Why would a valuable media brand want to be associated with a hallucination, or any kind of low quality knock-off? If ChatGPT hallucinates some news which is false, before long, they’ll find themselves in court. But publishers have legal obligations to consider. If ChatGPT is hallucinating a spaghetti bolognaise recipe for you, it might create one that’s a bit too salty. It “hallucinates”, and this can have catastrophic reputational consequences. The output of generative AI, like ChatGPT, is still very unreliable. There are other eminently practical reasons for refusing to take the bait. The AI systems only need to ingest the material once, and after that, they have everything they need – they can create AI clones and train further models on the clones, and so on. And even if Google does make what seems like a generous offer, there are very good reasons to say no.Įxposing the depths of a picture library or the archives of a newspaper so they can be absorbed by a greedy AI slurper is a very bad deal, one executive with a long history of licensing newspaper content told me, because it’s a one-off agreement. Today, Google isn’t a plucky startup any more, as it was 20 years ago – it banks $280bn (£227bn) in revenue every year. The music industry is grappling with exactly this problem – what’s to stop Spotify promoting banal AI-generated music instead? But the deals on offer to publishers are actually much worse this time around. However, as Professor Mike Wooldridge – author of The Ladybird Expert Guide to Artificial Intelligence – acknowledged in a recent Telegraph interview, the AI operations may be running out of free stuff to scrape.įor their part, publishers are right to argue that giving valuable stuff away destroys the market for content – particularly when the rival is a low-quality clone. Pearson is also in litigation, but won’t say against who. These AI models have already scraped what they can, and are very coy about how they obtained it, although sometimes they are careless and give the game away – Stability AI didn’t even bother to remove the watermarks from the Getty Images it scooped up, which made it very easy for Getty to sue. The large language models which underpin generative AI services like ChatGPT and Midjourney are useless without training. They want content not just from newspapers, but the owners of image libraries, commercial illustrators, visual artists, and database owners. No, and why?īut history rhymes, and today Google and the AI companies are singing a very similar tune. The publishers regarded this proposition in the same way that Saturday shoppers might consider a compulsive exhibitionist streaking down Bond Street naked, demanding that everyone else strip off. If only those foolish publishers had exposed their material to Google’s search crawlers, this awkward episode would never have happened. The problem lay with the owners of newspapers, magazines, periodicals and books who had arrogantly – in the bloggers' view – refused to give Google what it wanted, on the terms it demanded. It wasn’t their fault and it wasn’t Google’s fault that this had happened, they protested. It showed that just as in Animal Farm, some animals were more equal than others. Googlewashing demolished the proposition that the new internet utopia would be a fair and an egalitarian one. Within weeks, alas, spammers had noticed too, ending that brief golden age when Google was a pristine and breathtakingly efficient source of knowledge.īut surely the bloggers must have been pleased with their discovery? Google had anointed these obscure self-publishers as Kings of the Hill, giving them extraordinary reach and exposure, at the expense of the biggest brand names in media. I gave this a name, “Googlewashing”, and the term went viral: newspapers including Le Monde and The New York Times weighed in. It was very easy: all they had to do was quote each other frequently from their own weblogs. Twenty years ago, a small group of utopian internet enthusiasts discovered that they could rig Google’s search results so that they came out on top.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |