At some point during the 2024 Presidential Election battle, a damning tape or audio clip will probably surface. And one or more of the candidates will have to prove that it is an AI deepfake. 

The evidence may be real or possibly a sinister AI creation built by astonishingly lifelike generative AI models.

Generative AI experts assure us that they possess the tools to detect these fraudsters, but proving the authenticity of a recording remains an elusive endeavor. 

The real question is whether such evidence will truly matter to partisan voters, especially those quick to discard any information that contradicts their deeply entrenched worldviews.

NARA Documents: White House Invokes ‘Executive Privilege’ to Shield Biden Family's Controversial Past – USA Herald

Deepfake audio, those eerily authentic yet entirely false recordings constructed from snippets of someone speaking, have achieved a level of realism that can deceive even those who know you best.

This poses a grave threat to the political landscape, offering nefarious actors the potential to manipulate and deceive on a grand scale.

AI developers are sounding the alarm. They warn that the rapid advancement and widespread use of this technology risks ushering in an “epistemological dystopia”, one that threatens the very foundations of representative democracy. 

AI deepfake and politics

Hany Farid, a generative AI expert at the University of California, Berkeley, emphasized the high stakes on political campaigns. This is where state-sponsored actors, campaigns, trolls, supporters, and PACs all vie to gain the upper hand through any means necessary.

During testimony before a Senate Judiciary subcommittee, OpenAI CEO Sam Altman labeled AI's capacity to generate personalized disinformation as one of his most profound concerns. 

A United Nations adviser went so far as to express deep apprehension about the possibility of a deepfake October surprise.

The manipulation of media to mislead voters is not a new concept, and it doesn't necessarily require AI. Campaign attack ads have a long history of employing unflattering imagery and creative editing to cast opponents as shifty-eyed deceivers.

However, generative AI will supercharge these abilities, enabling campaigns and their supporters to produce eerily convincing fabrications.

Audio and visual deepfakes

Replicating a specific person's voice in a convincing manner was a near-impossibility just a few years ago. Even the most talented impressionists could only get close. 

Today, companies like Respeecher can analyze mere minutes of someone's voice to generate a convincing sonic clone, capable of saying anything. 

While computers alone may not yet be capable of producing visually convincing videos, skilled editors can enhance AI-generated content to create something that appears entirely authentic on a small screen. Given the relentless growth of computing power, it's only a matter of time before AI text-to-video platforms can seamlessly deceive viewers.

The experts and companies in this field unanimously agree that the most effective way to combat the damage caused by deepfakes is for the industry to adopt proactive standards. And use "passive" methods for analyzing media to uncover generative AI markers.

Farid's lab focuses on "passive" detection, researching methods to reveal AI's fingerprints in media content. 

While Farid is confident in his lab's ability to trace AI's involvement in specific clips, scaling this effort to the vast expanse of online content is an impossible feat.

Standards and regulation

This two-pronged approach appears to be the favored strategy within the burgeoning industry. Established firms like Adobe, Microsoft, and the BBC lead the Coalition for Content Provenance and Authenticity (C2PA), which is actively developing technical standards for certifying the authenticity of digital content sources.

During a White House gathering of AI corporate leaders in July, these firms pledged to implement active detection protocols.

 Anna Bulakh, head of ethics and partnerships at Respeecher, likened this ongoing effort to the migration of websites to more secure encrypted protocols, highlighting the potential for AI firms to collaborate with governments in mitigating technology abuse.

However, even the firms that have established conduct standards for their products have been unable to prevent users from abusing their technology. Recent investigations revealed that OpenAI's ChatGPT allowed users to generate personalized arguments for manipulating political views, despite efforts to prohibit such misuse.

The inherent challenges of industry self-regulation have led some AI companies to advocate for government intervention. Opinions vary on the form that government involvement should take.

Some suggest imposing "know-your-customer" rules on chipmakers who provide AI firms with computing power, akin to the way banks flag suspicious transactions. Others, like AI detection firm Reality Defender, have called on the Federal Elections Commission to develop methods for scanning all political materials for AI deepfakes.

In the United States, government concern over AI has grown, but tangible action has been limited. 

The FEC is reviewing a petition to ban politicians from using generative AI to deliberately misrepresent their opponents. But the authority to do so remains contested. 

In Congress, discussions on AI's impact have been sporadic, with only a few hearings and a handful of bills introduced on the topic.

As the 2024 election draws nearer, the looming threat of AI deepfakes casts a shadow over the political landscape. 

Whether technology will outpace regulation, and whether voters will be swayed by these digital deceptions, remain questions that demand urgent attention from policymakers, tech leaders, and the public at large.