Using AI in Political Campaign Ads

YOU’LL SEE A CAMPAIGN AD THIS ELECTION YEAR. IT’S PROJECTED THAT ABOUT $16 BILLION WILL BE SPENT ON POLITICAL ADS, 30% MORE THAN THE LAST PRESIDENTIAL ELECTION. BUT WHAT’S BEING WATCHED ESPECIALLY CLOSELY IS HOW A.I. IS BEING USED IN POLITICAL ADS FOR EXAMPLE, LAST AUGUST, THE REPUBLICAN NATIONAL COMMITTEE RELEASED THIS AD FEATURING AI GENERATED IMAGES OF A DYSTOPIAN FUTURE. SCOTT BABWAH BRENNEN IS HEAD OF ONLINE EXPRESSION POLICY AT THE UNIVERSITY OF NORTH CAROLINA’S CENTER ON TECHNOLOGY POLICY. IT’S SO NICE TO HAVE YOU IN STUDIO. THANK YOU FOR JOINING ME. OF COURSE. CAN YOU EVEN TELL WHEN A.I. IS BEING USED IN AN AD? NO, NOT NECESSARILY. THERE ARE VERY REALISTIC LOOKING DEEPFAKES OUT THERE THAT IT IS NEARLY IMPOSSIBLE TO TELL. THERE ARE A NUMBER OF TOOLS THAT CAN SOMEWHAT RELIABLY DETECT THOSE, BUT NOW, FOR EXAMPLE, MANY OF THE SOCIAL MEDIA PLATFORMS ARE RELYING ON SELF-DECLARATION. THEY’RE RELYING ON ADVERTISERS CHECKING A LITTLE BOX THAT SAYS THAT THEY INCLUDED GENERATED IMAGERY IN THE AD. SO IS THAT A GOOD THING? HONESTLY, THAT IS THE THAT IS THE KEY QUESTION IN CONSIDERING THE REGULATION HERE. THE PROBLEM IS WE WE DON’T KNOW IF THOSE LABELS WILL WORK OR WHAT EFFECT THAT THEY MAY HAVE. AND THIS GOES BACK TO THE POINT ABOUT HOW HARD IT IS TO KNOW WHEN SOMETHING IS AI GENERATED OR WHEN IT’S NOT. SO YOU’RE LEFT IN A REALLY DIFFICULT POSITION OF FLAGGING EVERYTHING THAT MAY BE AI GENERATED, AND THEN YOU MAY HAVE THE SITUATION WHERE, YOU KNOW, VOTERS SEE SO MANY LABELS THAT THEY BECOME MEANINGLESS OR MAY MAKE ASSUMPTIONS THAT CONTENT THAT IS NOT LABELED MUST BE TRUE EVEN WHEN IT’S WHEN IT IS IN FACT INCORRECTLY UNLABELED. NETWORKS GET AN AD AND THEN THEY DECIDE, IS THIS AD FACTUAL? IS THIS AD APPROPRIATE FOR OUR NETWORK? HOW DOES AI CHANGE ALL THAT FOR A NETWORK THAT CAN’T NECESSARILY DETERMINE IF EVEN THE PICTURES ARE REAL? YEAH, IT’S A DIFFICULT PROBLEM, BUT I THINK REALLY THE BIGGER CONCERN IS NOT WHAT HAPPENS ON THE MAJOR NETWORKS. POLITICAL ADS FOR MAJOR CAMPAIGNS. IT’S RATHER WHAT HAPPENS IN A DOWN BALLOT RACES. WHY? WHY IS THAT A BIGGER CONCERN? WELL, THERE’S THERE’S FAR LESS ATTENTION. THERE’S LESS JOURNALISTIC COVERAGE OF OF DOWN BALLOT RACES. IT’S POSSIBLE VOTERS MAY NOT EVEN KNOW THE NAMES OF THE CANDIDATES WHO ARE RUNNING. AND IN THAT CONTEXT, A SINGLE PIECE OF DECEPTIVE CONTENT. IT’S POSSIBLE, MAY HAVE MORE IMPACT, CERTAINLY, THAN IN A NATIONAL CAMPAIGN THAT’S ABSOLUTELY FLOODED WITH INFORMATION. WHEN YOU START LOOKING AT WHAT REGULATION THEORETICALLY COULD LOOK LIKE, WHAT ARE SOME OF THE SOLUTIONS THERE ARE TWO SOLUTIONS THAT REGULATORS NOW REALLY CONSIDERING. ONE WOULD BE TO BAN DECEPTIVE SYNTHETIC MATERIAL IN POLITICAL ADS. THE OTHER IS TO REQUIRE LABELS ON POLITICAL ADS. SO BAN OR LABELS. THE PROBLEM HERE IS THAT IT’S VERY POSSIBLE EITHER SOLUTION WOULD RUN INTO FIRST AMENDMENT CHALLENGES. POLITICAL SPEECH. TRUE OR FALSE, HAS HISTORICALLY RECEIVED THE HIGHEST LEVEL OF FIRST AMENDMENT PROTECTIONS IN THIS COUNTRY. AS A PERSON WHO STUDIES AI IN POLITICAL CAMPAIGNS, ARE YOU NERVOUS ABOUT WHAT THE FUTURE WILL HOLD? YES AND NO. I THINK IT ABSOLUTELY RAISES SOME VERY SERIOUS RISKS. DOWN BALLOT. I THINK THERE’S A HUGE RISK ABOUT AS CAMPAIGNS BEGIN TO INTEGRATE THESE TOOLS. AND JUST THE DAY TO DAY WORK OF CAMPAIGNING, THAT THEY MAY BE SORT OF REPLICATING OR AMPLIFYING THE BIASES THAT WE KNOW ARE BUILT INTO THESE, THESE AI MODELS. AND THEN THERE’S A LOT OF TALK ABOUT WHAT’S CALLED THE LIAR’S DIVIDEND. AND IT’S WHEN A CANDIDATE FOR EXAMPLE, IS ABLE TO DISCREDIT REAL CONTENT AND SAY THAT IT IS THAT IT IS AI GENERATED. SO YES OR NO, I MEAN, I THINK WE’RE IN A DIFFICULT POSITION WHERE, UH, THESE TOOLS ARE BEING USED WIDELY IN THE CAMPAIGN, BUT IT’S SO EARLY THAT WE DON’T HAVE GREAT EVIDENCE ABOUT THE IMPACT THAT THEY HAVE. SCOT