Since OpenAI’s release of ChatGPT towards the end of last year, many similar products have been introduced to the market, making productive AI mainstream. Productive AI goes beyond providing answers to complex questions with just a few lines of dialogue. It can also generate incredible images that appear almost unbelievably real, leading to potential concerns about deception and fraud. The solution to this issue comes in the form of a new MIT software!
MIT’s new software: PhotoGuard
Now that we can create incredible images with AI, there is a need for safeguards built into photos to make it difficult for someone to use them for generating fake images. MIT has taken the first step towards this by offering a software solution called PhotoGuard. This feature can prevent AI from convincingly altering your photos.
Researchers from MIT CSAIL (MIT Computer Science and Artificial Intelligence Laboratory) have published their innovation in a research paper. PhotoGuard makes it impossible for artificial intelligence to recognize certain pixels in an image by altering them. This feature ensures that, at least visually, the photo remains unchanged for humans.
When attempting to generate fake images using elements from protected images through PhotoGuard, AI cannot interpret the pixel alterations. As a result, the fake images created by AI will have evident sections that indicate the image has been modified. You can see how the program works in the video below.
As you can see in the video, with PhotoGuard, when attempting to create a fake photo, noticeable pixel distortions are visible, clearly indicating to users that the photo is fake. The researchers have discovered two more protection methods that can thwart AI’s efforts, both integrated into PhotoGuard.
The “Encoder” method makes it impossible for AI to understand the image’s segments.
The “Diffusion” method camouflages parts of an image as a different image for artificial intelligence. In both cases, AI will not be able to produce a flawless fake photograph.
Hadi Salman, the lead author of the paper, suggests that companies like Apple and Google should consider adding this technology to their stock camera applications on iPhone and Android, respectively.
{{user}} {{datetime}}
{{text}}