Here comes Nightshade, not necessarily the first attempt on holding AI trainers accountable but a notable move. If you are unfamiliar, in simple terms it poisons the image file ingested into AI models, effectively confusing the training and ultimately the results the model will return.
Read more here - https://nightshade.cs.uchicago.edu/whatis.html
This purpose-built software to defer AI is in my opinion a marked ‘line in the sand’ when it comes to the ownership of content in a place that is arguably public. One that I fear puts the world of content monetization and access at odds and effectively is the start of content “weaponization”. I bring this perspective from a unique position as a leader in an AI company that required the use of publicly posted content for our training and prior as an advocate and expert in consumer data privacy.
It is important to recognize the issues that generative AI causes creators. I genuinely feel bad for artists and authors to whom these must seem like existential threats to their livelihoods. I also recognize the terrible cycle in which the content created from those same threatened individuals is used to build the models which in turn threaten them, during which they are not compensated. This situation is made more complex as the traditional web model of free content for advertising dollars deteriorates as supply goes undervalued without accurate tracking and 3rd party data to provide meaningful targeting and transparency. It’s no surprise that ownership, attribution and the monetization strategies of content have become such hot topics.
It's not an easy road ahead and I don’t have all the answers. However, solutions like Nightshade, in my opinion, are a threat to the basis of how we share on the web and further pose a threat to accountability in the availability of public facing materials. While I believe in the need for consumer transparency and basic control of the data they create online, there is still a conversation to be had around the public nature of the actions that we take on the web. We should collectively realize we don’t have privacy in public places, and well the web is just that, a public place.
I don’t specify this point to make a case for tracking without consent or as a call to end web privacy, but rather a general point of view that web users must consider and one that is even more important in the action of creators sharing their content on the web. I also am not making a case that creators shouldn’t have rights of ownership and control of use to monetize their content. Rather my point is that content that is posted and made available on the World Wide Web is in a sense for “everyone” in a “public place”. This purposeful and directed public action by creators is in contrast to the efforts of software like Nightshade that aim to restrict the usage of this public based content to match the limited purpose of the creator. I know this conflict gets to the heart of content ownership and the legality of rightful use, however there is a more simple answer. If creators are concerned with how their content is used then don’t post it publicly. If you can’t accept the public's use of your content then don’t offer it in the first place.
My concern with this “weaponization” is that it has an underlying tone that a creator can determine the way in which their content is used even in this public setting. The consequence of this extends far beyond the annoyance of bad AI models. In an extreme sense, it opens the door for creators to control the narrative around their content. Meaning they can silence detractors or adverse opinions, ultimately allowing public delivery without public commentary which will lead to a lack of accountability.
I know this sounds like an extreme case, however Nightshade sets the bar for it. Imagine a world in which politicians, authors, artists and so called experts can’t be referred to in competing opinions because the creator disapproves of the motivation of the referring author, and programmatically attacks the use of the detractor? Imagine public figures not being quoted, or the inability to fact check a politician's claims or the use of articles and science briefs in testimonies. In a sense that is what Nightshade does… it attacks the use of content outside the consent of the author even though the content has been made available to everyone.
In a world of so-called “fake news”, political biased statistics, pseudoscience shrouded by religion, deep fakes and a seemingly endless string of false facts, the world needs accountability. Public content needs to remain publicly accessible for all use cases, maybe even specifically AI. The value of AI has yet to be realized, but it requires the right nurturing to be successful. Weaponizing content to ruin training data will at the start limit or eliminate the value that AI could have or worse introduce bias as these models will end up without “real world” training data. We should focus on the best ways to ensure proper creator attribution and a share in the monetization, not technical means of restricting use.