Google Unveils New Feature To Detect AI-Generated Images

Google is adding features to its search engine to help users identify images that were generated through artificial intelligence, the New York Post reported.

Last Wednesday, Google announced that it will be adding a “markup” to the metadata of images produced by its artificial intelligence models to show that they were AI-generated. The search engine will then use the metadata to display a label on any AI-generated images that appear in search results.

According to a recent company blog post, the image will include a caption below it noting that it is “self-labeled” as being generated by artificial intelligence.

Eventually, images from other prominent photo publishers, including Shutterstock and Midjourney, will also include similar warning labels.

In addition to the labels, Google will debut another tool called “about this image” within its search engine sometime in the “coming months,” the blog post said. This tab will give users easy access to details about a particular image that appears in the search results, including where the image has appeared online and when it was initially “indexed” by Google.

As an example of how the tool will work, Google presented an AI-generated image depicting the faking of the Apollo 11 moon landing in July 1969 that features a film crew. When a user clicks on the “about this image” tab, the user will see links to articles debunking the image as well as other key details to confirm that the image is fake.

According to Google, this tool will provide “background information” on images that will allow users to better determine if the image is “reliable” or if they “need to take a second look.” 

Earlier this year, AI-generated images of Pope Francis in a puffy, white Balenciaga jacket and Donald Trump being hauled away by the NYPD went viral online.

The so-called “Grandfather of AI,” Dr. Geoffrey Hinton recently warned that artificial intelligence could hasten the spread of misinformation online by producing images that are almost indistinguishable from real photos or providing inaccurate answers.