Making AI Work For You
Amongst all the heated debate on artificial intelligence, it’s clear that the technology is rapidly becoming a fixture in our lives. We as creatives must consider how we can leverage it and protect ourselves.
At the start of the year, the Labour government published their review of the country’s AI opportunities. Commissioned by science & technology secretary Peter Kyle and written by entrepreneur Matt Clifford, the paper identifies Britain as the third largest AI market in the world and AI is proclaimed as the “government’s single biggest lever” to deliver its five missions for change (economic growth; a functional NHS; safe streets; opportunity for all; and becoming a “clean energy superpower”). Among Clifford’s recommendations, which include vague references of mitigating “the sustainability and security risks of AI” as well as the establishment of AI growth zones, is the setting up of a copyright-free British media data set. Kyle is hoping to implement this as part of a new data bill which would allow developers access to all copyrighted works to train their models, so long as the creative doesn’t opt-out. It is on this point that the House of Commons and House of Lords have been stuck butting heads, stalling a mammoth bill that would also provide bereaved parents access to their deceased children’s data and create a 3D underground map of the UK’s pipes and cables to speed up roadworks. Backlash to this part of the bill has been led by crossbench peer and former film director Beeban Kidron. “It is astonishing that a Labour government would abandon the labour force of an entire sector,” said Kidron, “My inbox is filled with individual artists and global companies who are bewildered that the government would allow theft at [this] scale.” This opinion is mirrored by head of the entertainment trade union Equity, Paul Fleming, who warned in an interview with PoliticsHome, that the government’s plans would “legalise theft” and floated the idea of industrial action similar the 2013 SAG-AFTRA strikes across the pond. Baroness Kidron has repeatedly introduced an amendment to the bill which would require AI firms to notify copyright holders when their work is used. This has been repeatedly supported by the Lords and then shot down by MPs. In a letter signed by over 300 industry titans, including Paul McCartney, Elton John and Dua Lipa, the creatives cautioned, “We will lose an immense growth opportunity if we give our work away at the behest of a handful of powerful overseas tech companies and with it our future income, the UK’s position as a creative powerhouse and any hope that the technology of daily life will embody the values and laws of the United Kingdom.” John went as far as to call the government “absolute losers,” who will “rob young people of their legacy and their income,” and Kyle “a bit of a moron” to the BBC’s Laura Kuenssberg.
The government are said to no longer favour the opt-out approach and other options are being floated, including making no change at all to allow the bill to pass, allowing AI firms access to copyrighted material with no opt out and creating licenses for AI companies to attain copyrighted work. It is clear that there is no one consistent approach to the issue as the debate must consider the side of the creatives, who need to protect their livelihoods, and the side of overexcited Big Tech firms and politicians, who view AI as not just key to economic growth but to the potential transformation of society. If AI is to be this force for unforeseen societal change, how do we protect the creative worker?
“It’s easy for us to pinpoint this as a discussion around AI specifically but, actually, this is something that has been floated around for as long as Silicon Valley has become this dominant space,” said Margarita Louca, digital lead of the fashion course at renowned London art college Central Saint Martins. And it’s true. People have been sounding the alarm for years about how our data is used by big technology companies such as Google, Meta and Amazon. (This is something that the UK and EU’s General Data Protection Regulation (GDPR) sought to confront in 2018, but the pace with which AI training models are being developed calls for more site-specific legal protections). Louca views the ‘opt-out’ model as a continuation of the canon of user-blaming, where the user is made responsible for how their data is used despite the deeply opaque mechanisms constituting these firms’ methods for scraping said data. When Meta introduced their AI tools, Louca made it her mission to try and opt-out of their training models. She soon discovered that this was no small feat, having to go through layer after layer of user preferences to restrict the company’s access. She explainedhow, “I think it puts the burden of labour on the user, forcing us to opt-out… Everything was buried behind clicks and this and that and they really rely on the fact that we won’t see these things through.” This is a scepticism shared by Joshua Rothman, writer of the New Yorker’s ‘Open Questions’ column who often covers the ethical concerns of AI. “I strongly believe that artists and writers should be able to opt out,” said Rothman, “but I’m not aware of a practical mechanism through which this could be achieved.” In order to circumvent this insufferable bureaucracy, Louca proposes that we reverse the process and make it voluntary. “My take is that the current model doesn’t work, opting-out again is not really acting in our favour because they know consumers won’t do that,” she argued, “So opting-in is the best choice.”
For Rothman, an opt-in system is not just a favourable option but something he already engages in. “I used to want to be exempted from the training data,” he said, “but I have been increasingly open to being included.” He explained how he would “like the models to know what I think, and to be able to answer questions about me. I see being included in the training data as an opportunity to have a wider influence.” This point on influence sits at the heart of the issue. In a landscape which has systemically devalued the rights of the creative worker, it’s difficult to see how one can excise any supposed influence. Companies like the big Silicon Valley firms, as well as China’s Alibaba and Baidu, hold so much power on the web, and so much money, that it can appear difficult to confront them. Although, Louca once again clarifies that this is just a further exacerbation of pre-existing problems for artists. “It’s not as simple as a handful of these companies. Obviously, they’re like the harbingers of doom but it’s something that artists have been fighting since before AI or any other issues with the Internet,” she told me, “Which is that it’s really hard to protect your intellectual property and it’s really hard to compete against this conglomerate if they steal your art.” In this moment, it feels almost Sisyphean. Creatives have always been fighting these uphill battles, AI just made the hill even steeper.
Again though, Rothman proffers a more optimistic viewpoint on the matter. “I’m not persuaded that [companies like] Google and Meta have monopolies anymore. We’re at a transitional moment where AI is upending a lot of settled hierarchies,” he said, “Perhaps AI will enable creative people to handle their day jobs more easily, increasing the amount of time they have to make art. It might expand their agency.” It is true that, increasingly, creatives are coming together to try and protect each other through open-source AI models. Spawning.AI, which was co-founded, among others, by artists Mat Dryhurst and Holly Herndon, operates haveibeentrained.com— a platform that analyses LAION-5B, the most popular image data set for AI, and allows users to identify where their images have been used to train large language models and instruct these models not to train using them. The for-profit company have also developed Kudurru, which tracks data scrapers’ IP addresses to block them or send back a middle finger. At the University of Chicago, computer scientist Ben Zhao has led teams that are literally poisoning AI models. Nightshade is a “prompt-specific poisoning attack” which pollutes text-to-image models, such as DALL-E and Stable Diffusion, to mismatch image and text. For example, it will shift a dog into a cat, or a hat into a cake. Zhao hopes that this will break the models, forcing companies to revert to older versions or simply stop stealing the data. Speaking to NPR, Zhao said, “I would like to bring about a world where AI has limits, AI has guardrails, AI has ethical boundaries that are enforced by tools.” Zhao’s team have also created Glaze, a tool that changes the pixels of an image ever so slightly so that it makes it harder for AI models to understand and mimic specific styles. Glaze has been embedded into Cara, a new online community that promotes man-made art, since December 2024.
To see creatives importing direct action into online space is an exciting and encouraging step in reasserting some agency. It’s the element of the AI debate that makes Louca tick the most. “Where I think it is interesting is when artists go in and re-wire it and try and break things,” she believes, “That’s where it gets really exciting. I think a lot of this stuff is happening just not in public spaces. I have hope.” Whilst these tools show us an emerging school of resistance to look forward to, there are vital limitations. Some are sceptical about the long-term efficacy of Nightshade and Glaze, as developers, or the models themselves, attempt to overcome the challenge. Furthermore, at least in the case of Spawning.AI, their models are hosted on Amazon Web Services (AWS), a variation of cloud computing resource which contain the algorithmic output of AIs. Cloud computing is required for these algorithms as they host the data, models and applications needed for them to exist. This layer in the AI technology stack is dominated by the big firms: AWS, Google’s Cloud Platform and Microsoft Azure. If the creative is still reliant on these big companies in their resistance, then it is difficult to see how we can disrupt these hierarchies without solid government oversight.
Back in April, creative collective Studio Halia hosted ‘Infinite Play’, “a symposium on the evolution of storytelling and creative process,” at Manhattan’s trendy WSA office tower. The event saw leading voices in this field, including writer and researcher Phillip Lyle, photographer Erica Snyder and Behnaz Farahi, creative technologist and assistant professor at the MIT Media Lab, share their thoughts on how creatives and brands should engage with the shift to AI. In a talk exploring how AI is changing the creative process, Farahi had some thoughts. “AI is a mirror. It tells you how you perform or what is your pattern of behaviour. It’s become a mirror to your flaws,” she explained, “Important work we can do [as creatives] is keep having conversations about where the blind spots are.” This is a position shared by Louca, who says, “The only way we can change anything is through conversation, through advocacy, through education. Anytime you introduce a tool, you need to explain that there are going to be repercussions. One thing I really struggle with is the lack of understanding that all digital stuff has a physical real-world impact.”
Dialogue is a central part of any struggle, but particularly when the struggle is as muddled as the one against the algorithm. Much like how many feared the death of painting upon the advent of photo-mechanical printing, a lack of understanding of AI and its real-world effects is stoking the perilous fears of many and hindering the ability for impactful conversations. The tension over the data bill aids no-one involved. It merely permits tech companies even more time than they have already had to affirm their power. It means that the unfair and even exploitative practices striking creatives down are allowed to persist. It distracts from the opportunity for AI to enrich society. But this enrichment can only take place with proper oversight on the technology, and inclusive discussions that don’t over- or under-hype what AI can do for us.