Text, images and music made by machines are the newest field in art and the newest headache in the business of it. This week a large American newspaper tested AI-written articles and found them competent. For working artists the news is not good: the same tools can produce an imitation of any illustrator’s style in seconds, having learned that style from her work, without asking. Some call that theft. Investors, who have put billions into generative AI, call it the product. Three separate disputes will settle who is right.
1. Do the people whose work trained the model get paid?
In January a group of artists sued Stability AI, the London company behind the image generator Stable Diffusion, alleging it used their work as training data and infringes their rights every time it imitates them. “Artists should have a say in how their data is used, and they should be compensated,” one plaintiff, the cartoonist Sarah Andersen, told the New York Times. Getty Images has sued the same company in Britain and the United States for “brazen infringement” of millions of its photographs — pointedly, because Getty already licenses its archive for machine learning to companies that ask. Universal Music has asked Spotify and Apple Music to stop AI companies scraping its catalogue. Stability AI has not yet answered the claims.
2. Is training a machine “fair use”?
American law lets copyrighted work be used without permission for commentary, criticism and “transformative” purposes, and courts have generally treated computers reading text as transformative. Mark Lemley, who runs Stanford Law School’s programme on the subject, expects judges to be less sympathetic when the machine’s output competes with the artist it learned from. He proposes a new standard: training for the purpose of learning may be allowed; training for the purpose of reproducing a particular creator’s work may not. Unresolved either way: when infringement happens, is the user who typed the prompt liable, the company that built the tool, or both?
3. Can anyone own what the machine makes?
Copyright, so far, belongs only to humans. In February the US Copyright Office refused to register the AI-generated images in a graphic novel while allowing the human-written text and arrangement — treating the AI, in effect, as a hired illustrator whose contribution the author cannot claim. Some engineers say they will claim ownership of what their models produce; the office’s position, for now, is no.
For a designer in Dhaka
Bangladesh’s copyright law, like America’s, protects works of human authorship, and the 2023 Copyright Act does not mention AI. Two practical consequences follow. A logo or illustration generated entirely by Midjourney is probably unprotectable here — a client who pays for it cannot stop a rival using the same image. And a Bangladeshi illustrator whose style is being imitated by a model has, today, no court to go to; the fights above are being fought in London and San Francisco, and their outcome will arrive here as fact.
Source: The New York Times, via Prothom Alo.
