Google’s New Tools Can Identify AI-Generated Fakes Images

How can you tell if an image is real in a world that is increasingly populated with lifelike images that have either been changed using AI editing tools or produced by a generative AI bot like Midjourney or Stable Diffusion? A new feature called “About this image” that Google is introducing this summer for English-language searches in the US is one thing that might be helpful.

It is comparable to the “about this” drop-down that is now accessible in Google image searches and can be found on links in standard search results. You may now find out when a picture and others like it were initially indexed by Google, where on the web it first appeared, and other sites it has since appeared on when you run a “reverse image search” by uploading an image of unknown origin.

However, that is not the only situation in which this would be helpful. In Google’s example, a photograph of a faked Moon landing is uploaded, and the tool then displays how the image has appeared in publications debunking them.

For instance, it would seem reasonable to assume that a picture of a breaking news event is authentic if it initially appeared when it was uploaded to Getty, Reuters, or CNN. No matter how amazing the pope’s new Balenciaga attire appears, a photo that initially appeared in a random humour subreddit and has a news organization’s watermark is more likely to be a fake.

Assuming it functions as planned, this kind of technology will hopefully be imitated by rivals and immediately made accessible to individuals outside of the US, as the propagation of false information is an issue in more than just a single country and language.

Google also revealed that, regardless of whether you view a picture on a Google platform, its own generative AI capabilities would include metadata to each image to show that it was produced using AI rather than a photograph. The company added that other publishers and producers will be able to tag their photographs with the same technology, though it’s unclear how widespread the use will be.  According to Google’s blog post, the markup will be implemented by Midjourney, Shutterstock, and other companies in “the coming months.”