[Tech Thoughts] Realistic, but fake? It’s getting harder to tell the difference!

This is AI generated summarization, which may have errors. For context, always refer to the full article.

With video-generating AI tools gaining traction among AI enthusiasts and disinformation peddlers, it's become a lot easier to make videos that put words into imaginary people's mouths

Advancements in generative artificial intelligence have made the task of distinguishing reality from fiction harder, and it’s making the job of telling the truth even more difficult, it seems.

OpenAI’s Sora, Google’s Veo 3 and, more recently, Midjourney joined the ranks of video-generating AI tools.

With such tools gaining traction among AI enthusiasts and, unfortunately, disinformation peddlers, it’s become a lot easier to make videos that put words into imaginary people’s mouths.

For instance, this video of an “emotional support kangaroo” being denied entry onto a plane is cute, but entirely fake.

The source, as noted by Mashable, is an Instagram account called InfiniteUnreality, and it concerns itself with posting AI and VFX content.

But if you got it from somewhere else, like a stolen share without any sourcing or citations, you’d be hard-pressed to have enough facts on hand to reason through the “Is It Real?” process adequately.

What happens when you make the same output, but excessively more political? You get a can of proverbial brainworms messing with reality as we know it.

Amplifying fakery

Recently, Vice President Sara Duterte and Senator Ronald “Bato” Dela Rosa have gotten attention for sharing or otherwise amplifying such fakery.

Dela Rosa did so by sharing an AI-generated video of “students” opposing Sara Duterte’s impeachment and was promptly criticized for not being more discerning about sharing information online.

The video was apparently generated using Google’s Veo video-generation AI

Meanwhile, Sara Duterte followed this up by supporting Dela Rosa, saying she sees nothing wrong with sharing an AI-generated video supporting her as long as it’s not being used for profit.

Another Veo-generated video also made the rounds suggesting critics of the Dutertes are drug addicts, by having an “interviewee” in the generated video appear drug-addled and disheveled while apparently speaking in support of the impeachment of Sara Duterte.

Telling the difference

Unfortunately, there’s no easy way to tell an AI-generated video from a fake if an unscrupulous user has cropped out identifying watermarks.

That said, video-generating AI is still in its infancy, so you can always try to train yourself to spot inconsistencies that wouldn’t occur in reality, as if spotting optical illusions, except the illusion is the whole thing.

In late 2024, PCMag came out with a guide to help people spot AI videos, and it’s a helpful exercise in knowing some of the tricks. It’s not all-encompassing, but a worthy start to the process of training oneself, at the very least.

Sadly, this is only the beginning and we should expect — and be conscious and wary of — social media videos of this sort moving forward, even if it’s a tough order to tell what an AI-generated video designed expressly to appear real should resemble. – Rappler.com