A world where AI-generated videos and images can dupe the public on a large scale — a fear of the “Godfather of AI” — has become a reality, according to an artificial intelligence writer and podcast host.
“That moment is already here,” said cognitive scientist Gary Marcus, who hosts the AI-centric podcast, “Humans vs Machines with Gary Marcus.” “The techniques will only get better and better over the coming years, but they’re already good enough that they can probably fool at least some of the people some of the time.”
WATCH: AI EXPERT SAYS AI CONTENT ALREADY REAL ENOUGH TO DUPE PUBLIC
Computer scientist Geoffrey Hinton, who is widely considered the “Godfather of AI” and helped develop systems used in software like ChatGPT, recently told The New York Times he feared AI-generated photos, videos and text will soon flood the internet. The average person, as a result, will “not be able to know what is true anymore,” he said.
Rapid AI advancement over the past few months has made it easier to create deepfakes, hyperrealistic but fake content. Marcus pointed to a recent Republican National Committee ad arguing what the world would look like if President Biden. It used AI to generate realistic visuals of China invading Taiwan, 80,000 illegal aliens surging the southern border and skyrocketing crime and drug use forcing San Francisco to declare martial law.
“The RNC made a pretty compelling ad for the 2024 election that used deepfakes and looked pretty plausible,” said Marcus, who led Uber’s AI labs from 2016-2017 and co-authored the book “Rebooting AI: Building Artificial Intelligence We Can Trust.”
Marcus said an RNC anti-Biden campaign ad created entirely from AI imagery supported his belief that we’re ‘already’ living in a time when AI deepfakes could dupe the public. (SAUL LOEB/AFP via Getty Images)
Other viral deepfake examples include “Heart on my Sleeve,” an AI-crafted song that cloned the voices and styles of Drake and the Weeknd, and a fake photo of Pope Francis sporting a large puffer coat.
Meanwhile, NewsGuard, which rates news and information sites, recently identified 49 news and information sites that were either totally or mostly AI-generated. The websites, according to NewsGuard, used a hallmark of text created by artificial intelligence: “bland language and repetitive phrases.”
NewsGuard said it identified nearly 50 websites spanning seven languages purporting to report the news were either entirely or mostly generated by AI. (STR/AFP via Getty Images)
Marcus said the May 1 report further supported his belief that Hinton’s fear, that AI-generated content would regularly deceive the average person, “is basically here.”
To watch the full interview, click here.
To listen to Marcus’s new podcast, click here.