The world is awash in images, but some are not real. Fake photos have been used to smear people’s reputations and even to blackmail them. This has led to cyber criminals creating a lot of misinformation online, which is also being used by governments for surveillance purposes.
To combat this issue, Google has been working on a solution, and now it has launched a new tool to help spot AI-generated images. The tool is called SynthID, and it works by stamping images with a digital watermark that can’t be rubbed off or edited. The company is currently testing the tool with a limited number of users. The tool has been developed by Alphabet’s artificial intelligence (AI) division, Google DeepMind.
It has been challenging to distinguish between the fake and the actual image, especially regarding video. However, if the tool is successful, it can significantly reduce the amount of misinformation and false information published online. Government agencies could also use it to monitor the spread of false information and protect copyrighted images.
Google’s watermarking system is similar to the anti-piracy systems used by music and movie studios to track and identify copies of their content. The company claims that its watermark cannot be rubbed off or edited and can remain detectable despite cropping, adding filters, or changing the file format. The technology is based on an algorithm that uses the pixels in an image to add a watermark. The algorithm considers an image’s color, brightness, and texture to calculate its unique signature. It then compares the watermark to a list of watermarks stored in a database and returns the matching results.
However, it is essential to note that the tool has several limitations, and some experts say it can easily be defeated. According to UC Berkeley computer science professor Hany Farid, “Any method of marking AI content is going to be subject to tampering and hacking.”
Google says it will continue testing the tool and collecting feedback from its beta testers. It hopes to make it available to more individuals and organizations eventually. In the meantime, it will continue enhancing its AI systems. This includes structural improvements to its computing systems and improving its language models. These changes will result in more accurate translations and better search results. It will also increase the speed at which it can process information.