In A World of Deepfakes, Ridge Senior Offers Truthful Solutions

Samantha Liu ‘22

Angelina Xu ’21, Ridge High School senior and Ridge Devil’s Advocate’s very own Co-Editor-in-Chief, now boasts a new addition to her accolades: a semifinalist for the 2021 Regeneron Student Talent Search. Her project was one of 300 recognized by the nation’s oldest student research competition, and her prize includes $4,000 to be split between herself and the school [1].


Xu’s research centered around detecting deepfakes, scarily realistic videos fabricated in the imitation of another person. For Xu, the implications of such technology were harrowing. “A friend in [Governor’s] school first told me about deepfakes and it honestly shook me,” she explains. “Misinformation was already a major issue and deepfakes could completely alter the arena.” 


Specifically, she was concerned about the potential spread of misinformation regarding the COVID-19 crisis, including false cures or pseudo-authoritative orders. Additionally, although it did not turn out to be an issue, Xu had originally feared the possibility of malicious deepfake videos imitating either of the presidential candidates before November 2020, circulating virally via the internet. Such a story would be not unprecedented; in 2019, a digitally altered video of Nancy Pelosi appearing drunk garnered 98,000 likes and millions of views before it was debunked [2]. 

In the past, programs relied on recognizing irregular eye-blinking or facial movement to detect synthetic videos. But Xu’s research utilized a micro approach: she decomposed over 12,000 images into different wavelengths and inputted them, until the computer picked up patterns to discern real from fake. Although the process might sound complicated, Xu finds that “this method requires simpler machine learning algorithms and offers greater accuracy.” 

“Greater accuracy” is an understatement. The formerly highest accuracy, determined in an international competition held by Facebook and Google, was 65%, which Xu’s algorithm raised to 98% and 100% accuracy for low and high-resolution images respectively [3]. 

Most might believe such a large and successful undertaking would require cutting-edge laboratories, professional guidance, and all sorts of special knowledge. But this project, technical as it seems, was researched and created completely independently, and Xu had only started coding a couple months prior. She attributes her accomplishments to the abundance of resources on the internet and urges other curious students to do the same: “We have YouTube and Google nowadays so saying that you can’t figure it out is an excuse. There’s so many resources out there. You just have to go find it.” 


And find it she did, reaping the benefits when her work, after a rigorous vetting process, was awarded semifinalist status in the Regeneron Student Talent Search. Xu will now go on to join a network of other alumni from the nation’s oldest and most prestigious science competition from which student-conceived research ideas often manifest into real-world technology and large-scale implementations. For Xu, if a version of her project were commercialized and adapted, she muses over possibilities for how companies could curb the misinformation that fuels racial tensions, political polarization, and social prejudices—issues to which, being a student journalist as well, Xu is no stranger. She expresses concern over the politicization of science and wonders if a program that fact-checks audio in real-time could be created in the near future. “We need to promote reliable news sources and make sure that news sources uphold truth over profits, she explains. “There’s definitely a fine-line between fact-checking and the first amendment but maybe science can provide an objective benchmark.”