Since the advent of the Internet, this advice has been true: “Don’t believe everything you see on the Internet.” Whether it’s a personal blog, a tweet, a YouTube video, or a TikTok, everyone has something to say here , and it’s hard to know whether he’s right or not (or even telling the truth).
But we are at a tipping point in internet literacy: generative AI has reached a scary point, and the technology is good enough to mimic the likeness of celebrities and make these “clones” say whatever they want. These deepfakes may not be convincing to those in the know, but what about the average social media user? It seems that we are quickly approaching a point where the general public is starting to believe that these fraudulent videos are real, and that is a scary thought.
There have been three eye-catching examples of this in the last week alone. The first is Tom Hanks: The actor posted a screenshot on his Instagram of a video promoting a “dental plan” with a speaker who looks like Tom Hanks if he were getting his teeth put back in. The video itself does not appear to be public and Hanks has not shared it or the name of the company promoting it. However, it is believed that a company or person faked Tom Hanks to sell a dental product. As I mentioned earlier, if you look closely at the image, you can already tell that something is “off” about his face, and if we had seen the video, it’s likely that the movement looked so stiff and eerie in many AI deepfakes have.
But Mr. Hanks was far from the only celebrity to address a deepfake issue this week. Gayle King, one of the hosts of CBS Mornings, posted a deepfake on her Instagram account explaining that she was being sent the video over and over again and that she had nothing to do with it. This time she shared the original video the deepfake is based on, an innocent video of King talking about having Bruce Springsteen on her show:
Careful observers will be able to tell that the first video is a fake: the lip movements don’t match the audio, and while the audio resembles King’s voice, it’s too stiff, as if she’s pretending to be a bad actress while reading Script. I’m sure any of us familiarizing ourselves with these limitations of deepfakes and generative AI will immediately notice the telltale signs, but I’m not convinced everyone will immediately see this as a fake.
Third, internet sensation Mr. Beast is currently going viral with a fake ad being shown to TikTok users, including me. It says: “Mr. Beast” congratulates the viewer on being one of the 10,000 selected users who won an iPhone 15 Pro for just two dollars. Lucky you!
The video in question is one of the better ones I’ve ever seen, although it also has obvious problems. Whoever made this made sure that Mr. Beast was more expressive when he spoke to make the whole interaction seem more natural. I think this could be effective for some, but again, it’s not 100%. When you watch the video and know it’s fake, all the imperfections come to the surface.
Even if you think it’s obvious that these examples are fake, as Marques Brownlee says: This is the worst thing this technology will bring. Deepfakes will continue to improve, with the ultimate goal of becoming indistinguishable from real videos.
If you’ve enjoyed any of the covers of AI songs making the rounds all over the internet, you know how good the technology is getting. The voice actor for Plankton from Spongebob should rightly be concerned about how excellent these covers are. like Plankton’s cover of Begin‘. Frank Sinatra might have died when Dua Lipa was two years old, but this AI cover of him singing Levitating is a blast.
This technology can even translate your speech and overlay it with your voice in real time. Although there are a variety of apps with this function, Even Spotify is testing it Translate podcasts into the voices of the presenters.
Currently, the best deepfakes are audio-only, and even then they still have their accuracy issues. But what happens when a bad actor can create a fake Mr. Beast ad that most people fall for? Imagine a truly convincing Mr. Beast saying directly to young and impressionable fans: “To enter my competition, all you have to do is enter your bank details so I can transfer the winnings directly to you.” Maybe the “competition” will take place in a “Mr. Beast App” that actually installs malware on your device.
Of course, there are even scarier scenarios to consider. We are facing a presidential election next year. How good will deepfake technology be by November 2024? Will anyone open TikTok before voting to watch a deepfake of Joe Biden saying his ultimate goal is to lock up his political enemies? Or maybe one of Donald Trump telling his supporters to go to the polls armed?
Be careful when watching videos online
Social media companies need to be more proactive in attacking these fake videos before they are shared with others, but we also have a role to play in all of this. Now more than ever, we need to be careful when casually scrolling and clicking on our vast internet. Just because you see a video of a “celebrity” saying something or endorsing a product doesn’t mean it’s real. Check the account it was posted on carefully: if it’s supposedly the real person, their account should be checked (unless it’s a useless platform like X, in which case all posts should be as be treated incorrectly unless the opposite is proven beyond doubt).
While we wait for deepfakes to get really good, there are still plenty of warning signs that something is illegitimate. Eye and mouth movements will seem strange, for one thing. Check out the Mr. Beast video: While they tried their best to make him expressive, his eyes are pretty blank in the first half of the clip. And while they matched the lip movements well, many deepfakes are still not good at that.
Many of these videos also appear in very poor quality. This is because increasing the resolution makes it clear how jerky the video is. Deepfakes are based on a real video of a person, whether they are a celebrity or not. The celebrity’s face is then superimposed over the video and manipulated as desired. It’s quite difficult to do this in high resolution without blending issues, so you’ll see layers blending into and separating from each other.
A healthy dose of skepticism is enough on the Internet. Now that generative AI is taking over, you should increase your skepticism as much as possible.
Comments are closed.