AI Camera with No Lens???

[Zack Birlew]

Okay, this is getting freaky now! Someone has come up with a “camera” that doesn’t take an image but generates an image based on geographical location data, so kind of like an intricate Google Earth snapshot device. My only concern is when is it going to happen that you will try to take a photo of something and then find that there’s a black monolith or pixelated blur of something or someone that’s been flagged to not be photographed by the government, business, media, etc.?

Perhaps we could go down one road where AI makes taking photos easier like exposure adjustment and focusing or perhaps we could go down another road where professional cameras don’t have AI at all, whichever way we go, I think we need to clearly define and be able to technologically identify what is an image and what is AI clickart going forward. If this ever gets made, I feel sorry for the customers that buy into the creative fraud of it.

https://www.sonyalpharumors.com/cur...s-ai-generated-images-based-on-your-location/

 

[Alex Zeder]

My first thought was: why do I even have to go to the places and use a device to get these "photos"?? I could as well just point my finger on a map.

We'll probably see the use of AI diverge between documentation in all its forms, and fiction. Simple AI-enhanced grading for a doc is acceptable, changing relevant content is not. For fiction, all rules are off. There is certainly some overlap like trying to sell fiction as documentation, as has been pre-AI too. But that's not new nor does it require AI and it depends on the intention and use of the content, whether it's made to entertain or to mislead. Going from simple image "enhancements" to content modifications like removing objects, adding a person etc. up to fully fictional content.

Unless you legally forbid publishing images without proper tagging that tells the level of 'fiction' used, there won't be a way to determine just from the content whether and how much AI was used as the algos/models get better over time. And there are always folks who then pretend that they hadn't heard about these legal requirements when they published their untagged pictures with the intention to mislead. Now you could try to train yet another AI (or even the same one that generates such images) to detect such generated/modified images, but that's like playing chess against yourself and a dead end very soon. The solution needs to be on the 'use' side, not the technical side, in that you can maybe forbid publishing images with the intent or great likelihood to mislead. Some long existing laws might even be applicable for this as well. But proof for this kind of fakery then probably can't come from the image itself but from fact checking around it. AI can certaily help with this part also (prompt "Compare this image to all other images from the same place at about the same time.")