Two images of Pope Francis show a faked photo (left) of the pontiff in a puffy coat and (right) in a news photo taken of the pope last May at the Vatican. (Bloomberg)

Two images of Pope Francis show a faked photo (left) of the pontiff in a puffy coat and (right) in a news photo taken of the pope last May at the Vatican. (Bloomberg)

Comment: AI, deepfakes may offer advantage to old-school media

With deepfakes raising doubts about social media, the is ‘legacy’ media’s chance to regain lost ground.

By Leonid Bershidsky / Bloomberg Opinion

Russian pop music producer Iosif Prigozhin (not to be confused with Yevgeny Prigozhin, founder of Russia’s Wagner mercenary army) publicly supports Vladimir Putin and the Russian invasion of Ukraine. Earlier this month, however, a recording, ostensibly of Prigozhin’s expletive-laced telephone conversation with sanctioned billionaire Farkhad Akhmedov, was leaked on YouTube; and on it, both parties curse “Satan” Putin, his greedy, incompetent team and his dumb war.

The content of the conversation is mildly interesting and the insults and complaints are good for a laugh, but ultimately, the recording offers nothing particularly revealing. A year into the invasion, it’s hardly new knowledge that Putin dragged Russia into the war against the will of its hedonistic business elite, that this elite resents the inconvenience of sanctions and that it nevertheless won’t stand up to Putin publicly because that’s a sure path to ruin. What’s notable about the leak is Prigozhin’s defense: He claims the conversation, or at least parts of it, was generated using artificial intelligence; and today, it’s impossible to prove otherwise.

“Neural networks work all kinds of miracles today,” Prigozhin said in an interview. “This recording is a mix of phrases we said and those that were generated but never said.”

Now that the world has seen realistic images of Putin apprehended and on trial in The Hague, Putin bending his knee before China’s Xi Jinping, Donald Trump trying to resist arrest but ending up in jail anyway, Pope Francis wearing a faux Balenciaga puffer coat, why not a deepfake phone call between two wealthy Russians slagging off Putin? Now that a large language model can convincingly imitate the style of any writer, living or dead, why not the style of a slightly tipsy music manager?

Some people will claim to be able to tell immediately if a digital object — a picture, a text, an audio or video file — has been created by generative AI, and they already do: They’ve noticed that the pope in the viral image has a weird-looking right extremity, his glasses aren’t quite right, and anyway, he wouldn’t wear anything this fatuous. Experts claim that the Prigozhin-Akhmetov recording would have been next to impossible (not 100% impossible) to fake in full or in part.

I wouldn’t, however, lend much credence to such claims, based as they may be on expertise rooted in experience. These are revolutionary times, and the quality leap of AI services such as ChatGPT and Midjourney has been so enormous that it would be nearsighted not to expect fully realistic output in the foreseeable future. The ease of generating any kind of media already appears to be increasing exponentially. We will doubtless be force-fed this stuff; “until false is true, until day is night, until wrong feels right,” as Iggy Pop once sang. Most people won’t even bother to look closely enough, let alone use whatever intricate tools will be required to authenticate a media file.

Until recently, the emergence of a good deepfake was newsworthy. The fake could be relished and debunked, a dark future could be half-seriously predicted. Even today, the pope image and some others of its kind have been treated like this. But because the future is already here, these takes are behind the curve. The fakes are part of a new reality, and Prigozhin’s defense is destined to become ubiquitous;and as easy to resort to as the fakes are for a lay person to create.

This development should spell the end of social networks as a news source. Today, about half of Americans use them as such, but once they realize it’s no longer possible to tell if anything on them is real in the old-fashioned way — meaning that it’s created by the same person who signed it or that it depicts what it claims to depict — they should stop. Not that they necessarily will, of course; many of us prefer to look for a comfortable version of the world rather than confront the disagreeable reality. For those who seek a reasonable approximation of the truth, however, the use of Twitter, Facebook and the rest of them will become less reliably useful and more time consuming. Even now, filtering out the noise is a challenge. In an instantly falsifiable digital reality, it will require constant attention.

Twitter owner Elon Musk has argued that paying for account verification is the only option for social network users who want to keep interacting with humans rather than artificial entities. “It’s now trivial to spin up 100k humanlike bots for less than a penny per account,” he tweeted. “Paid verification increases bot cost by ~10,000%.” He also predicted that paid authorization will soon make Twitter the only trustworthy network because it will stop showing unverified (that is, unpaid) accounts in a user’s ‘For You’ algorithmic feed. The arguments may help Musk drum up some revenue, but they make little sense from any other perspective. Depending on how useful AI-generated content is to its mass-producers, the $7 per account that Musk is charging may be an acceptable cost, and plenty of actual human users will help spread the content if they fall for it, find it useful for their own purposes or simply for fun.

The AI revolution will also wreak havoc with web search, the preferred research method of most students, some academics and many journalists today. Even if links are provided in the search results, they may lead to fake sources; current tools can fill entire books and academic journals with content created by large language models. What is seen as a trusted source today may be irretrievably corrupted tomorrow by the trigger-happy use of the new “magical” techniques. Thus, search as a brain extension — the way most people use it today — potentially becomes a grotesque reality modifier.A true revolution often wipes out patterns of ingrained habit. It may, however, bring back some even older, forgotten ones.

What the tech world has come to call “legacy media” have an opportunity to claw back much of their lost credibility and clout. Taking to people and working in real-world archives, analog or digital, are again, as they were 30 years ago, almost the only methods of finding out the real state of affairs. The ability to get information from the direct source suddenly becomes more valuable to the curious information-seeker and the information consumer worried about veracity. Journalists are trained communicators, interviewers, seekers of newsworthy grains of meaning in countless boring documents. Even those of us who have not practiced these skills recently because of fascination with modern tech tools can quickly recall how things used to be done before these tools existed.

The “legacy media” we work for have lost their role as information gatekeepers as technology erased entry barriers. Now, the verified and proven ability to get real information and strict rules against faking it are suddenly relevant again. A leaked audio posted on the web is worthless when it cannot be authenticated; but a recording made by a specific journalist with a job to protect and a news organization with a reputation to safeguard is a different matter. Iosif Prigozhin’s AI defense will probably be lost on Putin, known for his vengefulness. As a mere news consumer, however, I simply cannot be sure that he actually said the things his voice is heard saying on the recording; and I wish a journalist I could trust had made the tape.

The media may still waste this opportunity: To use it, a certain return to the basics is required in working with human sources and being at the scene, and that can be an expensive undertaking. The investment can be worth it, however. At its core, news is an artisanal product with a broader appeal than, say, handmade shoes or custom-built bicycles; and, like them, it cannot be replaced with the output of the most perfect AI model.

Leonid Bershidsky, formerly Bloomberg Opinion’s Europe columnist, is a member of the Bloomberg News Automation Team. He recently published Russian translations of George Orwell’s “1984” and Franz Kafka’s “The Trial.”

Talk to us

> Give us your news tips.

> Send us a letter to the editor.

> More Herald contact information.

More in Opinion

toon
Editorial cartoons for Friday, March 29

A sketchy look at the news of the day.… Continue reading

Initiative promoter Tim Eyman takes a selfie photo before the start of a session of Thurston County Superior Court, Wednesday, Feb. 10, 2021, in Olympia, Wash. Eyman, who ran initiative campaigns across Washington for decades, will no longer be allowed to have any financial control over political committees, under a ruling from Superior Court Judge James Dixon Wednesday that blasted Eyman for using donor's contributions to line his own pocket. Eyman was also told to pay more than $2.5 million in penalties. (AP Photo/Ted S. Warren)
Editorial: Initiative fee increase protects process, taxpayers

Bumped up to $156 from $5, the increase may discourage attempts to game the initiative process.

Schwab: Who was Langerhans? And when’s the ferry to his islets?

The Herald’s resident retired surgeon slices into the anatomy of the etymology of our anatomy.

Comment: Cervial cancer treatable; if you’re screened for it

A screening for cervical cancer can detect cancerous or precancerous cells and direct treatment.

Comment: Framers gave us Goldilocks Constitution; let’s use it

It was meant to be resilient, not perfect, but it has to be used as designed toward workable solutions.

Comment: GOP in Congress isn’t fighting crime; it’s arming it

Budget cuts to the FBI and ATF and other riders have made it easier for criminals to get firearms.

Washington state senators and representatives along with Governor Inslee and FTA Administrator Nuria Fernandez break ground at the Swift Orange Line on Tuesday, April 19, 2022 in Lynnwood, Washington. (Olivia Vanni / The Herald)
Editorial: Community Transit making most of Link’s arrival

The Lynnwood light rail station will allow the transit agency to improve routes and frequency of buses.

An image of Everett Mayor Cassie Franklin is reflected in a storefront window during the State of the City Address on Thursday, March 21, 2024, at thee Everett Mall in Everett, Washington. (Ryan Berry / The Herald)
Editorial: State of city address makes case for Everett’s future

Mayor Franklin outlines challenges and responses as the city approaches significant decisions.

FILE - The massive mudslide that killed 43 people in the community of Oso, Wash., is viewed from the air on March 24, 2014. (AP Photo/Ted S. Warren, File)
Editorial: Mapping landslide risks honors those lost in Oso

Efforts continue in the state to map areas prone to landslides and prevent losses of life and property.

Protecting forests and prevent another landslide like Oso

Thank you for the powerful and heartbreaking article about the Oso landslide… Continue reading

Boeing’s downfall started when engineers demoted

Boeing used to be run by engineers who made money to build… Continue reading

Learn swimming safety to protect kids at beach, pool

Don’t forget to dive into water safety before hitting the pool or… Continue reading

Support local journalism

If you value local news, make a gift now to support the trusted journalism you get in The Daily Herald. Donations processed in this system are not tax deductible.