A think tank has developed a tool to find original content stolen by artificial intelligence. It also assesses the fair compensation owed to authors. This has thrown a spanner in the works for the business model of AI operators.
Will we ever be able to reconcile robots and content creators? The standoff between these two links in the artificial intelligence (AI) value chain, a market estimated at $700 billion by 2030, has turned into a sterile trench war. On one side, there are operators of generative AI (conversational agents like ChatGPT, but also image, video, and sound generators, etc.) who, under the "right to search," harvest web content (often protected by copyright) to feed their language models. On the other hand, creators (artists, publishers, media, etc.) who denounce the greatest plundering of intellectual property in history, for the benefit of a handful of technological giants.
The equation, which is readily presented as insoluble, an existential threat, or a necessary evil for human progress, was not avoided at the AI Action Summit held from February 6 to 11 in Paris. This initial argument was quickly dismissed by Sam Altman, the head of Open AI, who insisted that any "pro-culture" attempt was synonymous with anti-innovation.
Not so fast, retorted more than 34,000 artists, a thousand media outlets, and 25,000 journalists in a forum, recalling a few basic principles in their eyes: the possibility of closing the door to crawlers, the traceability of information sources, and the payment of content exploitation rights. "Impossible," the AI operators tirelessly respond, citing both the survival of their sector and the technical impossibility of meeting these demands.
Really? "False," argues economist Vincent Lorphelin, founder of Controv3rse, a think tank and research group on the digital economy, which, on behalf of the French National Assembly, co-led a report on the remuneration of rights holders by generative AI operators. One building block was missing: the prototype demonstrating the technical feasibility of fair compensation. With the support of a "repentant" engineer (who had worked for Gafam), the think tank therefore unleashed its "reward simulator" (a compensation simulator) at the Paris Summit.
"The iTunes Moment of AI"
The result is astonishing: upload an image generated by AI (a cat or a rose, as in our illustration), click. And in a few seconds, the interface returns a list of rights holders whose work contributed to the synthetic artifact. It also suggests the equitable distribution of copyright. The concept was built on an open source database of 10 million photos. But it could run on any database, such as those of rights management organizations, with images, videos, text, audio, mixed content, etc.
A slap in the face for AI operators like OpenAI, Google, and Meta? "We're providing a basis for negotiations for collective management organizations here," adds Vincent Lorphelin, who sees this as "the iTunes moment of AI." "The period we're currently experiencing with AI is comparable to what happened with Napster in the world of music publishing. There were three phases. The first was massive piracy, which led to a collapse of the music markets. It was a time of shock. Then came the legal and regulatory phase. We tried to develop a kind of internet police force (with Hadopi in France), but that didn't stop piracy. The pivotal moment came when iTunes and Spotify proposed a new economic model that brought everyone together. The solution was therefore entrepreneurial. We're now in the second phase for AI."
"If we go along the lines of Musk-Trumpism, AI is there to best extract know-how, knowledge, and culture. We call this cultural capital or human capital. This raw material, which Europe abounds in, is extracted to be concentrated in Gafam models. We are witnessing a concentration of wealth. This cognitive raw material seems elusive. It doesn't fit well into the economy; we can't give it value. But the promise of AI is to harvest all of this. The message hasn't gotten through yet, but in reality, the entire cultural and human heritage will be affected."
With the exception of a few deals between AI operators and press publishers (between OpenAI and the Associated Press, El Pais, Le Monde, Die Welt, etc.), "everyone remains stuck in a game of political posturing, where everyone defends their rights, and will get lost in the legal arena. We won't get out of this for many years," notes the co-founder of Controv3rse. "We need to prepare the entrepreneurial solution now. We just need to make sure we're smart enough to ensure the solution is European."
How much is this cat worth?
Subsidiary question: how much is this raw material worth? "Here, we have a tool to inform the public debate. The second step is to analyze the value chain and assess how much content contributes to its commercial success. This is where we invent the economic model, to share the pie. They're not making a profit yet? That's not a problem. Their stock market valuation alone gives an idea of the value of our author stocks." Economist Ernst Fehr thus estimated the value that press agencies and media contribute to Google Search, the precursor to IAG, at 14%.
The verdict on AI learning content? "15%," suggests Vincent Lorphelin. It's up to rights management organizations to collect and redistribute it. "To take a more established comparable, the 15% rate is the one that oil-producing countries apply to the price of a barrel. It's been said often enough that intellectual property is the oil of the 21st century without drawing inspiration from it."
Afterwards, "we're at the thick of it." "But not enough to disrupt the Gafam business model," assures Vincent Lorphelin. "AI operators aren't in danger. It's just that it bores them. Today, the markets are still unclear, uses are highly variable, and competition is not yet established. So, market segmentation is still chaotic. We're in a vibrant phase where everything is being invented and reinvented every two days. We're starting to realize that ChatGPT is worth around $20 per month. That's an order of magnitude."