Artists are fighting back against AI
This past spring, we witnessed strong criticism of ArtStation, which quickly snowballed into a mass "No to AI-generated images" protest joined by thousands of professional and amateur artists demanding the removal of AI-generated content. The antagonism of the artistic community towards AI is on the rise, as the initial opposition swiftly escalated into dozens of class action and individual lawsuits in the USA. Various individual artists and authors' guilds are suing providers of generative AI on account of copyright infringement, and more and more are joining the battle every day. Today, there are dozens of ongoing lawsuits brought by different types of artists in the USA dealing with this matter.
The artistic community has already voiced several fundamental concerns, some of a moral nature, others purely legal. The ethical arguments revolve around the fear of human work becoming devalued. They can be summed up as AI-generated content is contrary to art because simply clicking on a button to generate an image is not a creative endeavour. In that sense, AI does not really "create" anything new, but simply looks at available art and then mixes it into something else. Against that backdrop, artists feel that AI-generated images demean their work, skills and efforts, and undermine the time and talent that goes into art.
Beyond the philosophical and moral aspects of creativity, there are also deeper legal issues. One aspect concerns the potentially unfair competition that traditional artists are now facing. Mass-produced AI products can be created at the click of a button and cost nothing compared to classic art, something artists will struggle to compete with in the long run. But above all, it seems that all the commotion around AI might boil down to IP-related issues. The concerns voiced by aggrieved artists already encapsulate burning IP issues. What happens when AI art infringes the IP rights contained in prior human art? To generate images from prompts, AI relies on available databases that comprise billions of images and text trawled from the web. While those databases ought to contain content from the public domain, many artists argue that databases also host many copyrighted images. So, at least in part, AI relies on pirated human-made art, thereby infringing the IP rights of various rightsholders.
Artists also have stressed that many AI tools use human art without the knowledge or permission of rightsholders. Not only is AI "trained" on account of intellectual property, but it also effectively mashes elements thereof without observing the attribution and integrity rights of the copyright holders. Finally, a fundamental critique concerns the status of AI output, with many at odds as to whether intellectual property rights can apply to AI-generated content.
While the courts are yet to give their say in this matter, the artists have not remained idle. They are starting to rely on tools that directly contaminate and confuse AI systems, such as Glaze, Nightshade and Kudurru. Nightshade, for example, confuses the matching of images with textual prompts by creating a discrepancy between the image and text, thereby causing the AI to pair, for example, the prompt "car" with an image of a cow. "You can think of Nightshade as adding a small poison pill inside an artwork in such a way that it's literally trying to confuse the training model on what is actually in the image," says Zhao, the leader of the research team that built the tool. Working on the same principle, Glaze modifies the pixels in an artwork so that AI cannot produce works in the style of a specific artist, while Kudurru tracks scrapers' IP addresses to either block them or send them back unsolicited content (such as the middle finger).
All in all, these digital tools allow artists to disrupt future AI by "poisoning" the copyright works that may be included in the training datasets. Data poisoning attacks manipulate training data to introduce unexpected behaviour into machine learning models at training time. As such, they will not help artists revert and "untrain" the existing AI models who have already digested tons of artworks, but they might be able to prevent future training of their creative works without permission. The idea is to eventually pollute and break future AI models to such an extent that AI companies will be forced to either stop training on copyright works or be forced to seek permission from authors for data scraping.
These tools have come at a time when the debate on the use of copyright works for training purposes has intensified between two contrasting positions. One, voiced by authors but also other stakeholders that such use requires the triple C (Consent, Credit, Compensate), and the other asserted by AI developers who rely on "fair use" or other legal grounds to train AI tools on vast amounts of copyright materials without the consent of the authors. But it seems that the authors are not entirely alone in their position. An executive of Stability AI, Ed Newton-Rex, recently resigned over the company's view that it is acceptable to use artwork for training purposes without the permission of the artists. He told the BBC he thought it was "exploitative" for AI developers to use creative work without consent. "I think that ethically, morally, globally, I hope we'll all adopt this approach of saying, 'you need to get permission to do this from the people who wrote it, otherwise, that's not okay'," he said.
The use of copyright materials for training purposes thus remains one of the key issues in the present IP vs. AI debate. While we await the first decisions of the US courts that should finally clarify whether the use of copyright works for training purposes will be allowed in the USA under the "fair use" principle, Europe remains silent, as there is still no indication of how this question will unfold in practice. At the same time, European countries may end up with a heavily fragmented approach, as certain countries (such as France) are considering an author-friendly approach requiring the AI to seek prior permission, credit all individual authors and pay back a fair tax for the works used for training purposes, as opposed to the current "text and data mining" exception introduced by the EU Digital Single Market Directive, which allows scraping of copyright works for certain purposes. And while it remains for legal theory and practice to address the issues raised by artists and resolve the identified problems, the underlying idea (that it is unfair to reap what you did not sow, to steal another's labour of mind) continues to be as relevant today as it was at the dawn of the printing press.