What essential feature did Google provide to identify the image generated by artificial intelligence?

Artificial intelligence is almost everywhere today and helps people do their jobs, as well as write letters, create content, learn new things, and more. This raises fears that the technology is too ubiquitous, as it could potentially replace jobs for some people. AI should be used with caution as it does not necessarily provide correct information and can be biased, racist, or insulting. Therefore, it is important to use it intelligently, knowing its shortcomings and possible failures.

AI images are getting harder to spot. Google thinks it has a solution.

AI-generated images are becoming harder to distinguish from the real thing as tech companies rush to improve their AI products. As the 2024 presidential campaign progresses, concerns are rapidly growing that such images could be used to spread false information.

Google announced a new tool, called SynthID, which it says could be part of the solution. The tool embeds a digital "watermark" directly into the image that cannot be seen by the human eye but can be picked up by a computer trained to read it. Google said its new watermarking technology is tamper-resistant, making it an essential step in controlling the spread of false images and curbing the spread of misinformation.

AI image generators have been around for several years and are increasingly being used to create "deep fakes" - fake images that pretend to be real. In March, fake AI images of former President Donald Trump running from police went viral online, and in May a fake image showing an explosion at the Pentagon sent stock markets momentarily tumbling. Companies have placed visible logos on AI images, as well as attached text “metadata” that indicates an image's origin, but both techniques can be cropped or edited relatively easily.

As AI gets better at creating images and videos, politicians, researchers, and journalists worry that the line between what's real and what's fake online will further erode, a dynamic that could deepen existing political divisions and make it more difficult to disseminate factual information. The improvement in deep fake technology comes as social media companies are giving up trying to control misinformation on their platforms.

Watermarking is one of the ideas that tech companies are pushing as a potential way to lessen the negative impact of “generative” AI technology that they are rapidly powering millions of people. In July, the White House hosted a meeting with the leaders of seven of the most powerful artificial intelligence companies, including Google and OpenAI, the maker of ChatGPT. All companies are committed to creating tools to watermark and detect AI-generated text, video, and images.

Microsoft has started a coalition of technology and media companies to develop a common standard for watermarking AI images, and the company has said it is researching new methods for tracking AI images. The company also places a small visible watermark in the corner of images generated by its artificial intelligence tools. OpenAI, whose Dall-E imager helped start the wave of interest in AI last year, also adds a visible watermark. AI researchers have suggested ways to embed digital watermarks that the human eye cannot see but a computer can identify.

Kohli, the Google executive, said Google's new tool is better because it works even after the image has been significantly changed, a key improvement over previous methods that could easily be thwarted by modifying or even flipping an image.

Take a closer look at people.

One of the first things you should pay attention to is how humans are represented in the image. It is difficult for AI to accurately reproduce parts of the human body because they are complex, so paying close attention to them can help you identify if there is something wrong with the image.

For example, there may be inconsistencies, such as an unusual number of fingers, an abnormal shape, or a peculiar position. Likewise, keep an eye out for facial details that may look strange, especially around the eyes and ears, as these are often harder for the AI to generate. A difficult feature for AI to reproduce is wrinkles and lips, which must be consistent across the entire face, and a discrepancy in these can also be a sign that the image is not necessarily a real photo.

Check all the details.

In addition to the bodies of the people, it is also important to look at all the elements of the photo, such as clothing and accessories. Check if they make sense or if the shading and details are rendered correctly. If there are animals or flowers, make sure their size and shape make sense, and look for items that may seem too perfect, as they could also be fake.

But now, keeping this problem in mind, Google has introduced a tool called SynthID to fix it. The tool is developed by Google Cloud in partnership with Google DeepMind and Google Research.

SynthID leaves a secret code (watermark) on AI-generated images that is invisible to the human eye but can be detected by a computer.

This tool is available for special users using Vertex AI and Image. Image is an intelligent system that converts text to image.

SynthID doesn't work very well, but it's the first step in detecting fake images.