May 22, 2024

New AI Voice-Cloning Tools Have Potential to Spread Misinformation on Social Media

5 min read


In a video clip from a January 25 news report, President Joe Biden talks about tanks. But a doctored model of the video clip has amassed hundred of hundreds of sights this week on social media, producing it surface he gave a speech that attacks transgender folks.

Digital forensics experts say the movie was developed applying a new era of artificial intelligence resources, which allow for anybody to rapidly generate audio simulating a person’s voice with a number of clicks of a button. And when the Biden clip on social media could have unsuccessful to idiot most buyers this time, the clip demonstrates how quick it now is for men and women to make hateful and disinformation-crammed “deepfake” video clips that could do actual-globe damage.

“Tools like this are heading to generally incorporate additional fuel to fireplace,” reported Hafiz Malik, a professor of electrical and computer engineering at the College of Michigan who focuses on multimedia forensics. “The monster is currently on the unfastened.”

It arrived past month with the beta section of ElevenLabs’ voice synthesis platform, which authorized customers to create reasonable audio of any person’s voice by uploading a several minutes of audio samples and typing in any textual content for it to say.

The startup states the technology was produced to dub audio in various languages for movies, audiobooks, and gaming to protect the speaker’s voice and emotions.

Social media customers promptly began sharing an AI-generated audio sample of Hillary Clinton reading the same transphobic text highlighted in the Biden clip, along with phony audio clips of Bill Gates supposedly expressing that the COVID-19 vaccine causes AIDS and actress Emma Watson purportedly reading Hitler’s manifesto “Mein Kampf.”

Shortly just after, ElevenLabs tweeted that it was viewing “an rising range of voice cloning misuse instances,” and declared that it was now exploring safeguards to tamp down on abuse. One of the to start with methods was to make the feature obtainable only to those who offer payment information and facts. Originally, nameless consumers were being able to obtain the voice cloning software for free. The corporation also statements that if there are troubles, it can trace any produced audio back again to the creator.

But even the means to keep track of creators will never mitigate the tool’s harm, mentioned Hany Farid, a professor at the University of California, Berkeley, who focuses on electronic forensics and misinformation.

“The damage is performed,” he mentioned.

As an case in point, Farid mentioned terrible actors could go the stock market place with pretend audio of a best CEO stating revenue are down. And now there is certainly a clip on YouTube that employed the instrument to change a online video to make it seem Biden explained the US was launching a nuclear assault in opposition to Russia.

Totally free and open up-source software program with the same capabilities have also emerged on the internet, that means paywalls on commercial tools are not an impediment. Working with 1 free on the web product, the AP produced audio samples to sound like actors Daniel Craig and Jennifer Lawrence in just a couple minutes.

“The question is exactly where to level the finger and how to place the genie again in the bottle?” Malik reported. “We are unable to do it.”

When deepfakes initial made headlines about 5 many years in the past, they were being quick sufficient to detect because the subject did not blink and the audio sounded robotic. That is no for a longer time the scenario as the applications grow to be extra innovative.

The altered video clip of Biden producing derogatory feedback about transgender men and women, for occasion, merged the AI-generated audio with a serious clip of the president, taken from a January 25 CNN stay broadcast announcing the US dispatch of tanks to Ukraine. Biden’s mouth was manipulated in the online video to match the audio. Although most Twitter buyers acknowledged that the content was not something Biden was likely to say, they ended up nonetheless shocked at how sensible it appeared. Some others appeared to think it was authentic – or at the very least didn’t know what to imagine.

Hollywood studios have very long been equipped to distort reality, but accessibility to that technological know-how has been democratized with no contemplating the implications, mentioned Farid.

“It’s a mixture of the incredibly, quite potent AI-dependent engineering, the relieve of use, and then the fact that the design appears to be: let us put it on the web and see what transpires next,” Farid explained.

Audio is just one particular region exactly where AI-created misinformation poses a threat.

Free of charge online AI picture generators like Midjourney and DALL-E can churn out photorealistic images of war and purely natural disasters in the design of legacy media outlets with a straightforward textual content prompt. Last month, some college districts in the US commenced blocking ChatGPT, which can create readable textual content – like college student time period papers – on need.

ElevenLabs did not answer to a request for remark.

Affiliate links might be routinely produced – see our ethics statement for specifics.


Supply hyperlink The growth of artificial intelligence (AI) has increased rapidly over the last few years, with new applications being developed every day. One of the most powerful AI tools is the ability to create voice clones using text to speech technology (TTS). AI voice-cloning tools have the potential to spread misinformation on social media, a consequence that could be devastating for democracies.

TTS software allows users to type in a text string, and then generates a realistic sounding voice that reads out the text. This technology is becoming increasingly sophisticated, and will soon enable a range of powerful applications. For example, it could be used to create convincing fake audio recordings of famous people, or to impersonate someone’s voice.

These capabilities are especially dangerous in a world where the speed of digital communications vastly exceeds the speed of fact-checking. A single false audio clip, created with a voice clone generated by AI, could spread within minutes on social media. It could be used to spread misinformation, disinformation, or hate speech.

It is clear that AI voice-cloning tools have the potential to cause great harm. There are several steps that can be taken to reduce this risk. First, social media platforms should employ automatic detection tools to help identify suspect audio clips and take action to remove them. In addition, users should be educated on how to identify suspicious audio recordings, and governments should pass laws that make it a criminal offence to spread false audio recordings.

The rise of AI voice-cloning tools creates a real and present danger to democracies. It is incumbent on everyone to ensure that these powerful tools are not misused to spread misinformation, disinformation, or hate speech on social media.