IRYNA HUMENYUK, COHOST: New York voters went to the polls on Tuesday to vote in the primary elections. Joe Biden and Donald Trump won their respective party’s delegates, which comes at no surprise to those following politics.
GIULIA LEO, COHOST: But what might be surprising in this election cycle is that it will be the first in the era of mainstream Artificial Intelligence. That means uncharted waters, especially with the emergence of audio deepfakes that seem to be getting ever more convincing.
HUMENYUK: Pascal Hogue reports on managing the perils of AI during a Presidential election.
PASCAL HOGUE, BYLINE:
One mystery element this year is if – or how – artificial intelligence could be used to spread misinformation to voters in the election.
In fact, last month a Democratic consultant was caught using the technology to create fake Biden Robocalls ahead of the New Hampshire primary. The consultant said it was an act of “civil disobedience” to draw attention to the dangers of AI.
Will audio deepfakes have an impact on this election cycle? I did a little test with voters at the shops at Columbus Circle.
I played them this real quote from Joe Biden.
BIDEN (REAL)
No you haven’t. You’re a lying dogfaced pony soldier. You said you we’re. Now you’ve gotta be honest. I’m gonna be honest with you.
HOGUE:
Then I played them this fake quote created by the Guardian using AI.
BIDEN (AI)
My granddad used to say you’re short on ears and long on mouth.
HOGUE:
Do you think that’s real or AI?
GABRIELA RAMIREZ
I think AI.
DIEGO RAMIREZ
Real. Not AI made.
HOGUE:
So that one is actually AI.
DIEGO RAMIREZ
Oh really. Yeah I couldn’t really tell.
HOGUE:
Gabriela Ramirez correctly guessed the fake AI quote, but she says it’s getting harder to tell.
GABRIELA RAMIREZ
It’s pretty good. You can still kind of tell with the rhythm. It’s a little concerning that it’s very similar.
HOGUEAnd that’s what’s got Silas Jonathan worried. He’s a Digital Investigator at the Centre for Journalism Innovation and Development.
SILAS JONATHAN:
When deep fake first came, it was laughable because people could see that, ah, this is not a real human being.
HOGUE:
Jonathan says Audio deepfakes have been around for a while, but they were never particularly good. That made for great meme material. Like this clip of Trump and Biden playing Wii Sports that went viral last year.
JOE BIDEN (AI):
Golf is fun but the 9-hole takes too long.
DONALD TRUMP (AI):
Oh well Joe, are you gonna fall asleep or something?
HOGUE:
To be clear what you just heard were fakes. And audio Deepfake have become a lot better lately. Technologies like Parrot AI and Eleven Labs can mimic natural speech patterns and intonation.
JONATHAN:
AI has become so good in mastering people's voices that even for the most expert listener, hardly can you be able to distinguish what is real or not.
HOGUE:
Jonathan says this allows politicians to weaponize AI. To make statements and then later deny having done so.
JONATHAN:
Not just because it can be used to spread misinformation is because another way is that those who should be held accountable to what they say or what they admit can now say, Oh, this was not me. This was AI generated. This is not my voice.
HOGUE:
Jonathan says this means the process to detect AI misinformation has had to become a lot more sophisticated. When analyzing an audio clip, fact-checkers now take into account – politicians’ speech patterns, how they naturally pronounce words and even the acoustics of the room where the clip was allegedly recorded.
JONATHAN:
So in all of these things, it is not something that you do in a day or two. It is something that requires time commitment and a lot of energy
HOGUE:
So with the election approaching, listen carefully. Pascal Hogue, Columbia Radio News.