 |
| AI tools easily create fake technology products. (Illustration photo) |
“The age of deepfake”
On the flip side, AI, in the wrong hands, can easily steal images from individuals’ social media accounts and turn them into unwanted references. For example, pornographic websites can use images of individuals without consent.
“We are in the age of deepfakes,” said Hadi Salman, a researcher at the Massachusetts Institute of Technology. “Now anyone can manipulate images and videos to make people ‘do’ something they didn’t actually do.”
Deepfake is a combination of the words “deep learning” and “fake”, which is a method of using AI to create fake technology products in the form of audio, images or even videos.
In just a few seconds, with a few simple commands, AI can create any image the “bad guys” want. For example, someone could ask AI to create a photo of the Pope wearing a Balenciaga coat and then post it on the Internet before it turns out the photo is fake.
AI technology has also amazed users with its ability to create works of art in the style of a particular artist. For example, you can create a painting of a cat that looks like it was painted in the style of famous painter Vincent Van Gogh.
For visual artists, these tools risk putting them out of work as AI models mimic their unique styles and create artwork without asking for permission.
AI Propeller
Eveline Fröhlich, who lives in Stuttgart, Germany, is one of a growing number of artists who are fighting against the overreach of AI and trying to find ways to protect their visual works online from manipulation.
Recently, Ms. Fröhlich learned about a tool called Glaze, developed by computer scientists at the University of Chicago (USA) to prevent abuse of AI models.
Ben Zhao, a professor of computer science at the University of Chicago and one of the lead researchers on the Glaze project, told CNN that the tool is intended to protect artists' unique work from those training AI models.
Glaze software uses machine learning algorithms to make small changes to digital images that are invisible to the naked eye, but create a protective layer for the work in the online environment, fooling AI models, making them unable to copy the author's style.
Thanks to Glaze, works of art are "covered with an invisibility cloak", preventing AI from "reading" the image.
For example, Zhao explains, an artist could upload an image of their own oil painting, which is run through Glaze. When the AI models “read” the painting, they would see it as a charcoal sketch—while a human would immediately recognize it as an oil painting. After running it through Glaze, the artwork would now look significantly different to the AI’s eyes.
Zhao's team released the first version of Glaze last March, and it has been downloaded more than a million times. Just last week, the team released a free version of Glaze online.
The team hopes the Glaze tool will help protect artists' copyrights until specific regulations and laws are established.
Since introducing Glaze, Mr. Zhao said, he has seen an increase in the number of voice actors, novelists, musicians, and journalists reaching out to his team to ask about a version of Glaze for their field.
California-based visual artist Jon Lam is using the Glaze tool to help protect his online artwork from being used to train AI models, and says he now uses the app for all the artwork he shares online.
For years, artists like him have been posting their work in full resolution online as a matter of pride, he said. “We want people to see how detailed and amazing full HD photos are. However, artists don’t know that their work can be ‘eaten up’ by AI models, which then copy their style and expose them to unfair competition.
“Now we know that people are taking our high-resolution work and feeding it into AI machines to compete in the same space as us. So we have to be more cautious and start thinking about how to protect ourselves,” he said.
Salman and his team have just released another tool to protect images from AI. The app, called PhotoGuard, allows users to place an invisible layer of protection over images to prevent AI models from manipulating the images.
PhotoGuard's goal is to protect photos that have already been uploaded online by their authors from “malicious AI manipulation,” Salman said.
The PhotoGuard app works by adjusting the pixels of an image in a way that is imperceptible to humans. However, this change, which is imperceptible to the naked eye, is powerful enough and carefully circumvents any attempts by AI models to manipulate the image.
This means that if someone tries to edit a photo using AI after it has been protected by PhotoGuard, the resulting photo will be “not realistic at all,” Salman explained.
Not licensed
On the other side of the coin, even if AI tools take and manipulate images, what they create is not recognized as copyrightable.
Since its emergence and popularity, AI has simultaneously sparked a debate: are these works copyrighted? If so, who do they belong to?
The United States Copyright Office (USCO) has made an official decision on these works. According to the USCO, any images generated by AI based on commands - (like the current apps Midjourney, Dall-E, and Stable Diffusion) - will not be copyrightable in the US.
US law states that intellectual property can only be copyrighted if it is the product of human creativity. The USCO currently only recognizes works created by humans. Therefore, AI machines will not be considered authors, and the products they create will not be copyrighted.
“If the traditional elements of copyright in a work are created by a machine, then the work is not copyrighted by a human, and the USCO will not register it,” said USCO Director Shira Perlmutter.
Source
Comment (0)