There is a danger that politics will be distorted because AI can write sentences
The AI that generates sentences is evolving day by day, and in October 2020, it turned out that the sentence generation AI ' GPT-3 ' was talking with users on the online bulletin board Reddit for a week without being noticed by anyone. did. WIRED, an IT news site, points out that the advent of AI has increased the risk of political distortion.
AI-Powered Text From This Program Could Fool the Government | WIRED
In 2019, the Government of Idaho solicited opinions on the health care system online as part of the process of obtaining federal approval for health care reform. As a result, a total of 1810 comments were collected, but it was later discovered that more than half, 1001 comments, were 'deep fake comments' generated by AI.
In response to this case, Harvard Medical Student Max Weiss conducted an experiment in which 108 subjects were asked to distinguish between sentence generation AI and human-written sentences. The correct answer rate obtained in this experiment was 49.63%, which was no different from the result of throwing coins and selecting them appropriately, and there were some deepfake comments that no one could see as being due to AI. From this result, Mr. Weiss concludes, 'The government's public comment site has been shown to be significantly vulnerable to mass postings from AI.'
'Weiss's experiments highlight that there are more serious threats than the 2019 case,' said Will Knight, a WIRED reporter who specializes in artificial intelligence. Because, Mr. Weiss used the sentence generation AI ' GPT-2 ' released in 2019 for the experiment, but in 2020, the successor model ' GPT-3 ' is out in the world. Because.
GPT-3, which can generate more sophisticated sentences than GPT-2, which was regarded as ' too accurate and dangerous ', has already had a great impact on society, such as the article created being ranked first on the news site. It's starting to exert.
An article written by 'GPT-3', which was regarded as dangerous as 'creating sentences with too high precision', has risen to the top of social news sites --GIGAZINE
'New technology is always a sword,' said Renee DiResta of the Stanford Internet Observatory , which studies technology abuse on the Internet . In the future, more government sites will be the target of deep fake comments by AI, he said.
Experts also say that AI threats are especially dangerous in today's difficult situation. Dr. Joan Donovan, Principal Investigator at Harvard Kennedy School, told WIRED that 'when people's emotions are worn out, people start jumping at convenient sweet words rather than complicated truths. He pointed out that even simple sentence generation AI is becoming a society that can easily deceive people.
in Software, Posted by log1l_ks