Meta Is Tagging Real Photos As 'Made With AI,' Says Photographers (techcrunch.com) 25
Since May, Meta has been labeling photos created with AI tools on its social networks to help users better identify the content they're consuming. However, as TechCrunch's Ivan Mehta reports, this approach has faced criticism as many photos not created using AI tools have been incorrectly labeled, prompting Meta to reevaluate its labeling strategy to better reflect the actual use of AI in images. From the report: There are plenty of examples of Meta automatically attaching the label to photos that were not created through AI. For example, this photo of Kolkata Knight Riders winning the Indian Premier League Cricket tournament. Notably, the label is only visible on the mobile apps and not on the web. Plenty of other photographers have raised concerns over their images having been wrongly tagged with the "Made with AI" label. Their point is that simply editing a photo with a tool should not be subject to the label.
Former White House photographer Pete Souza said in an Instagram post that one of his photos was tagged with the new label. Souza told TechCrunch in an email that Adobe changed how its cropping tool works and you have to "flatten the image" before saving it as a JPEG image. He suspects that this action has triggered Meta's algorithm to attach this label. "What's annoying is that the post forced me to include the 'Made with AI' even though I unchecked it," Souza told TechCrunch.
Meta would not answer on the record to TechCrunch's questions about Souza's experience or other photographers' posts who said their posts were incorrectly tagged. However, after publishing of the story, Meta said the company is evaluating its approach to indicate labels reflect the amount of AI used in an image. "Our intent has always been to help people know when they see content that has been made with AI. We are taking into account recent feedback and continue to evaluate our approach so that our labels reflect the amount of AI used in an image," a Meta spokesperson told TechCrunch. "For now, Meta provides no separate labels to indicate if a photographer used a tool to clean up their photo, or used AI to create it," notes TechCrunch. "For users, it might be hard to understand how much AI was involved in a photo."
"Meta's label specifies that 'Generative AI may have been used to create or edit content in this post' -- but only if you tap on the label. Despite this approach, there are plenty of photos on Meta's platforms that are clearly AI-generated, and Meta's algorithm hasn't labeled them."
Former White House photographer Pete Souza said in an Instagram post that one of his photos was tagged with the new label. Souza told TechCrunch in an email that Adobe changed how its cropping tool works and you have to "flatten the image" before saving it as a JPEG image. He suspects that this action has triggered Meta's algorithm to attach this label. "What's annoying is that the post forced me to include the 'Made with AI' even though I unchecked it," Souza told TechCrunch.
Meta would not answer on the record to TechCrunch's questions about Souza's experience or other photographers' posts who said their posts were incorrectly tagged. However, after publishing of the story, Meta said the company is evaluating its approach to indicate labels reflect the amount of AI used in an image. "Our intent has always been to help people know when they see content that has been made with AI. We are taking into account recent feedback and continue to evaluate our approach so that our labels reflect the amount of AI used in an image," a Meta spokesperson told TechCrunch. "For now, Meta provides no separate labels to indicate if a photographer used a tool to clean up their photo, or used AI to create it," notes TechCrunch. "For users, it might be hard to understand how much AI was involved in a photo."
"Meta's label specifies that 'Generative AI may have been used to create or edit content in this post' -- but only if you tap on the label. Despite this approach, there are plenty of photos on Meta's platforms that are clearly AI-generated, and Meta's algorithm hasn't labeled them."
EXIF (Score:2)
Re: (Score:2)
We don't know how many detection layers Meta has implemented. Here it was an obvious case of an image self-declaring as made by AI, so obviously it got tagged.
Regarding cameras, big manufacturers are developing digital signatures to tag real photographs https://slashdot.org/story/24/... [slashdot.org]
Re: (Score:2)
The catch with their idea for signatures is including time, date, and location of when the photo was made. I distinctly remember when that almost got someone killed. [wired.com]
Additionally, the signature is supposed to be embedded into the pixels. Which means it's likely loss on lossy compression. So it'll be up to the social media websites to "trust us bro" because I highly doubt they'll be lossless enough to transmit the embedded pixels onward to consumers in mass. Which that's exactly what we want, "trust me"
Re: (Score:2)
Indeed all those technologies have an error margin and can possibly be hijacked as well. I am pretty sure some non-guilty people rest in jail after being convicted with DNA samples. I remember some clerk telling me once that the computer is always right and never makes mistakes, LOL
Re: (Score:2)
Including the data in the pixels can survive lossy compression, but is itself a degradation of the image, similar to lossy compresson in that it allegedly is perceptually imperceptible. Except it isn't.
Free! (Score:2)
But you don't seem to realize the part they are getting.
So you get your photos indexed.
Call me cynical, but.. are they using all that "data" for AI training?
No?
Re:Free! (Score:5, Interesting)
Given their past behavior, I wouldn't be surprised if this is Meta's operating plan.
1) Meta leadership becomes aware that hoovering up and using everyone's data in their for-profit AI work, without permission and without attribution, is becoming a PR problem and may soon become a legal problem.
2) Meta leadership sees recent stories about courts declaring that an AI cannot assert copyright
3) Meta leadership uses back channels to inform their coding team they need a "bug" introduced that will erroneously tag professional photographers' photos with a "made with AI" identifier.
4) Meta continues hoovering up everyone's data and feeding it to the AI.
5) Meanwhile, Meta leadership works with Meta's legal team to prepare for the eventual "we made a mistake, we're terribly sorry, we won't do it again" testimony in front of Congress.
6) Mark Zuckerburg's Facial Manipulation team begins new preparations for making Mark's expression look at least marginally human by the time the Congressional questioning is likely to begin.
Working Correctly (Score:3)
Re: (Score:1)
Celebrity photos have been cleaned (doctored) since the dawn of photography.
Re:Working Correctly (Score:5, Interesting)
Some of the example photos look straight out of camera. Some pre-date the existence of Photoshop.
Re: (Score:2)
AI getting jealous already now? (Score:2)
Can't keep track of all the new AI developments.
Re: (Score:3)
No. If you mark a photo as "AI made", you obviously don't have to pay royalties or even acknowledge the maker. Profit!
Remember, AI does very little except being stoned digitally.
Almost all the "consequences of AI" are consequences of human actions.
So what happens with a photo edited with AI? (Score:2)
Like CA prop 65, everthing is AI (Score:5, Interesting)
Yo dawg... (Score:2)
It's hard to tell what's AI (Score:2)
Remember this real photo [cnn.com] that won an AI photo contest?
AI and meta (Score:2)
I'm sorry to say this to photographers, but (Score:1)
their business is finished. Meta probably made a (semi-) honest effort, and the result was less than optimal. With the progress of the AI models, pretty soon neither Meta nor the human eye will be able to distinguish a real photography from the sh!tload of AI-generated stuff.
As a consolation to photographers, their business will be far from the only one obliterated by AI. Interesting times are coming.
Re: (Score:2)
Comment removed (Score:3)
Does Meta really expect their (Score:1)
...guessing engine to be sufficiently accurate? Sounds like a fool's errand from fools.