Artificial intelligence is sophisticated enough to create latent deepfakes from live video, AI developer Tom Graham said on stage at a conference in Vancouver on Tuesday. demonstrating sex and the thorny ethical issues it raises.
This ad hasn’t loaded yet, but your article continues below.
Graham, CEO of Metaphysic.AI, used the company’s technology to translate the image and voice of TED Talks host Chris Anderson into his own face in live video projected onto the screen above. rice field. , oddly enough, is thrilling.
“We are really pushing the boundaries of AI,” Graham says eerily in Anderson’s voice, who crouches beside him.
“I’m deeply sorry,” Anderson roared to a video image of his twins projected onto the screen.
Digitally manipulated videos that make celebrities speak a language that is not theirs are nothing new.
However, Graham describes Metaphysic’s business as “creating artificially generated content that looks and feels exactly like reality,” arguing that it will enable amazing advances in entertainment and human interaction. I’m here.
This ad hasn’t loaded yet, but your article continues below.
But the CEO also acknowledged a darker possibility stemming from the reality that, as Anderson put it, people “can no longer trust the evidence in our own eyes.”
“Worrying is a good instinct that everyone has,” Graham said.
But Graham believes safeguards need to be put in place to give people control over the data in their images, and his company still has potential.
And the possibilities and challenges AI poses is one of the big topics TED (an acronym for Technology, Entertainment and Design) focused on at its signing conference in Vancouver this week.
AI has permeated digital assistants like Siri and Alexa for years. But, as Anderson puts it, it has exploded in public awareness of technology’s “transformation moment.”
This ad hasn’t loaded yet, but your article continues below.
For better or worse, Anderson said, “the biggest in our lifetimes.”
Canada is beginning to work on ways to control the downsides of AI with the federal Artificial Intelligence and Data Act introduced into Congress last summer.
However, many believe that advances by AI developers are progressing at such a rapid pace that machine learning could come with the potential benefits of sifting through data and writing text in the same way humans do. I am concerned that the problem is becoming more difficult for the general public to understand.
Or realistically generate realistic images of humans to convince someone that the AI they’re looking at is a person they can recognize.
“If you can fake a liveit becomes a real problem for journalists,” said Alfredo Hermida, a professor of journalism and digital media expert at the University of British Columbia School of Journalism, Writing and Media.
This ad hasn’t loaded yet, but your article continues below.
“Because how can you prove that what you’re doing, what you’re broadcasting, what you’re streaming on YouTube is actually real? I don’t have the answer,” Hermida said. Told.
Hermida has no doubt that broadcasters will face the reality of live deepfakes “as technology gets better, faster and more powerful.”
“However, ethical concerns are not being discussed and addressed at the same speed that technology is being developed,” said Hermida.
But Greg Brockman, creator of the infamous ChatGPT application and co-founder of OpenAI, argued that AI developers need to address these ethical concerns as the technology develops. .
This ad hasn’t loaded yet, but your article continues below.
ChatGPT is a so-called generative AI known for its ability to sift through information from the internet and compile it into, for example, a compelling college essay.
Brockman said in a TED talk on Tuesday that AI “just looks different, it’s not what people expected.”
“One of the things I really believe deeply in is that getting AI right requires everyone’s participation,” said Brockman.
And for him, everyone needs to be involved in determining where AI tools fit in society and what the “rules of the road” will be for AI tools.
The TED conference attracted a wide range of speakers, from optimists like Graham and Brockman to skeptics and outright pessimists.
University of Washington AI scientist Yejin Choi argued that generative AI being developed by big companies is too big for less-wealthy researchers to access the technology.
This ad hasn’t loaded yet, but your article continues below.
“To make this powerful AI sustainable and humane, we need to teach it common sense norms and values,” she said.
At the last minute, TED asked doom-predicting engineer Eliezer Yudkowski to tell society, after 20 years of research, that society is ready for what will happen when AI becomes smarter than the humans who built it. Added warning that it is not
“My guess is that eventually we’ll face something smarter than us that doesn’t want what we want and doesn’t want to be perceived as valuable or meaningful. ‘ said Yudkowski.
Speaking at UBC, Hermida said the gradual approach doesn’t explain how the technology can be abused by bad actors, not just creating fake news and deepfake images, but also using AI to generate massive amounts of data. He added that it also did not address its ability to generate questionable information.
This ad hasn’t loaded yet, but your article continues below.
“And in this morass of disinformation, I don’t know what to trust when people don’t believe anything they read,” says Hermida.
Editorial picks
