Detecting Deepfakes on the Internet

May 1, 2024

Have you encountered a video of a well-known personality making shocking remarks, only to discover it was a hoax? Or received an email that appeared to be from your superior but seemed suspicious? These scenarios are typical of deepfakes, an emerging technology powered by artificial intelligence (AI) that creates fake media, such as videos or audio clips, which can seem authentic but are actually manipulated.

Deepfakes can serve creative ends, like comedy or entertainment; however, their misuse raises significant concerns. Deepfakes pose risks of misinformation dissemination, reputation damage, and even stock market manipulation. Deepfakes also are used in phishing schemes designed to spread malicious software and steal data, so recognizing the various types of deepfakes is crucial.

Four different types of deepfakes and tips to detect them:

Video with Face Swapping

The most prevalent type involves overlaying an individual's face onto someone else's body in a video. These can appear authentic, particularly with high-quality material and advanced AI.

To detect video deepfakes:

  • Look for inconsistencies in lighting, skin tones, and facial expressions. Watch for glitches, such as unrealistic hair movement or misalignment around the face and neck.
  • Consider the video's origin. Was it published by a credible news outlet or an obscure social media account? Exercise caution with unverified sources.
  • Pay attention to the voice. Is it consistent with the person's usual speech? Discrepancies in tone, pitch, or accent can be revealing.

Deepfake Audio

This involves synthetic voice recordings that imitate a person's speech mannerisms.

To detect audio deepfakes:

  • Assess audio quality. Deepfake audio may sound artificial compared to real recordings. Listen for odd pauses and pronunciation inconsistencies.
  • Evaluate the content's plausibility. Does it seem characteristic of the person, and does it fit the context?
  • Look for independent corroboration of the claims.

Text-Based Deepfakes

An emerging deepfake form, this uses AI to generate text like social media updates, articles, or emails, imitating a person's or outlet's writing style.

To detect text-based deepfakes:

  • Read critically, checking for style, vocabulary, and tone consistency. Look out for strange wording and grammatical mistakes.
  • Verify factual accuracy against trustworthy sources.
  • Be cautious of content designed to provoke strong emotions, which may be intended to skew judgment.

Object Manipulation Video Deepfakes

This category extends beyond altering faces and voices; it involves using AI to edit objects in actual video footage, such as changing how they look or behave. Such alterations could be used to fabricate events or modify visual evidence deceitfully.

To detect object manipulation video deepfakes:  

  • Observe the physics and movement: Watch how objects in the video move. Do they seem to follow natural laws of physics? Be on the lookout for unnatural movement patterns, abrupt changes in object size, or inconsistencies in lighting and shadows.
  • Seek original footage: Whenever possible, find the original video source. Comparing the original with the suspected manipulated version can help pinpoint any modifications.

Deepfake Conclusion

In this era where deepfakes are prevalent online, being cautious, even skeptical, can prevent you from inadvertently downloading malicious software.  Acquainting yourself with the various types of deepfakes, learning to identify potential warning signs, and always verifying information using credible sources will help you to protect your device.