AI tools that are undermining our own eyes and ears – POLITICO

This article is part of a series. Bots and voting: How artificial intelligence is reshaping elections around the world.

Have you ever seen a deepfake? More importantly, can you tell the difference between these AI-generated images, audio clips, and videos and the real thing?

As more than 2 billion voters in 50 countries prepare for national elections in 2024, this issue, and the ability of such deepfakes to distort the decisions of potential voters at the polling stations, has once again It has become more important than ever.

Case in point: In recent months, more people on According to a study by the Brookings Institution (Verification Initiatives) on the platform.

Politico decided to test you.

POLITICO is powered by Midjourney, an AI research lab with technology that can create lifelike images simply by entering suggestions into its online platform, to create a series of real-world and artificial intelligence-generated images. Collected. Tests repeated around the world have shown that, on average, digital forgeries can be detected by comparing them to legitimate images about 60 percent of the time, according to tech company officials POLITICO spoke to.

While the technology is still in its infancy, politicians, policymakers, and disinformation experts alike are excited that anyone, including POLITICO reporters, can create such realistic images with just a few clicks of the keyboard. has concerns.

In the coming months, if AI puts such power in the hands of anyone with a laptop, an internet connection, and $50 to access these powerful tools, Such deepfake political content could flood people's social media feeds.

How well does it work? Take the quiz before reading the rest of the story. (There are spoilers below!)

Who wants to be a clone?

Of all the potential deepfake threats this year, audio is the one that worries cybersecurity and disinformation experts the most.

So far, almost all controversial AI-generated images have been debunked within hours. This is largely thanks to the power of social media, which allows us to quickly crowdsource errors in these photos that are often unrecognizable in any other way. Big tech companies and independent fact-checkers have also made it a priority to uncover and remove harmful politically motivated falsehoods.

But audio, particularly the grainy AI-powered clip that was unsuccessfully used to smear UK Labor Party leader Keir Starmer, remains uncharted territory. Because of the disconnect between what people hear and what they see, inflammatory deepfake audio clips often fool people into thinking they are legitimate.

To test this theory, POLITICO saw how easy it is to generate deepfake audio clips using off-the-shelf technology that costs less than $50 in total to purchase. Initially, we intended to clone real politicians. But since such falsehoods are legally dubious and pose a direct threat to this year's election cycle, we decided to imitate the POLITICO reporter's voice instead.

You can decide for yourself whether these AI-generated clips are good enough to fool you.

AI Biden vs AI Trump

The next frontier for AI deepfakes is video, especially content that can interact with humans in real time.

And when it comes to politically motivated AI-powered photography, a Soviet-era business district near the German-Polish border has become ground zero to demonstrate how the technology is evolving.

There, within an activist group known as the Singularity Group, researchers created an ongoing, real-time online video debate between an AI-powered Joe Biden and an AI-generated Donald Trump.

The project has been running for almost nine months and uses so-called open source technology, meaning AI models that are freely accessible to the public. This will allow anyone to enter debate questions through the Amazon-owned streaming service Twitch, and the Biden/Trump bot will launch and calculate the answers through Singularity's his AI system, allowing politicians to voice and You can imitate an image and spit it out.

“Deepfakes are a real concern,” said Reese Reisen, one of the activists behind the project, which, importantly, has been classified as a parody on Twitch. “I wanted to focus on politicians to get people's attention.”

POLITICO asked the fake Biden and Trump multiple questions in a real-world debate. Most were too racy or profane to be published. That's not surprising, considering the AI ​​system has been trained for almost a year on people asking random questions on the internet.

However, the two videos below have the least amount of graphics. Is the technology perfect? Absolutely not. But this is a snapshot of where things are heading.

Ask the bot a question

We asked fake Donald Trump and Joe Biden some real-world debate questions. They answered:

Quick question #1: Which Disney character best represents your political opponent and why?

---