Deepfakes as a Form of Harassment

The term “deepfake” (“deep learning + fake”) seems to have originated when online communities of users began sharing fake celebrity pornography with each other. To create these, artificial intelligence software is used to produce incredibly realistic-looking fabricated content (e.g., photos and videos) intended to come across as real. Learning models are created by using computing power to analyze significant amounts of content (e.g., hours of video of a person, thousands of pictures of a person) with specific attention to key facial features and body language/position.

Next, what is learned is algorithmically applied to images/frames one might want to manipulate or create (e.g., superimposing lip movements upon original content (and dubbing in sound) to make it seem like a person is saying something they actually never said). Additional techniques such as adding artifacts (like “glitching” that appears normal or incidental) or using masking/editing to improve realism are also employed, and the resultant products are surprisingly convincing. If you perform a web search for deepfake examples, you’ll likely be surprised as to how authentic they seem. Below are some important points to know as you seek to safeguard your child from any possible deepfake victimization, and equip them to separate fact from fiction.

While deepfakes are becoming increasingly realistic as technology advances, detecting them is often done by looking carefully for certain information in the photo or video content (for example, eyes that don’t seem to blink naturally). It can be super helpful to zoom in and look for unnatural or blurred edges around the mouth, neck/collar, or chest. That is often where misalignments and mismatches between the original content and the superimposed content can be seen.

On videos, one can slow down the clip and watch for visual inconsistencies such as possible lip-syncing or jittering. Furthermore, keep an eye out for any moments the subject displays a lack of emotion when there should be emotion based on what is being said, seems to mispronounce a word, or is part of any other strange discrepancies. Finally, running reverse image searches on photos (or a screenshot from a video) can point you towards the original video before it was doctored. At that point, carefully compare the two pieces of content to determine which one has been manipulated. The bottom line is that you should trust your senses; when we slow down to look and listen very carefully at content, we generally can sense when something is amiss.

It is important to remind teens that everything they post online could be used to create a deepfake. On their social media accounts, they likely have created a library of content that others can access and manipulate without their consent. Their face, movements, voice, and other characteristics might be appropriated and then superimposed onto the likeness of someone else – someone engaged in a behavior that may cause extensive reputational harm. To facilitate dialogue in this regard, here are a few questions to ask them in a nonjudgmental and understanding way:

• Is it possible that you’ve accepted some individuals who, in time, may have a conflict or be in competition with you?
• Have you ever been hurt by someone you never thought would hurt you? Could it happen again?
• When new follower or friend requests come in, do you check their profile to make sure they are legitimate? Can you trust them?
• Have any of your friends’ posts ever been used by another in an unauthorized way? Could this possibly happen to you?

Deepfakes have the potential to compromise the well-being of teens when one considers the emotional, psychological, and reputational damage they can inflict. While aural, visual, and temporal inconsistencies may miss observation by the human eye, software is being refined to identify and flag any non-uniformity in image or video content. As these technologies continue to improve, parents, caregivers, and other youth-serving adults must raise awareness about the reality of deepfakes, and work to prevent the consequences of their creation and distribution. At the same time, regularly remind your teen that you are always there to help them find their way out of any deepfake situations (and, of course, any other online harm they experience).