Robo-photos are still making waves

THE QUESTION of the legality of robo-photos continues to produce headaches. Though the images produced by machine-learning systems are entirely irreal, they have real-world consequences.

In the US, back in September the US Copyright Office granted registration of the copyright in a comic book called Zarya Of the Dawn [update]. In the US, and only in the US, you have to register your works to get effective copyright protection.

The comic was written by Kris Kashtanova and illustrated by Midjourney, a machine-learning system trained on many millions of actual illustrations. In November the Copyright Office announced that it is reviewing the registration.

Crucially, review is based on the US policy that copyrightable works require human authorship. Several people have linked this to the notorious case that held that a monkey could not hold copyright in a self-portrait that it had taken itself.

Kashtanova is said to have concluded that the Copyright Office "overlooked" the lack of a human author of the images.

The law of the UK is unusual in that it specifically provides a form of copyright in works that are entirely computer-generated. The one court case that dealt with this wasn't an enormous help in determining what Parliament meant.

We take exception...

Also in the US, we noted earlier that programmers were planning to sue Microsoft over a machine-learning system pilfering their computer program code uploaded to its GitHub service. We are thankful to Neil Turkewitz for his analysis of Microsoft's outline defence. It claims that "OpenAI's training of Codex is done in accordance with global copyright laws which permit the use of publicly accessible materials for computational analysis and training of machine learning models, and do not require consent of the owner of such materials."

The Freelance shares Neil's view that the above is... not exactly the case. European Union and UK laws, for example, have an exception to copyright allowing such use for non-profit research purposes. The application of this is concerning image-makers: the LAION dataset of images, used to train that Midjourney machine-learning image generator, claims that it gathered images for non-profit research under the German implementation of EU law.

Labelling required

Meanwhile the government of China is reported to have issued a regulation specifying that from 10 January any machine-learning output - whether images, robo-voice, chatbot or virtual reality - must be "marked prominently to avoid public confusion or misidentification".

The Freelance suspects - from skimming a machine-learning translation of the announcement - that the main motivation for this is concern over the potential use of "deep-fakes" in anti-government propaganda. We hope this perception will not impeded the adoption of similar labelling requirements elsewhere: propaganda based on a fake is propaganda based on a fake whatever its motivation.

© Mike Holderness

• 8 August 2023 update: the Register of Copyrights withdrew registration of the immages in Zarya of the Dawn, stating that the original application did not disclose that the images were created by an "AI model". The text remains registered.

Repair robot rights! UK government pauses free mining for robots - and more US lawsuits