OpenAI released ChatGPT in late November 2022 and it’s almost laughable to think about how excited many of us were over AI-generated text. Now OpenAI is taking its talents to the video world. Sora is the company’s latest tool and will allow users to create high-definition video just off of a simple paragraph of text. OpenAI offers some examples of AI-generated video on its site and the footage ranges from impressive to creepy to “I can’t believe AI can turn my words into visuals so easily.”
(It’s worth noting that Sora isn’t available to the public yet, so we haven’t tried it. Yet.)
How To Tell If Video Was Generated By AI
You may be worried about living in a world where anyone can make a hyper-realistic video with just a few sentences. What will this mean for political elections? What will this mean for deepfake scams? What will this mean for trolls who are all too eager to misuse a public figure’s likeness?
For now, Sora-generated AI videos contain a watermark in the bottom right corner. (The watermark looks like a waveform that transforms into OpenAI’s logo.) But it’s not hard to imagine ways people can hide this watermark if they wanted to (simply cropping it out, for example) or fail to see it altogether. Are there other ways to tell if a video was generated by AI?
Valentine Goddard is a Mozilla Creative Media Awardee and lawyer who specializes in AI. We we spoke to Valentine, their advice was to pay attention to the impossible. “Obvious tells include physical impossibilities like running in the wrong direction on the treadmill, [people with] too many fingers or tree branches that are too perfect,” says Valentine. The treadmill video on OpenAI’s site (in addition to looking like a fitness horror movie) hints at a lack of physics we often see in AI-composed footage. Valentine also notes to keep an eye on living things — humans, pets, etc. — and check if they look and feel genuine. “I also find the blankness in the eyes and slow speed of the characters very telling in videos. There’s something mechanical to the movement.”
Is there a way to spot misinformation within AI-generated videos?
It’s all fun and games when a nameless person is walking down the street or running backwards on a treadmill, but what happens when someone uses AI-generated video to impersonate President Biden?
Tools like Sora are changing the rules when it comes to spotting misinformation on your social media feed and in your group chats. The above advice still applies: look for OpenAI’s watermark, look for blank stares in the eyes and slow moving speeds and look for the impossibilities — how many fingers are present? Are they attached to a body?
Along with a finger check, consider the content’s origin. Is it a reputable news source sharing the video or is it Cousin Larry? If it’s Cousin Larry, maybe double check the video’s source.
AI-generated video is here. Now what?
As usual, it will be impossible to put the AI-video toothpaste back in the tube. So where do we go from here? Valentine recommends we all just take a beat and consider the ramifications of what could come. “I think the companies that deploy AI irresponsibly need to take a pause and use some of their profits to engage with stakeholders,” says Valentine. “As someone who works in AI policy, as an artist, as a citizen, I find it very irresponsible to keep on piling funding into technical developments when the capacity for society to adapt isn't equally funded.”
Valentine wants AI researchers to consider groups of people who may not look like them or share their circumstances. “Research shows that AI hurts more women — female entrepreneurs in AI, for example, make less money — and that women will have fewer opportunities and equal access to work in digital economies,” says Valentine. “They keep saying AI is dangerous, but what are their concrete plans to make it safe for women, for racialized, marginalized, LGBTQ, Indigenous communities?”
In the meantime, this tech is nearly here. Fingers may be hard to draw for professional artists and AI alike, but over time AI will only get better at smoothing out the impossibilities. Make sure you’re relying on trusted news sources when trying to suss out whether footage is fact or phony.
Was This Video Generated Using AI? Here’s How To Tell
Written By: Xavier Harding
Edited By: Audrey Hingle, Kevin Zawacki, Tracy Kariuki, Xavier Harding
Art By: Shannon Zepeda