A woman walks down a Tokyo street at night wearing a red dress and a black leather jacket.
But if you watch the video closely, the woman's right leg suddenly becomes her left leg, a giveaway glitch that indicates the clip was created using artificial intelligence.
It's called Sora, a new program from OpenAI, the same company behind ChatGPT. It works by taking a user prompt, such as "tour of an art gallery with many beautiful works of art in different styles" and quickly rendering a completely photorealistic scene. The Sora clips can be realistic, or fantastically bizarre.
"Clearly, it's by a long way the best video generation model that's been created," said Ed Newton-Rex, CEO of nonprofit Fairly Trained.
But what has some AI experts like Newton-Rex concerned isn't the quality: It's the worry that Sora and other AI programs only get better by scraping the internet for copyrighted material to make something new.
"These technologies, many of them — not all of them, but many of them — are exploiting human creators' work without those creators' permission and without paying them," Newton-Rex told Scripps News.
OpenAI won't say how it trains Sora, but on its website demonstrates how the program improves over time, greatly improving video renderings of a dog, for example.
Sora is not yet available to the public. The company is asking select, so-called "red teams" to look for potential harms or risks. "We are also granting access to a number of visual artists, designers, and filmmakers to gain feedback," the company says on its website.
Concerns over AI taking the jobs of artists is one of the main reasons Hollywood writers and actors recently went on strike.
"These tools will help creators," Newton-Rex says. "There will be people who use these tools to do amazing things, but at the same time, the same model that, you know, can assist one video creator is going to put, you know, 10 more out of work."
And there is growing concern AI may be used to generate deepfakes to sway elections, like a recent robocall before the New Hampshire primary featuring a phony President Joe Biden.
Newton-Rex believes new laws should require AI programs to disclose how they are trained. His new nonprofit company, Fairly Trained, works with companies pledging to buy — not just take content.
"We certify companies that don't go and claim that they can use whatever they like, and it falls under the fair use exception to copyright law," says Newton-Rex.
OpenAI says it's focused on safety and ways to detect AI-generated videos, knowing the ethics, legality, and potential of AI are mammoth issues that aren't going away.
Tech companies sign pact to fight against AI election disruption
Adobe, Amazon, Google, IBM, Meta, Microsoft, OpenAI, TikTok and X adopted new policies for handling election-related AI fakes.
