December 1, 2023

AI Watermarks Can Be Easily Removed, State Researchers – Shiv Telegram Media

Artificial intelligence (AI) has taken on a new role in combating the rise of deep fakes and misinformation through the use of AI watermarking technology. Traditional watermarks, which are visible logos or patterns, have been transformed in the context of AI to help computers determine if text or an image has been generated by AI.

Recognizing the importance of addressing this issue, major tech companies such as Google, OpenAI, Meta, and Amazon have pledged their commitment to developing watermarking technology. By implementing AI watermarking, these companies aim to prevent the spread of false information and protect intellectual property rights.

However, recent research conducted by the University of Maryland (UMD) suggests that the current methods of adding or removing watermarks are easily evaded. The UMD researchers discovered that potential perpetrators can bypass existing techniques with relative ease, undermining the effectiveness of watermarking as a solution.

Nevertheless, the same UMD researchers have managed to develop a watermark that is highly resistant to removal without compromising intellectual property. This breakthrough would enable the detection of stolen products and adds an extra layer of security to combat counterfeiting.

In a separate collaborative study conducted by the University of California, Santa Barbara and Carnegie Mellon University, it was revealed that watermarks can be easily eliminated through simulated attacks. The researchers noted that both destructive and constructive methods can be employed to remove watermarks, with destructive methods greatly impacting the quality of the image.

While these findings suggest that there is room for improvement in watermarking AI-generated content, efforts are underway. Google, for instance, is developing a tool called SynthID to address the issue of simulated tests.

As deep fakes and misinformation continue to pose significant challenges in today’s technologically advanced society, the development of effective AI watermarking technology remains crucial. By further refining these techniques and overcoming their limitations, the tech industry can actively combat the spread of falsehoods, safeguard intellectual property rights, and ultimately preserve the integrity of information in the digital age.