content from the top ten leading authoritative sources such as the WHO and the CDC.
Do not believe in everything you read, especially not online. The majority of news you find on Facebook and Twitter is probably fake. This post has covered how writers can conceal their identities behind deepfake journalists with GAN-generated photos, and how prevailing fake news statistically is on social media.
Another aspect of deepfake journalism that I did not have space to cover in this post, is how AI technologies such as GPT-3 can generate texts, on any subject, that is indistinguishable from articles written by humans. I will dive into that topic in my next post, where I will also write about Artificial General Intelligence with skepticism.
 Noah Giansiracusa (2021), How algorithms create and prevent fake news – exploring the impact of social media, deepfakes, GPT-3, and more, pg. 18.
 Giansiracusa (2021), pg. 21.
 See https://www.pewresearch.org/journalism/2018/10/30/younger-europeans-are-far-more-likely-to-get-news-from-social-media/ (EU) and https://www.statista.com/statistics/1124119/gen-z-news-consumption-us/ (US).
 Soroush Vosoughi, Deb Roy, and Sinan Aral, “The spread of true and false news online,” Science 359 no. 6380 (2018): https://science.sciencemag.org/content/359/6380/1146.
 Allcott, Hunt, and Matthew Gentzkow (2017) “Social Media and Fake News in the 2016 Election.” Journal of Economic Perspectives, 31 (2): 211-36.
 “Facebook’s Algorithm: A Major Threat to Public Health,” Avaaz, August 19, 2020:
https://secure.avaaz.org/campaign/en/facebook_threat_health/ and Giansiracusa (2021), pg. 184.