You’ve probably heard a ton about AI-generated stuff lately, from deepfakes to AI-created art, right? 🤖 Well, here’s a question that’s been bugging us: Are our photos secretly fueling the training of these AI robots? Could they be snatching our pics from our sites or social media to level up their game? 🤨
See, the internet’s a free-for-all that allows generative AI to snatch art and use it for training and it’s normal to have that spooky feeling that your art is under surveillance. 🎨👀 With AI tools like Midjourney constantly devouring the web’s content buffet, that feeling just keeps intensifying. There might not be a cream for that itch, but there is a website that can help you see if AI has laid its digital hands on your creative genius. 🕵️♂️🌐
There’s this site called Have I Been Trained? created by Spawning AI, who make cool image-generation tools for artists. You can use this site to dig through a whopping 5.8 billion images that were used to teach AI art models. You just search your name and it shows you the closest matches from the LAION-5B training data. It’s like peeking into the AI’s secret stash. Talk about a digital eye-opener!
Image courtesy of Have I Been Trained?
Of course, I gave it a shot, starting with BubblegumClub (as seen in the images above). The results are shady. While it’s hard to trace irrefutable evidence, the types of results are aesthetically aligned with the BubblegumClub brand. However, the examples of direct knicking are so vague, it appears reasonable to doubt whether AI has used publication to train itself.
Still, the concerns about where the data used to teach AIs comes from are valid. An Ars Technica report recently uncovered private medical records in LAION-5B, which raised serious privacy concerns because removing them is quite the challenge as LAION isn’t organised as files, but is simply a collection of web image URLs.
In response, tech gurus like Mat Dryhurst and Holly Herndon are leading Source+, a movement that says “Nope, you can’t use my stuff for AI training!” Even Dungeons & Dragons wizard Greg Rutkowski faced off against AI. But the catch is: individual efforts might not have that big of an impact. Have I Been Trained?, however, has only been in the game for about a year and keeps getting better.
Image courtesy of GIGAZINE
This website keeps it simple by spilling all the binary beans on whether your art or photos got caught up in AI training. Plus, it shows you precisely which art has been swiped. And here’s the icing on the cake: they’ve got a ‘once-only opt-out tool.’ Hit that button, and you can take your shit and bounce.
And here’s some more good news: if you’re looking to armour up your art against AI, there are more solutions on the horizon. Take Glaze, for example, a fresh tool that throws AI off the scent when it tries to sniff out your art, making sure the results are all kinds of wrong. This freeware jazzes up your image to confuse AI – it might think your work’s a Van Gogh, but the changes are so low-key, even humans can’t spot ’em.
The quest to “opt-out” seems to be the most problematic part of the entire ordeal. Even if your images are removed during the main AI training, others may throw them back in during their own sessions. AI training operates on legally shaky ground where it might breach copyright laws because it has little regard for the permission of the content owners. If this AI training results in financial gain, it could be classified as commercial use, raising more legal red flags. 💼🧐🚩
Image courtesy of CHIP
As for the future, who can predict what’s on the horizon? Lawsuits, debates about tech harm reduction, digital consciousness, and the age-old question of free will and creativity are in the mix. But controlling one’s images in the online realm will always be a thorny issue. Whether or not he actually said it, Pablo Picasso is often given props for dropping the line: “Good artists borrow, great artists steal.” 🤔🕵️♂️Some argue that AI using images is just like how artists have always been inspired by each other. But using another person’s work without credit is still a no-no, whether done by humans or machines. 👀
Whether you see it as cool or controversial, AI recognising your work does have its appeal. After all, you put it online to be seen, and if AI appreciates it, mission accomplished. 🤖🖼️ But here’s the kicker: using your images, AI can whip up “new” art with minimal effort, and that leaves traditional artists a tad uneasy. 😨🎨 The shift in creative power toward the “normies” or “NPCs” could mean less attention for what traditional artists could bring to the table. 🎨🙅♂️ This could spell the end of art as we know it and potentially, the final nail in the coffin for originality.
All artists can do is stay woke outchea and sites like Have I Been Trained? can help at least spot the digital bandits that wanna make off with your creative capital.
Here is the link if you want to search for your own pictures in the database.