Scientists have developed a new tool that is to be a "poison" for generative AI. Nightshade allows artists to change their images so that they become useless for algorithms. However, a human will not notice the difference.
Free access to this article is provided to you by MU Group.
Creators of digital art have found a solution to the problem of copyright theft by generative AI learning models. A team from the University of Chicago has presented the Nightshade program - a tool that "poisons" visual information with pixels invisible to the human eye, making images useless for AI models. The team's leading scientist, Professor Ben Zhao, hopes that the developed program will return control to the hands of artists.
Currently, tools based on generative AI such as DALL-E, Midjourney or Stable Diffusion download data from the internet without respecting copyright, paying fees for the use of works or the consent of the artists themselves, who feel threatened.
Nightshade is designed to solve these problems. The program introduces changes to each pixel of the image that are invisible to the human eye. A model trained on these samples begins to hallucinate. This causes the artificial intelligence to start interpreting, for example, an image of a cat as a dog, a hat as a cake, or changing women's handbags into toasters. Since the data collection process is usually automated, over time more and more misleading data will be delivered to the model. Their removal will be a huge problem, as creators have to do it manually.
Nightshade is an open source project, which allows other programmers and researchers to introduce their own versions and improvements of this method. Scientists are not afraid that they will completely ruin the work on current AI models. These are trained on billions of images, so point "poisonings" should not be a problem for the general operation of AI. However, it will make it difficult to copy artists who are securing themselves with the system.
Nightshade is a tool that offers artists the opportunity to actively defend against unauthorized use of their works in AI models. Although this requires a significant number of "poisoned" images to effectively affect the operation of the model, it is an important step in the fight for artists' rights in the digital age. As representatives of Stability AI noted, the development of AI models is moving towards greater diversity and elimination of biases, but Nightshade shows that artists are not defenseless and can also act to protect their achievements.
The European Union has reached an agreement on the AI Act, Google is competing with OpenAI, and Kantar announces marketing trends for 2024. What else have we been living with in the past few days?
As the only company in the world, it is able to produce processors in the most advanced 3 nm technology. In addition, it owns the most expensive factory ever built, and its clients are Apple, NVIDIA and Intel. How did it happen that Taiwanese TSMC became one of the most important companies in the world?
Another scandal in the world of cryptocurrencies with money laundering in the background, Germany and France block the EU AI Act, and IAA Poland initiates the Round Table of Influencer Marketing. We invite you to another review of the week from the Digitized editorial team.
On the occasion of the Meta Threads app premiere, a discussion about the decentralization of the internet and handing it over to the users has flared up again. Why can the ActivityPub protocol play a key role in this process and what does the concept associated with it, fediverse, mean?
English