10th May 2024

TikTok Makes Steps In Creating Transparency With AI Content

By Vic Banham


A brand new update to TikTok means that we will now start to see AI-generated content automatically labelled, as well as new readable sources to help its users better understand how AI is being used, or misused across its platform and online.

What does this mean?

Any content that is now created on the app will be automatically tagged with ‘content credentials’, which is a digital watermarking technology created by the Coalition for Content Provenance and Authenticity. They will work together to establish new processes and implement them across their site.

This isn’t major news for TikToks as we’ve already seen major moves on the platform, by ensuring its creators must label any AI-created content accordingly before posting and implementing auto labels for its own AI effects.

Why was this a necessary move for TikTok?

While it’s a big step in ensuring safety and transparency across its network, there are still limitations within its data markers, so it may not be able to identify all AI content.

This being said, as we approach the upcoming election in the US, it limits the concern for deep fakes and false content, reducing the spread of misinformation across the platform.

TikTok is now also working with MediaWise to develop literacy campaigns which help its users understand and identify AI content, planning to release 12 new videos that highlight how tools and labels can add more context to content, as well as raise awareness around AI labelling.

How to identify AI-content

To protect yourself online against misinformation and AI-made content, it's important to identify its common characteristics, which include:

  • Inconsistencies within the text: Look for sudden changes in writing style, tone or vocabulary within copy. You may find the content goes from simplified language to using complex terminology that sounds unnatural.
  • Misplacements in imagery: AI-generated imagery will typically contain unusual placements of objects or people, showcasing distorted perspectives and inconsistent lighting.
  • Flawless Patterns or Symmetry: You’ll typically find AI produces images that have near-perfect symmetry or flawlessly repeating patterns, to a point where it feels uncanny or too perfect to be true.
  • Vivid or over-saturated colouring: AI imagery typically enhances colour within its imagery so it appears overly saturated and bright.
  • Uncanny valley: If you haven’t heard of uncanny valley, it is a spectrum of feeling that unsettles people who experience visual simulations that closely resemble humans but they aren’t quite as realistic as they appear to be. A lot of people tend to feel unsettled, or just have an off feeling.
  • Video deep fakes: There are many tells for a deep faked and AI-produced video, such as the face becoming momentarily smaller when there’s a blink, distortion when a head is turned
  • Misfeatured: AI-generated imagery can’t quite recreate human features yet. You’ll often find it produces too many or too few fingers, or an over-extension of limbs. There tend to be differences in the eyes too, particularly the reflections within them. Accessories will also disappear into the person as well, such as hats or glasses.
  • Misspellings: Text within the imagery is typically a massive giveaway because AI can’t produce correctly spelt words on imagery yet. You’ll often find misspellings or missing words.