Unmasking Deepfakes: How to Spot Faux Media

Published on Tuesday, 28 May 2024 at 2:43:35 PM

Have you ever come across a video of your favourite celebrity saying something outrageous, only to discover later that it was entirely fabricated? Or perhaps you’ve received an urgent email that claims to be from your boss, but something felt off. Welcome to the world of deepfakes—a rapidly evolving technology that leverages artificial intelligence (AI) to create faux media, often in the form of videos or audio recordings. For example, recently Hollywood film star Scarlett Johannsen launch legal action amid claims that OpenAI copied her voice. While the outcome is pending in the case of Scarlett Johannsen, it can be said meticulously manipulated  digital creations can  appear remarkably real, yet they are not.

The Deepfake Landscape

  1. Face-Swapping Deepfakes

The most common type of deepfake involves seamlessly superimposing one person’s face onto another’s body in a video. These can be highly convincing, especially when using high-quality footage and sophisticated AI algorithms. To spot them:

  • Look for inconsistencies: Pay close attention to lighting, skin tones, and facial expressions. Do they remain natural and consistent throughout the video? Look for subtle glitches, such as hair not moving realistically or slight misalignments around the face and neck.
  • Check the source: Consider where you encountered the video. Was it on a reputable news site or an unknown social media page? Be cautious of unverified sources.
  • Listen closely: Does the voice sound natural? Does it match the person’s typical speech patterns? Incongruences in voice tone, pitch, or accent can be giveaways.

In a recent case, a deepfake video emerged showing a prominent politician endorsing a controversial policy. The video spread rapidly on social media, leading to confusion and heated debates.

 

  1. Deepfake Audio

This type generates synthetic voice recordings that mimic a specific person’s speech patterns and intonations. Scammers use these to create fake audio messages or make it seem like someone said something they didn’t. To identify them:

  • Focus on audio quality: Deepfake audio may sound slightly robotic or unnatural compared to genuine recordings. Pay attention to unusual pauses, inconsistent pronunciation, or strange emphasis.
  • Compare content: Does the audio align with what the person would typically say or the context in which it’s presented? Consider if the content seems out of character or contradicts known facts.
  • Seek verification: Look for independent evidence to support the claims made. Approach unverified audio with healthy skepticism.

A company CEO reported they received a deepfake voicemail instructing them to transfer company funds urgently. Fortunately, the CFO recognised the inconsistencies in the voice and prevented a significant financial loss.

 

  1. Text-Based Deepfakes

An emerging type of deepfake generates written content, such as social media posts, articles, or emails, mimicking the writing style of a specific person or publication. These can be particularly dangerous, as scammers use them to spread misinformation or impersonate others online. To detect them: 

  • Read critically: Pay attention to writing style, vocabulary, and tone. Does it match the person’s usual way of writing? Look for unusual phrasing, grammatical errors, or inconsistencies.
  • Check factual accuracy: Verify information against reliable sources. Don’t rely solely on the content itself.
  • Beware of emotional triggers: Content evoking strong emotions (fear, anger, outrage) may be manipulative.

 

  1. Deepfake Videos with Object Manipulation

This advanced type goes beyond faces and voices, manipulating objects within real video footage. For instance, it can alter the appearance of objects or change their behaviour.

Remember, staying informed and vigilant is crucial in today’s world where deepfakes can deceive even the most discerning eye. If you encounter suspicious content, always verify and question its authenticity!

Need help with understanding AI and implementing detection tools? Speak to the team at Integrated ICT 6374 8200 or email hello@integratedict.com.au

 

Back to All News