A recent trending study suggests that a digital watermark, a covertly embedded digital marker to detect copyright infringements, can be easily bypassed letting the person who does use it, reuse it, or edit and use it on other platforms.
Study For Digital Watermark Protections
How it Works?
The study was conducted by professors of the University of Maryland who ran tests on digital watermark protection and found out that it can be bypassed easily. The study report is titled Robustness of AI-Image Detectors: Fundamental Limits and Practical Attacks.
One among the authors of this report Mr. Soheil Feizi said that currently there were not any watermarkings that are reliable. His team subjected the digital watermarks to certain procedures and were able to remove the watermark.
The process used by the team is referred to as diffusion purification which applies Gaussian noise to the given watermark. This distorts the watermark that will bypass the detection algorithms.
‘Diffusion Purification’ or ‘DiffPure’ is a method by which different levels of noise on the model undergo this process. The end result will be all the perturbations in the image getting cleaned. In the case of an image that has a protective digital watermark, when the noise levels are applied to it, the watermark will get distorted. When the digital watermark that the author or the creator has applied to the image gets distorted the image will appear as one without a watermark at all.
Digital Protections Algorithms
The detection algorithms identify the creator of an image and whether the image has been watermarked by anyone. When these watermarks are distorted, the detection algorithm will not recognize it as belonging to anyone or having been watermarked by anyone so it will bypass the algorithm. This will enable the fraudulent users to steal the original images from others, modify them or edit them, and use and pass them as their own.
Currently, it is viewed that it is challenging to design and embed a digital watermark that is robust enough to pass the distortion-causing processes. However, more studies and attempts have been made in this field to explore the possibilities and to develop a method to design a robust watermark.
When the digital watermark protection can be easily bypassed it can lead to the spending of false information and the present scopes of AI in this way more impactful.
Digital Protection against AI
One of the most disturbing things that AI is capable of doing is generating deepfake images that have the ability to convince people and mask truth and reality. People update their photos on various social media and other media platforms. AI can take the image of any person, let it be a celebrity of our times who is globally popular or an introvert who keeps minimal contact with people and can use the face on other platforms.
Recently, celebrities have been shocked to wake up to ads and promotions with their faces on them, when in fact they might not have even heard of such things! This leads to a society where AI has the power to deny reality to anyone and everyone and let individuals and communities believe things that might not even have happened.
It can even bring in the faces of national and international leaders to announcements and statements that they might never agree upon and this can be a much bigger problem. These are only a few of the threats and challenges that AI poses.
Find out more : Google Alters SEO Playbook To Recognize AI-Generated Content