Nightshade-With this new tool, artists are now defending themselves against AI companies

Artificial intelligence is increasingly being used in the art world, which on the one hand opens up new creative opportunities, but on the other hand displeasures many traditional artists.

Generative AI and AI tools have added a completely new dimension art You can imitate the style of well -known artists and create new works or even create their own unique works that can even win prices.

But not everyone likes this path.

With Nightshade and poisoned pictures, artists are now fighting against AI companies and generative AI

Researchers at the University of Chicago have developed a new tool with Nightshad that enables artists to defend themselves against the use of their works by generative AI models . This was recently reported by Technology Review .

Nightshade changes subtle training data for image-generating AI models and thus starts targeted attacks on these technologies.

This innovative tool gives artists the opportunity to make invisible changes in the pixels of their art before they publish them on the Internet. However, there is a risk that the resulting model will be damaged in an unpredictable and chaotic way when transferring to a AI training set.

The deliberate "poisoning" seriously harm future versions of image-generating AI models such as Dall-E , Midjourney and Stable diffusion Some results could become unusable - dogs could become cats or cars, etc.

Technology Review had exclusive access to a research study that was submitted to the Peer review at the Usenix

AI companies such as Openai, Meta, Google and Stability AI are facing a number of lawsuits from artists. These claim that your copyrighted material and your personal data were used without consent or compensation.

Ben Zhao , a professor at the University of Chicago and head of the team behind Nightshade, hopes to restore the balance of power between the AI ​​companies and the artists .

This should be achieved by a strong deterrent against the disregard of the copyrights and intellectual property rights of the artists. The companies Meta, Google, Stability AI and Openai have not yet responded to the request from Technology Review regarding their possible response.

Nightshade and Glaze as a combination weapon against copyright infringement of the large AI companies

Zhao's team has also Glaze -a tool for artists -with which they "mask" . This prevents this from being copied by AI companies.

Similar to Nightshade, subtle changes are made in the pixels of an image. These are invisible to the human eye; However, they manipulate machine learning models so much that they interpret the image as something else than it actually is.

The team plans to integrate the data tool Nightshad in Glaze . Artists have the choice whether they want to use this tool or not. In addition, Nightshad is also offered as an open source, which enables others to get around it and create their own versions of them.

By using and creating different versions, the tool becomes increasingly powerful, emphasizes Zhao. The data records for large AI models can consist of billions of pictures. The more poisoned images can be fed into the model, the greater the damage from this technology.

How does the use of Nightshade and Glaze work?

Artists who want to share their works online, but do not want AI companies to use them for their purposes, have the solution called Glaze ( here is the tool ).

Here you can upload your pictures and alienate them with a different style. In addition, there is now also the possibility to use Nightshade

When AI developers search the Internet to collect data for improving or creating models, these manipulated samples are subjected to you and thereby lead to malfunctions.

These poisoned data samples have the effect of influencing the AI ​​models in such a way that hats, for example, are recognized as cakes and handbags as a toaster. It is extremely difficult to sort out such false information - technology companies would have to find and delete every single damaged sample.

The researchers tested this attack on both the latest models from Stable diffusion and on a self-developed AI model. When she only fed 50 poisoned dog pictures and then asked to generate more pictures of dogs, the result was strangely: beings with excessive limbs and cartoon -like faces.

With 300 poisoned samples, an attacker can even make stable diffusion to create cat pictures instead of dog pictures.

What effect do these tools have on the power structure between authors and AI companies?

Junfeng Yang , an computer science professor at Columbia University, which researched the security of deep learning systems and was not involved in the project, convincingly explains that Nightshad could have a potentially great impact-especially for deterrent.

For example, AI companies be made to better respect the rights of artists and rather be ready to pay license fees .

AI companies such as Stability AI and Openaai, which have developed text-to-image However, many artists believe that this is not sufficient.

Eva Toorentent is an illustrator and artist who worked Evaboneva opt-out methods would still request to subordinate to the technology companies.

Toorenent hopes that Nightshad can change the status quo:

It will move [AI company] to rethink because they are now able to destroy their entire model by using our work without consent ”,

Commented them with Technology Review.

Another artist named Autumn Beverly reported that tools like Nightshade and Glaze gave her the trust to publish her works online again. Before that, she had removed it when it came out that it was copied Laion

I am really grateful to have a tool available that can help the artists to regain control of their work ”,

she says.