Imagine you are in a video call where you are just introducing yourself. The person on the other side has already sought permission to record the interaction. You being naive to the range of possibilities agreed to it. It’s not uncommon for companies to record their Zoom meetings for future reference. In fact, it’s the norm until you get shocked by what comes next.
Imagine you are in a video call where you are just introducing yourself. The person on the other side has already sought permission to record the interaction. You being naive to the range of possibilities agreed to it. It’s not uncommon for companies to record their Zoom meetings for future reference. In fact, it’s the norm until you get shocked by what comes next.
Simply by recording your voice samples and using a free AI Software, the person on the other side plays an audio where you are heard speaking passionately on the Israel-Palestine issue. You know it’s not you because you don’t share that opinion or have strong views or knowledge of the Middle East. The voice recording is so perfect and disorienting that you can’t see how you can deny it.
Welcome to the age of AI where risks of Deepfakes are high and personal.
Deepfakes are extraordinary technology that can replicate the original so impeccably that it appears high on credibility and therefore prone to misuse. Among Deepfakes, audio and video are sought-after while manipulating. Last year, within hours before the closure of polls in Slovakian elections, an audio fake of one of the candidates claiming to have rigged the election went viral. Within hours, enraged voters lined up in the polling booths in huge numbers attempting to reverse the rigging. They ended up delivering a heavy defeat to that candidate. The Deepfake audio stirred the opposing forces to act fast and unitedly changed the electoral fortunes of the candidate. This year, London mayor Sadiq Khan was targeted with fake audio of him making inflammatory remarks calling for pro-Palestinian marches.
While audio and video deepfakes are equally harmful in ruining hard-earned reputation, audio scores higher than video due to its remarkable ease in producing it. Video take more effort and precision to be deployed by scammers. Video requires sophistication and finishing skills so that lip sync appears accurate. Further, video Deepfakes are easy to debunk as viewers can easily go to YouTube and find the Original video to crosscheck. In case of audio Deepfake say a recorded telephone conversation, there is no medium to verify the authenticity. The background and context in which the statement was made can be thoroughly erased in the case of audio recordings. In the case of video, the objects and people in the background can also turn useful reference points in reverse image search of the original content. In the case of audio Deepfakes, the absence of such options make it difficult to decipher. Moreover, voice samples are easier to acquire and fake than videos. Creating a video requires someone pointing the camera at individual discreetly or with spy cameras while an audio is relatively easier to record without permission.
In the recent Indian elections, Deepfakes were used extensively to influence the poll outcomes. Some candidates had created customized videos seeking votes. Deepfakes were deployed to ensure that every voter was addressed in the video by their first name so that it appears like a personalized request. Deepfakes can enable the election candidates by using their voice samples and disseminating millions of videos with minor Deepfake i.e the part of video where they address the voter by their first name. Using AI, voters would feel that their candidate invested time in personally creating a video for them. Further, a candidate’s voice samples can be used to make videos in different languages. A Tamil voter in Delhi can be sent the candidate’s video of him seeking votes in Tamil. Addressing people in their mother tongue can improve the appeal of the candidate. However, it’s harder for candidates to fake in the video format because of the extreme precision required to make the lip sync seem natural. That’s where audio trumps the video format. Robocall in candidate’s voice with the audio addressing the voter by his first name in his native language thus became a potent tool to use.
Audio is equally potent to misuse. In January, a voice message in the sound of US President Joe Biden went viral. The audio urged the voters in the New Hampshire to not vote in the Democratic Primary and instead “save your vote for the November election”. The message was alarming as it demonstrated the ease with which it was edited, cheap manner it was produced and difficulty in tracing the creator. It demonstrates that all it takes is a notorious individual, a free software, and a big database to wreak havoc on the election process. In the Deepfakes ecosystem, audio is beating video in terms of impact. It’s only a matter of time that AI develops to produce equally cheap and easy to edit versions in the video format. Till then, audio will remain the more destructive threat.
At pi-labs, we are on a mission to keep the internet clean using our advanced deepfake detection tool designed for enterprises. Authentify by pi-labs leverages cutting-edge AI++ technology to identify and mitigate the risks posed by deepfakes, ensuring the integrity of your digital content. Visit our website to explore the features of Authentify, try out our tool, and receive a comprehensive report on deepfake analysis.
Founder & CEO, pi-labs
ankush@pi-labs.ai
