TikTok exploring AI-Influencer and Meta starts labeling AI content

TikTok is toying with AI-influencers to help brands advertise on their platform

TikTok is exploring the development of AI-generated influencers to enhance advertising on their platform, allowing brands to create custom virtual characters for product promotion.

Although still in the experimental phase, this new technology could revolutionise advertising by creating more personalised and versatile ad experiences, targeting diverse global demographics with hyper-tailored messages.

How does it work?

According to the The Information the feature is supposed to work similar to an interaction with an LLM like chatGPT.

Advertisers create video scripts by typing prompts about their products and marketing message that can be edited and rendered into synthetic videos featuring an AI-Influencer of choice.

Why is it important?

Brands actively deploying synthetic models and influencers is most likely a trend proliferating in the coming months.

It will be interesting to see how the disclosures will be implemented and whether these AI influencers can truly replicate the connection and trust that human influencers build with their audiences.

However, in China, where live-stream shopping is hugely popular, AI-Influencers have already partly taken over, streaming relentlessly 24/7.

“Made with AI” - Meta will start labeling GenAI content on Facebook & Instagram starting May ‘24

Meta's updated policy will soon label AI-generated and manipulated media across its platforms, including Facebook and Instagram starting May 2024. The labels will identify various forms of digital content with focus on video and pictures. Source

This decision, influenced by recommendations from the Oversight Board and extensive consultations, emphasizes transparency by providing context (labeling) rather than removing content.

Will this actually work?

While the intention and idea is a step into the right direction, it will be very challenging for Meta to correctly label content as “Made with AI” as video detection tools have been notorious unreliable and provenance signals from standards like C2PA or metadata embedded in EXIF data are still being stripped when importing pictures or videos into any Meta app.

Video as evidence dismissed due to suspected tampering with AI in tripple murder case

A Washington state judge recently ruled against the use of AI-enhanced video evidence in a murder trial, marking what might be a first-of-its-kind decision in the U.S.

While the video is not suspected to be entirely deepfaked, the judge highlighted concerns about the reliability of AI technology, which uses opaque methods that can misrepresent events by adding or omitting visual data from the original footage. Source

Is this a new era?

The decision triggers a debate over the forensic and legal use of digital videos, photos and audio recordings as evidence as deepfake technology improves.

This is likely an uphill battle that can only be fought using technology like advanced AI-detection and tamper proof content provenance methods.

Interesting commentary on this topic → here.