Recent criminal cases related to artificial intelligence He came out of high school in Maryland last week.police said the principal was framed as a racist through a fake audio recording.
This incident is another reason not only for everyone to do so. politician and celebrity — Experts say we should be concerned about this increasingly powerful deepfake technology.
“Everyone is vulnerable to attack, and anyone can commit an attack,” says Hany Farid, a professor at the University of California, Berkeley, who focuses on digital forensics and misinformation.
Here's what you need to know about the latest uses of AI to cause harm.
AI has become very familiar
Manipulating recorded audio and images is nothing new. However, the ease with which someone can change information is a recent phenomenon. So is the ability to spread quickly on social media.
Fake audio clips that impersonate you are an example of a subset of artificial intelligence known as generative AI. Create surreal new images, videos, and audio clips. In recent years, it has become cheaper, easier to use, and available to anyone with an Internet connection.
“Especially in the last year, online services have become available to anyone, and really anyone,” Berkeley professor Farid said. “And for free or for a few dollars a month, you can upload 30 seconds of someone's audio.”
Farid said those seconds could come from voicemails, social media posts or secret recordings. Machine learning algorithms capture the human voice. Clone speech is then generated from the words entered on the keyboard.
The technology will become even more powerful and easier to use, including video manipulation, he said.
What happened in Maryland?
Baltimore County officials said, Pikesville High School athletic director Dazon Darien said. Cloning the voice of Principal Eric Eiswart.
Police said the fake recording contained racist and anti-Semitic comments. The audio file was included in an email in some teachers' inboxes before being spread on social media.
Police said the recording surfaced after Eiswart expressed concerns about Darien's work performance and alleged misuse of student funds.
Authorities said the fake audio forced Eiswart to take a leave of absence while police guarded his home. The school was flooded with angry phone calls and social media was flooded with hateful messages.
Detectives asked outside experts to analyze the recordings. One said the court records “contained evidence of AI-generated and human post-editing.”
Berkeley professor Farid's second opinion found that “multiple recordings were spliced together,” according to records.
Farid told The Associated Press that questions remain about how exactly the recording was created, and he has not confirmed whether it was entirely generated by AI.
But Farid said the Maryland case still serves as a “canary in the coal mine” given the growing capabilities of AI, and the technology needs to be better regulated. .
Why is audio so important?
Much of the disinformation generated by AI is audio.
This is partly due to the rapid advancement of technology. And while the human ear can't always discern clear signs of manipulation, discrepancies in videos and images can be easily spotted.
Some people have cloned the voice of Children allegedly kidnapped Experts say ransom money is obtained from parents over the phone. Another impersonated the CEO of a company in dire need of funds.
In this year's New Hampshire primary, AI-generated robocalls He tried to dissuade Democratic voters from voting by imitating President Joe Biden's voice.experts warn The proliferation of AI-generated disinformation Targeting this year's election.
But experts warn that disturbing trends extend beyond audio, including programs that create fake nude images of clothed people, including minors, without their consent.singer Taylor Swift was also recently targeted..
What can you do?
Most providers of AI voice generation technology say they prohibit harmful uses of their tools. But self-enforcement is different.
Some vendors may require some type of audio signature or require the user to recite a unique set of sentences before the audio is duplicated.
Big tech companies like Facebook's parent company Meta and ChatGPT creator OpenAI only allow a small number of trusted users to experiment with their technology, citing the risk of abuse.
Farid said more needs to be done. For example, all companies should require users to submit a phone number and credit card so that files can be traced back to the person who misused the technology.
Another idea is to require recordings and images to be watermarked.
“It changes the sound in a way that is imperceptible to the human auditory system, but it changes the sound in a way that is discernible by downstream software,” Farid says.
Alexandra Reeve Givens, CEO of the Center for Democracy & Technology, said the most effective intervention would be law enforcement action against criminal uses of AI. Further consumer education is also needed.
Another focus is encouraging responsible behavior between AI companies and social media platforms. But it's not as simple as banning generative AI.
“Adding liability can be complex because in many cases there are potential active or affirmative uses of technology,” Givens said, citing translation and book reading programs as examples. Stated.
Yet another challenge is finding international consensus on ethics and guidelines, said Christian Mattmann, director of the Information Retrieval and Data Science Group at the University of Southern California.
“People use AI differently depending on what country they live in,” Mattman said. “And it's not just the government, it's the people as well. That's why culture is important.”
___
Associated Press writers Ali Swenson and Matt O'Brien contributed to this article.
