Deepfakes are videos that glue one person’s face onto another’s body, making the former look like they’re saying or doing something that they never actually did — even if it’s something as harmless as Tom Cruise talking to the camera and hitting a golf ball. They’re hard to spot just from watching the video, but here’s the good news: you don’t actually have to watch the video to know you’ve encountered a deepfake.
The Tom Cruise deepfakes (a few brief videos posted to TikTok on an account called @deeptomcruise) were convincing because they applied Tom Cruise’s face to an actor who had already built a career as a Tom Cruise impersonator. These videos were professionally produced, and honestly, there was no way to tell they were deepfakes just by looking at them. Should we be worried? Are we now living in a world where anything can be faked?
I mean, yes, but we have been for years. For comparison, think about photoshopped images. Sure, a bad one can be obvious. But normally we don’t scrutinise photos to figure out whether they’ve been altered. Instead, we’re just aware that it happens pretty commonly. Do you need to even look at a celebrity magazine cover to know the photo is probably altered? You do not.
It’s the same with deepfakes: What matters is the context of the video and where it came from, not the specifics of how the pixels move. An expert on videos used for activism told Vice that the bigger problem is that “We live in a world with a lot of shallowfakes — simple, miscontextualised, or edited videos.”
So how can we spot fake videos without poring over every pixel? Mike Caulfield, an expert on digital literacy, tweeted about this recently, offering two frameworks for spotting fake anything, and they work well for deepfakes.
The first is SIFT, a concept he explains here. There are four steps:
- Investigate the source.
- Find better coverage.
- Trace claims, quotes, and media back to the original context.
The first step can be applied as soon as you notice yourself getting sucked in, worrying, or wondering — or when you realise that you’re spending a little too much time scrutinizing an image for telltale signs of deception.
The rest of the steps are commonsense questions that should have clear answers if the thing you’re looking at is real. Caulfield pointed out an example from last year of a fake reporter with a convincing looking social media presence. Lots of details checked out. The obvious sign of fakery, though? She said she wrote for Bloomberg, but no articles by her appear on Bloomberg’s website.
These are basics! I am sincerely sorry if the internet the past several years taught you you were a budding video forensics expert. But the simpler approach is to apply the who/what/when/where/why you learned as a fourth grader, and not share until those questions are answered.
— Mike Caulfield (@holden) March 2, 2021
The Five Pillars
Another checklist for checking out a video or photo is called the Five Pillars of Visual Verification, from anti-disinformation organisation First Draft.
As they put it:
The nice thing about teaching verification is that it is easy to break down. That’s because whether you are looking at an eyewitness video, a manipulated photo, a sockpuppet account or a meme, the basic checks you have to run on them are the same:
Provenance: Are you looking at the original account, article or piece of content?
Source: Who created the account or article, or captured the original piece of content?
Date: When was it created?
Location: Where was the account established, website created or piece of content captured?
Motivation: Why was the account established, website created or the piece of content captured?
As with SIFT, there’s no need to zoom in on the video or image itself; instead, you zoom out to judge its context in the real world. That will not only give you hints as to whether it’s real, but help you think through why it exists and who stands to benefit from its being shared.
Many of us already use these techniques, of course. For a perfect example, see this Lifehacker piece, in which Nick Douglas investigated the supposed trend of hipsters wearing tiny scarves around their ankles. The image was a photoshop, but the real tell was that if you follow the links (provenance), you find that it came from an Italian website (source) that publishes humour articles (motivation). It was never a real trend.
The SIFT and Pillars techniques work just as well for faked and mislabeled news images as they do for silly stories about ankle scarves. (Remember when we talked about those “breaking news” accounts on social media that circulate misinformation — sometimes by accident, sometimes on purpose — by tying it to current events?) Deepfakes and shallowfakes are out there, and you can spot them without any special training — just your own common sense, if you choose to use it.