In the past, spotting fake images wasn’t too difficult due to technology back then not being quite as good. These days, not only has the technique involved in making fake images become better, but so has the tech, where the rise of DeepFakes has us a bit worried for the future.
However, thanks to the work of researchers over at MIT and IBM, they have managed to develop an AI tool that is capable of detecting fake images. The AI was originally created to be used as a way to generate photographic images from scratch, but it seems that in the process of creating this tool, the researchers were surprised to find that the AI was capable of learning about the relationships between objects.
For example, it was able to tell that a door was not meant to be placed in the sky, and because of this simple knowledge of relationships between objects, it turned out to be useful in helping detect images that might be fake. According to Jun-Yan Zhu, who co-authored a paper on the tool, “You need to know your opponent before you can defend against it.”
The tool is available as a free demo for those who might be interested in trying it. In the meantime, other companies such as Adobe have also created tools to help detect fake images, while others such as Google are launching courses on teaching kids how to spot things that are fake on the internet.