Scientists may have found an effective and inexpensive way to detect deepfakes, video and audio recordings generated through the use of artificial intelligence.
At the Black Hat 2019 conference held in Las Vegas, a team from the University of Oregon's Institute of Neuroscience claimed that they trained laboratory mice to identify irregularity in recorded speech, which could be a sign that the audio or video was manipulated.
If true, the proposed system — albeit not convenient — could help sites such as Facebook and YouTube to detect deepfakes before they spread online.
How Lab Mice Can Detect Deepfakes
The researchers explained that mice have an auditory system similar to humans. However, unlike humans, mice cannot understand speech.
If an audio recording included a subtle mistake that most humans would simply gloss over, the mice trained to discern these inconsistencies would immediately notice.
To test, the team taught lab mice the difference between "buh" and "guh" sounds across different contexts and surrounded by different vowels. Whenever the mice identified speech sounds correctly, they were given a reward.
The researchers reported that the mice were able to distinguish different speech sounds with 80 percent accuracy.
"And because they can learn this really complex problem of categorising different speech sounds, we think that it should be possible to train the mice to detect fake and real speech," said Jonathan Saunders, one of the researchers, to the BBC.
Studying Mice Can Help Fight Deepfakes
Saunders and his colleagues do not expect to see an army of mice employed at Facebook and Google. While adorable, the researchers admitted that using mice for the sole purpose of searching for inconsistencies in recorded speech would not be practical. Instead, they hope that the study could help tech companies develop a more efficient way to detect manipulated audios and videos.
"[S]tudying the computational mechanisms by which the mammalian auditory system detects fake audio could inform next-generation, generalizable algorithms for spoof detection," the researchers wrote in a white paper.
The proposal came at a time when deepfakes are increasingly becoming more ubiquitous and sophisticated. While most are still obviously manipulated, experts fear that deepfakes would become undetectable in the near future and will be used as a tool to spread false information.
Ahead of the 2020 presidential elections, Rep. Adam Schiff, who heads the House Intelligence Committee, sent letters to Google, Facebook, and Twitter to ask the tech giants how they will deal with deepfakes.
