​Theft and compensation: on news publishers and AI models  

Large language artificial intelligence models are fuelled by content on the Internet, and much of this content comprises news reports gathered, curated and published by media professionals and organisations with decades of experience. As creative industries reckon with their labour getting diffused into unaccountable clusters of graphics processing units that reproduce styles and spit out human-level artwork in mere seconds, the news industry has reason to fear the compounding of permission-less innovation into an existentially threatening heist of several lifetimes of work. Previous waves of digitisation peeled away captive audiences from print and broadcast media by replacing these with a web-charged attention economy, and Big Tech platforms further squeezed news media’s place in these rapid transformations by often short-changing the very sources of information that their businesses relied on to be useful to the public. In a landscape where even precarious business models in the Internet age are threatened by a reluctance to pay for news and declining public trust in professional news-gathering, AI may very well be a body blow. It is clear: AI firms, with their billions in market capitalisation, must not be permitted to just take what they want from the Internet, synthesise these inputs into monetised insights, and pretend that the whole process is a form of victimless, innovative progress. Publishers have a clear right to decide who gets to hoover up their entire corpuses, and to ensure that their businesses benefit from the AI wave. To this end, the Department for Promotion of Industry and Internal Trade’s committee on copyright and AI is a welcome step.

This is not a decelerationist, or “decel” demand, as those who advocate for rapid AI development might be quick to say. The news industry has fought as search giants and social media companies profited enormously on the back of its content and set the terms for how the financial benefits flowed back to it. That cannot be permitted to happen — as social media platforms turn more and more into video-focused walled gardens, discouraging even a step outside their apps. For the news organisations, the avenues to earn are shrinking. As AI-generated overviews of news content with source links are reduced to a footnote, it is time for compensation to be negotiated at the time of publishers’ content being scraped from their websites in the first place. AI firms may claim “fair use” in model training, but there is nothing fair — morally or legally — about accessing and disseminating troves of news without taking the creators and processors into confidence. News publishers and policymakers must now fight for their share in the AI era.