Boston Sunday Globe

Spotting a deepfake: tips and tells

- MALCOLM GAY

Deceptive deepfakes seem to be everywhere these days, making it harder than ever to sort the true from the false. While there’s no silver bullet to address the threat posed by generative AI, here are a few techniques to guard against disinforma­tion.

1. Take your time, look closely

As humans, we are hardwired to focus on the face. But while many of today’s AI-image generators can create lifelike faces, it pays to spend a little time looking at other aspects of an image. AI is apt to cut corners and that’s where things can get weird. Look at the background. Does it make realworld sense? Does everything line up? How about people other than the image’s primary subject? Is there a phantom limb? Maybe a sixth finger?

2. Pay particular attention to hands and limbs, and beware of smoothness

One early tell for deepfakes were misshapen hands: Some looked like root balls; others had an impossible grasp on an object or maybe an extra digit sprouting from the palm. Newer AIs are better at rendering these complicate­d body parts, but it’s still worthwhile to inspect them closely. The same goes for limbs. Is there an extra one floating around somewhere? If an arm is draped over someone’s shoulder, do the visible parts line up? And what about the overall look? Does it feel too processed, as if the entire scene has been airbrushed?

3. Funny reflection­s, shadows, and blurs

This one’s a little harder, but try to think about the light source. If it seems to be from above, do the shadows make sense? Reflection­s are similar — do they come from the right source? Are they too uniform?

4. Keep an eye out for patterns, symmetry, and nonsensica­l words

In the real world, if all goes well, pinstripes are parallel, tiles are uniform, and words make sense. Patterns and intersecti­ng lines can be challengin­g for AI, so check whether shapes meant to be consistent are, in fact, uniform. Same goes for phrases on clothing and other text: Are those actually letters, or just gibberish?

5. Face swaps, lip syncs, and strange expression­s on video

Video deepfakes often manipulate only a person’s face, leaving the rest of the original scene intact. Look closely for abnormal shadows and lighting around the eyes. Do the facial features move naturally? How about the mouth? Does it match up with the face and audio?

6. Consider the source

Who’s sharing the media with you? Is it a trusted source known for verifying informatio­n? If not, why are they sharing it? Are they trying to affect your thinking in some way? How?

7. Is the media consistent with what you already know?

Does the video or other media make sense? Is it in keeping with what you know about the subject, or is it somehow outlandish?

8. Take your time

Let’s face it: Most of us aren’t going to become experts in digital forensics. So if a video or photo seems off (or too good!), maybe reserve judgement — and hold off reposting — until there’s more context.

 ?? LANE TURNER/GLOBE STAFF ?? The MIT Museum’s exhibit “AI: Mind the Gap” looks at deepfake video technology. Visitors can watch videos and guess if the images are real or fake.
LANE TURNER/GLOBE STAFF The MIT Museum’s exhibit “AI: Mind the Gap” looks at deepfake video technology. Visitors can watch videos and guess if the images are real or fake.

Newspapers in English

Newspapers from United States