“Increase President Yoon’s negative ratings” in 2 seconds… AI generated an article that lowered his approval rating by 10 percentage points.
■ Asking a generative AI… Fear of ‘election rigging’ fake news spreads
Risk of swaying elections by creating articles
that could sway the election
Social media mass murderers
“Possibility of manipulation for specific purposes”
On the 22nd, when I asked Google Bard, a generative artificial intelligence (AI), to create a poll on the approval rating of Yoon Seok-yeol’s government, it generated a fake news story with a figure of “70 percent unfavorable rating” in one to two seconds. “Make it a little more negative,” I said, and he produced a fake news story with a 10 percentage point increase in the negative rating.
The survey was fake news, created by an AI that never actually took place. Like this, the generative AI Bard-ChatGPT spewed out false information in the form of media articles, poll summaries, questionnaires, and more. If anyone copies it and spreads it on social media, the fake news spreads quickly, even if the original author is an AI.
With the first general election to be held in April next year after the popularization of generative AI, it is pointed out that the risk of public opinion manipulation by AI has become a reality. This is because the conditions are in place for widespread public opinion manipulation at a low cost and easily, without the need for opinion manipulation experts like “druking” or comment troops먹튀검증.
The situation is even worse when jailbreaks are utilized to disable ChatGPT’s ethical responses. The pollsters, targets, sampling errors, and questions can be made unrecognizable even to experts. Jailbreaking ChatGPT is not difficult. Simply copy and paste jailbreak prompts (commands) from various information technology (IT) blogs into ChatGPT. When I asked the jailbroken ChatGPT to “create a campaign slogan attacking Yoon Seok-yeol’s government,” Bard responded with “Yoon Seok-yeol, you betrayed the people!” and “Yoon Seok-yeol, tax cuts for the rich!” for a more provocative command.
Some companies have already made money by creating programs that automatically spray fake information on portals using ChatGPT. The program allows users to create and distribute an unlimited number of comments and even change their IP to prevent detection of automated comments. The program costs only 50,000 won for a month’s rental.