LIKE most New Scientist readers, you are probably a human. But, increasingly, publications like ours have a new breed of artificially intelligent consumers 鈥� an apt word, because these AIs don鈥檛 so much read as they do ingest, gobbling up vast quantities of text as part of their training.
Our human readers (we hope) derive value from reading New Scientist, whether that be for our informative news, mind-bending features, smart commentary or fiendish puzzles. In return for that value, all we ask is that you buy the magazine, subscribe digitally or even just let us show you a few ads. It seems a fair trade.
Advertisement
AI readers, or rather the researchers and companies behind them, also derive value from New Scientist. In a way, every piece of content consumed by an AI becomes a part of it, frozen in the numerical values that make up its neural networks. With this in-built knowledge, AIs like ChatGPT can spew out all sorts of text on command.
But New Scientist isn鈥檛 compensated for this value provision. As far as we know, no AIs have taken out a subscription. Yet the C4 data set, an enormous collection of 750 gigabytes of text used to train many AIs, contains around 7 million words or phrases derived from New Scientist content. You can also find great swathes of data from The New York Times, The Guardian and many other publications.
This, we believe, isn鈥檛 a fair trade 鈥� and we aren鈥檛 alone. Everyone from Twitter CEO Elon Musk to small website owners is increasingly feeling hard done by when it comes to AIs exploiting their data (see 鈥淗ow Elon Musk and Reddit are leading a war on AI web scraping鈥�).
What isn鈥檛 yet clear is the solution. When you, a human, read these words, you aren鈥檛 infringing New Scientist鈥榮 copyright, even if you have a perfect memory. But it is unclear whether AIs ingesting copyrighted material en masse could be subject to legal action (see 鈥淐hatGPT seems to be trained on copyrighted books like Harry Potter鈥�). The use of text to train AIs is nothing new, and until now no one has really cared about copyright because it was all small-scale research. With the latest generation of AI set to overturn entire industries, it is time to figure out who pays.
