AI Watermarks: Can We Trust What We See?

< Media >

 

Sua: Our topic today is ‘AI product watermark marking system’. It’s basically a system that requires every AI-created product to include a watermark that tells it has been generated by AI. Have you seen any content created by AI recently?

 

Seoyun: Yeah, like, a lot. There are so many AI-created videos on Instagram and YouTube. I can't tell if the product is made by a human or AI. Are there any AI product marking systems currently in place?

 

Yeonsoo: Yeah, I have some examples. In South Korea, SK Telecom’s “Adot” puts a small visible logo on AI-generated images so people can recognize they’re made by AI. The US announced the “AI Safety Initiative” in 2023 and is working with companies like ChatGPT and Google to add watermarks for AI content tracking. And the European Union has strict laws requiring clear labels on AI-generated content. These systems help everyone identify what’s AI-made and keep things transparent.

 

Chaeyun: They seem like good examples, but how could they be improved? Since only some countries have AI marking rules, I think a more convenient and universal system is needed.

 

Yulim: Yeah, especially because video platforms like Instagram or YouTube are used across the globe.

 

Yeyoung: I have an idea. Do you know how editing programs like Capcut or Kinemaster have their logos printed on the videos they edit?

 

Yeonsoo: Yeah, I’ve seen them a lot.

 

Yeyoung: Like that, the ‘AI product watermark marking system’ can make it mandatory for all AI-created products to have the logo of the AI video creating website or app that has been used to make the product.

 

Seoyun: Or, since consumers of the videos may not recognize the logos, words like ‘AI-made’, or ‘AI-creation’ can work instead.

 

Yulim: Pretty good. If some videos that didn’t print out the logos or whatever marks get caught, the creator can be punished, or the videos can be deleted.

 

Yeonsoo: Good idea, but will this system be effective? What could be the limitations?

 

Chaeyun: Hmm. Other than video contents, it would be impossible to add a watermark to AI-generated audio.

 

Sua: Right. And on video platforms like Instagram and YouTube, it’s hard to catch every production without a watermark and punish them.

 

Seoyun: Then how can we solve these problems?

 

Yulim: It is difficult to display visible watermarks for voice content, so it is necessary to reveal that it is an AI production by voice.

 

Chaeyun: And if the AI-generated application is set to display the logo unconditionally, there is no need to inspect all the work.

 

Sua: All very good. It seems like we’ve discussed enough ideas for the ‘AI product watermark marking system’.

 

Yeyoung: Since AI-related contents are gaining more attention, I really hope this system can be implemented as soon as possible.

 

Seoyun: I agree. So I guess that’s it for today. Everyone, great job!

 

By Staff Reporter Dong Chaeyun (1-6)

Kim Yeyoung (1-7)

Park Yulim (1-7)

Lee Yeonsoo (1-7)

Kim Sua (2-8)

Lee Seoyun (2-8)