MARY REICHARD, HOST: It’s Wednesday the 24th of April, 2024. You’re listening to WORLD Radio, and we thank you for listening. Good morning, I’m Mary Reichard.
LINDSAY MAST, HOST: And I’m Lindsay Mast. First up on The World and Everything in It: Washington Wednesday.
Today we’ll hear what options Congress has to combat disinformation from AI.
But before we get to that, we have more analysis of how Congressional Republicans are handling foreign aid. Here’s Washington Bureau reporter Leo Briceno.
LEO BRICENO: For over a year, Republicans in the House of Representatives have blocked funding new military aid to Ukraine. Late last week, a foreign aid package finally cleared the chamber, but with more Democrats than Republicans voting in favor.
While its passage has aggravated the most conservative factions of the House, U.S. Speaker Mike Johnson says he sees the package as the best way to prevent a darker alternative.
MIKE JOHNSON: The House has worked its will. I’ve said it very simply. It’s an old military adage, but we would rather send bullets to the conflicts overseas than our own boys, our troops.
The $95 billion bundle also includes help for Israel, some aid for Taiwan, and several other security-related provisions. It closely resembles a bill passed by the Senate in February with a few key differences: First, it’s roughly 20 percent smaller in terms of total cost; Second, the bill would allow the U.S. to seize Russian assets to pay for some of the aid; Third, a portion of the assistance is provided conditionally as a loan. And then separately, the package also contains language that would force the divestiture of TikTok.
But a handful of the most conservative Republicans in the House are angered by what’s not in the bill.
Georgia Rep. Marjorie Taylor Greene warned Johnson against passing any bill with Ukraine aid that didn’t also increase security at the U.S. southern border.
MARJORIE GREENE: He’s serving Ukraine first and America last. And that would be the worst thing to do for a United States House of Representatives Speaker—third in line to the president of the United States—to do nothing for American citizens.
Greene is one of three Republicans who have called for removing Johnson as Speaker.
Despite fierce opposition from these conservatives, other Republicans see the package as a fulfillment of American leadership internationally. Rep. Marcus Molinaro told me he represents a very strong Ukrainian-American, and Jewish-American district in New York.
MARCUS MOLINARO: I’m confident that my vote not only reflects what’s best for the people I serve, but also for this country. And it’s critically important as we are in a moment of significant crisis around the world that America shows stability and leadership to our allies and makes clear to our enemies that we mean business.
Tuesday’s vote marks the first time both chambers of Congress have passed military aid to allies abroad since 2022.
Reporting for WORLD, I’m Leo Briceno.
REICHARD: While foreign aid was able to clear Congress, Artificial Intelligence regulations haven’t. And with the general election coming up in November, Congress is weighing options to keep high-tech lies from influencing voters.
You may remember that back in February, a fake voice sounding just like Joe Biden called phones of registered Democrats in New Hampshire.
BIDEN AI DEEPFAKE: It’s important that you save your vote for the November election.
MAST: It’s unclear how much influence recordings like that have on primary voters. But lawmakers hope to prevent more sophisticated deepfakes in the months ahead.
Washington Bureau reporter Carolina Lumetta sat in on a recent Senate hearing about AI and election security. She brings us this report.
CAROLINA LUMETTA: As AI development surges, detecting deepfakes is getting harder. And time is running out to pass laws regulating AI before the general election in November.
RICHARD BLUMENTHAL: We need action to make sure that the public has notified when AI is used to distort images or voice.
That’s Senator Richard Blumenthal, a Democrat from Connecticut. He chaired a subcommittee hearing last week with industry leaders to discuss solutions. At the hearing, the witnesses played audio samples to see if the senators could identify the deepfake recording.
AUDIO: Hi my name is Richard Blumenthal, United States senator from Connecticut and I am a diehard Red Sox fan.
BEN COLMAN: And the second one please?
AUDIO: Hi, my name is Richard Blumenthal, United States senator from Connecticut, and I am a diehard Royals fan.
COLMAN: And the third and final one.
AUDIO: Hi, my name is Richard Blumenthal United States senator from Connecticut, and I am a diehard Yankees fan.
Turns out, all of those examples were AI-generated fakes. But one was obvious, Blumenthal said.
BLUMENTHAL: Anybody oughta know I’m not a Royals fan.
COLMAN: No comment.
Producing them is easy for anyone with less than one minute of real footage of their subject. And it costs an average of just 10 cents per minute. Here’s Ben Colman, CEO of Reality Defender, a deepfake detection company.
COLMAN: Anybody with internet access can create AI generated audio, video images or text to convince and persuade millions of people. This fake media can be distributed and shared instantly over social media platforms. The more incendiary the content, the faster it spreads.
So what can be done to keep deepfakes from taking over the internet?
One option is watermarking, where programs embed data into a video, image, or even audio to protect it from being regenerated. ResembleAI CEO and founder Zohaib Ahmed told the committee that his company requires verbal consent from the person whose voice is being cloned. Then they use an algorithm that encodes that watermark on the content. But this isn’t standard practice.
ZOHAIB AHMED: I think we need to hold a lot of generative companies accountable. You know, it shouldn’t be the case where someone can go in and, as unsophisticated as these attacks are at the moment, they’re largely there because they’re not even writing code. They’re actually going to online websites, sometimes even to Apple’s App Store, downloading an app.
Another solution could require disclaimers on all AI-generated content, especially when used for a political purpose, like a campaign advertisement. I asked Senator Blumenthal about this after the hearing:
BLUMENTHAL: As a candidate, I’m required to identify the ad that I put out there as to who pays for it, and out to be required also to say whether AI has been used and whether somebody’s image has been presented with their consent. So there’s no First Amendment issue here.
But scofflaws aren’t likely to respect disclaimer rules. Another solution is to quickly track and identify calls or media created using AI. A robocall of President Biden to New Hampshire voters was quickly identified only because it used a real Democratic official’s phone number on the caller ID. Here’s an exchange between Senator Amy Klobuchar of Minnesota and CEO Colman.
AMY KLOBUCHAR: Could you tell me, Mr Colman, how AI has this potential to turbocharge election-related disinformation and why we can’t just rely on the disclaimers and watermarks? I think you can do that for a set of it. I don’t think you should do it for all uses of AI. Tell us why it’s not enough to run a whole thing and then have a little label underneath when they think it’s the actual candidate but it’s not.
COLMAN: Yeah, we agree on that. To paint the larger picture, what we saw during the primaries was a single static deepfake, pre-recorded. Imagine a world where that was a one-to-one attack where instead of it being pre-recorded, it was actually live. Instead of being from one to many, it was one-to-one where it’s coming from your husband, your wife, your boss saying, “Hey Ben, we need you in the office at 6am. I know it’s a voting day.” Or to an election official,“Hey we’re moving your precinct. We need you across town three hours away.” And that’s where this is gonna go. It’s going to be a real-time custom deepfake in conversational language.
New Hampshire Secretary of State David Scanlan also testified at the subcommittee hearing…and he told me his state is launching voter education initiatives to teach them about AI threats like the Biden robocall. But they still need congressional help.
DAVID SCANLAN: I think that we need the technological assistance to help recognize it, stop it, and then take appropriate actions with whoever generated it.
Senator Blumenthal joined forces with Missouri Senator Josh Hawley, a Republican, last September to propose a framework that would create a new oversight body. AI developers would have to register with the group and license their work, as well as submit to audits.
But proposed legislation has failed to gain traction in Congress so far this year. Most of the subcommittee members did not even attend the hearing last week. Another bipartisan bill from Hawley and Klobuchar has been stalled in committee for seven months. Here’s Hawley:
JOSH HAWLEY: There are multiple bipartisan bills that are common sense bills that are ready to go… I mean, we can talk and talk and nobody has done a better job of servicing this issue and bringing facts into the public domain than the chairman next, but now it’s really time to vote, and I just call on the leadership of both parties in the Senate—both parties—the leadership needs to support an effort to get a vote. And I say an effort to get, but really they just need to schedule a vote. Let’s put these bills on the floor, and let’s vote. Let’s not allow these same companies that control the social media technology in this country, that control the news in this country, to also now use AI to further their hammer-hold on the United States of America and on our political process.
Reporting for WORLD, I’m Carolina Lumetta.
WORLD Radio transcripts are created on a rush deadline. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of WORLD Radio programming is the audio record.
