COMMENT: AI is out of control, we have the receipts | Copyright-denying OpenAI and Google pile pressure on UK | Angry authors to Meta: 'Get the Zuck off our books'
#48 | PLUS: Indy launches AI-enabled Bulletin | Music execs: 'watch out for quantum and agentic AI' | Musk's xAI swallows up X | ✨AND FINALLY: Matt Haig's new big hit, One April Morning
A WARM WELCOME to Charting Gen AI, the global newsletter read in over 100 countries that keeps you informed on generative AI’s impacts on creators and human-made media, the ethics and behaviour of the AI companies, the future of copyright in the era of generative AI, and the evolving AI policy landscape.
📣 COMMENT
Faked cartoon images and restaurant receipts are proof that AI is out of control
LAST WEEK social media was flooded with images imitating the style of Japan’s Studio Ghibli. The torrent came courtesy of OpenAI’s new image generator within its GPT-4o model that the AI developer said had been trained on content “reflecting a vast variety of image styles”. Ghiblified images were condemned on two grounds, (1) The apparent admission that OpenAI had trained its model on Studio Ghibli content, and probably every other studio’s content since images in the style of The Simpsons, Pixar, and The Muppets soon appeared; and (2) The fact that Ghibli’s founder, 84-year-old Hayao Miyazaki was on record as saying he considered AI-generated art as “an insult to life”. Studio Ghibli — famous for its hand-drawn animated movies Princess Mononoke, Spirited Away and Howl’s Moving Castle — hasn’t entered the fray with a cease and desist letter, though a fake one appeared this week generated by, yes, you guessed it, ChatGPT. Were the studio to assert its rights and enter into a licence agreement then it could be looking at an annual income of between $17 million and $20 million, according to Credtent, a leader in ethical AI and content licensing.
Something else we then learned: ChatGPT is really good at faking restaurant receipts. On LinkedIn digital entrepreneur Raphael Chenol posted a genuine receipt and an AI -faked one, complete with creases. “Yes, it only takes a few seconds where before you needed Photoshop skills,” said Chenol who stressed he wasn’t seeking to “set a bad example” but “denounce possible abuses”. What could be next? Faked photos on eBay, with items made to look perfect when in reality they might not exist at all? How about passports, driving licences, utility bills, all used to establish identity? Surely this breaches OpenAI’s policy of not sharing “output from our services to defraud”? TechCrunch asked OpenAI why it was letting people generate images of fake receipts. Spokesperson Taya Christianson said images contain metadata showing they were made by ChatGPT, but added it was the AI developer’s goal “to give users as much creative freedom as possible”. Those fake receipts could also be used to teach people financial literacy, create art and ads. So there you have it. An AI developer acting in a grossly irresponsible manner, refusing to impose guardrails on something that is bound to cause real-world problems. And why would it do this? Because it can. AI is out of control.
🏛️ AI POLICY & REGULATION
OPENAI AND GOOGLE have firmly rejected the UK government’s plans to water down copyright laws, essentially saying they don’t weaken them enough.
In its delayed response to a consultation that closed in February OpenAI said it backed a broad data mining exception, which would make it legal to scrape rightsholders’ content without consent or compensation. Google’s delayed response supported a version of the government’s opt-out for creators, but at the same time called for a new commercial research exception. Google made clear it did not expect to pay anything for training data. “To develop and improve our Al models, we train on publicly available data from the web. We believe the use of such content is non-expressive and that it enables innovation. As such, we believe training on the open web must be free,” said Google.
Speaking to Charting Gen AI Ed Newton-Rex, the ethical AI advocate and CEO of Fairly Trained, said: “The UK government has presented multiple options that would hand AI companies the life’s work of the country’s creators for free. It should come as no surprise that AI companies favour those options — who wouldn’t want their resources to have zero cost? The real question is why the government has put these options on the table at all, when they would be terrible for our world-leading creatives.”
Baroness Kidron, the independent peer who sought to reinforce current copyright law only to have her amendments removed by a Labour-dominated committee in the Commons, told Charting: “The frustrating thing about the consultation is that it asked the wrong question. These consultation responses from OpenAI and Google fundamentally add up to ‘we won’t pay’ for what they see as little more than raw material. The real question that the government should have asked is how do we build AI that will reflect the laws and culture of the UK, will help its workers — whether tech or creative — become skilled and prosperous, and will build a sustainable future for UK PLC?”
