The dangers and possibilities of new AI video technology Sora

Remember this time last year when the best thing Artificial Intelligence (AI) could conjure was a confusing — and mildly terrifying — video of Will Smith eating spaghetti? 

It's now a year later and AI generated videos are blowing away the internet's expectations. Last week, OpenAI — the company behind ChatGPT — unveiled its new text to video AI model called Sora.

With Sora, users can input text for a video they'd like to generate, up to a minute long. For example, users can write "A woman walks down a busy Tokyo street wearing a black leather jacket, a long red dress, black boots and sunglasses" and Sora will deliver a video that looks like this:

Upon closer inspection, it's easy to tell that the video above is AI generated. The pedestrians are walking far too slowly, the woman's face is wobbly, and the images reflected in the puddles on the ground don't look quite right. 

But from far away, the videos Sora is able to generate are hyper realistic and can easily fool the average person. And bear in mind, this is the worst that AI will ever be, so the technology will only get more refined. 

To discuss the potentials and pitfalls of OpenAI's Sora and what guardrails need to be put in place before releasing it to the public, futurist and tech commentator Sinead Bovell joins host Elamin Abdelmahmoud on Commotion.

We've included some highlights below, edited for length and clarity. For the full discussion, listen and follow the Commotion with Elamin Abdelmahmoud podcast on your favourite podcast player.

LISTEN | Today's episode on YouTube:

To Bovell, the Sora generated video that impresses her the most is of the grandmother who was blowing out candles on her birthday cake because it feels like she was there in the room with the grandmother.

"You felt like you were there. You've seen that grandmother and you've been to that birthday. It really transported you to that room," she says.

But as impressive as these videos are, Bovell says this tech is arriving at a rather dangerous time. With one of the biggest election years being just around the corner, deepfakes, misinformation and disinformation are going to get much more believable. 

"Now you can not only clone somebody's voice and create an image of them not doing something that never happened. You can have the deepfake video to go with it," says Bovell. "I think we all are a little scared by what happened to Taylor Swift with the AI generated imagery and what happens to many other victims. Now imagine that sort of content, but in video and also with some of the biases that are baked into these systems."

Before Sora is released to the public, Bovell says there needs to be stricter regulations and guardrails when it comes to technology.

"At the end of the day, the private sector has a fiduciary duty to shareholders, not to the general public first and foremost. And so that does come with inherent conflicts of interest and reasons for concern," she says.

"That is why the government and policies exist for these checks and balances. I don't think tech should be regulating itself."

Bovell says one barrier is when policymakers are not well versed in the tech they're trying to set up guardrails for. One clear example is when Meta and Facebook CEO Mark Zuckerberg testified in front of the senate recently. Senators didn't know Facebook runs ads to generate revenue.  

"The problem is, when it comes to industries like science and technology, there's a knowledge gap between people who are setting the policies and those who are building the technologies."

You can listen to the full discussion from today's show on CBC Listen or on our podcast, Commotion with Elamin Abdelmahmoud, available wherever you get your podcasts.

Interview produced by Jess Low