THIS week saw something that hasn’t happened in decades. On 1 January, the US copyright protection on a host of expired. Works from 1923, including silent movie The Ten Commandments and Charlie Chaplin’s film The Pilgrim, have now entered the public domain.
These works were granted a copyright extension by a 1998 law signed by US president Bill Clinton that retroactively added 20 years’ protection to works created in 1923 and later. That time is now up. If no new laws are passed, even more work will lose its copyright, including the original Mickey Mouse film, Steamboat Willie, in 2024.
At the same time, advances in technology are leading some to wonder about the future of copyright altogether. As artificial intelligence algorithms play an increasing role in media production, questions of ownership are becoming fuzzy.
Advertisement
Creating something with an AI takes three steps. First, someone codes the algorithm itself, then it must be fed masses of data to teach it to recognise and mimic patterns, and finally the AI produces some sort of output.
When it comes to copyright, the big question is who owns that output: the person who built the algorithm, the person who picked the training data or the person who selected the specific output?
In some ways, this isn’t a new conundrum: think of bands arguing over who should own the rights to a particular song when one member wrote the chord sequence and another the solo, says Tom Lingard, an intellectual property and technology lawyer.
Elevating current AI systems to the status of bandmate is probably going too far. Both artists and lawyers say they are more like word-processing programs: if nobody types into one, there can be no essay. The software might check your spelling, but the thing that makes an essay unique is the writer.
“Without an artist to actually collect the data, the algorithm has no agency”
By extension, we could say that anyone who feeds data into an AI owns the resulting work, and the AI’s creators have no claim to the outputs, says Andres Guadamuz, a lecturer in intellectual property law at the University of Sussex, UK.
AI artist Janelle Shane agrees with this view of algorithms as sophisticated tools for realising an artist’s vision. “It can seem like the machine is doing everything, but without an artist to actually collect the data, the algorithm has no agency,” she says.
That’s not the end of the issue. The data sets used to train AI generally consist of hundreds or thousands of images, songs or pieces of writing, some of which may be copyrighted themselves. Should the owners of these works have a claim on the output?
One argument is that this training process is equivalent to inspiration. “When musicians create music, they listen to music,” says Guadamuz. “In some ways, feeding a machine-learning algorithm music or images is almost the same.”
However, the act of a human listening doesn’t create an exact digital copy. By contrast, gathering a data set and feeding it into an algorithm requires the original work to be copied. Without the appropriate permissions, that is copyright infringement, says Lingard.
That doesn’t mean that people who use AI to create things are likely to be sued. “It’s really hard to prove anything in terms of the training data,” says artist Mario Klingemann. “[The AI] doesn’t cut out an eye and a mouth and make a collage of the new face. It learns to manipulate pixels so that a certain area looks like an eye.” That means it can be impossible to retrieve the original images once the AI is trained.
Of course, if your AI is solely trained on extremely identifiable works, you are probably asking for trouble. For example, an AI fed all of Dr Seuss’s children’s books will produce outputs that are very similar to the originals, perhaps even including Seuss’s made-up words. For AI, it is a thin line between inspiration and plagiarism.
Ideas machine
More advanced AIs could raise a thornier question: can an algorithm claim copyright on its own creations? Current laws suggest not. For example, the only recognises “the fruits of intellectual labor” that “are founded in the creative powers of the mind”.
Because previous US case law has established that copyright can only be extended to “original intellectual conceptions of the author”, it will only be granted to works with a human creator, meaning that AI-produced work can’t be copyrighted. “It’s sort of human fundamentalism, the idea that machines cannot do anything that we consider creativity,” says Guadamuz.
Things are different in the UK, where the 1988 Copyright, Designs and Patents Act , “the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken”. A few other countries, including India, New Zealand and South Africa, have similar laws, allowing a human to claim computer-generated artwork as their own intellectual property.
Giving an AI ownership of its work wouldn’t even make sense, says economist Marc Scheufen at the German Economic Institute. Copyright law is designed to encourage the production of new works by allowing their creators to profit exclusively from them and recoup their costs.
“As soon as you have an AI system that is able to create new art, the costs of producing art will be zero,” says Scheufen. “There’s no cost, so no incentive is needed.” In a world where everything is made by algorithm, there will be no need to compensate AIs for their work and thus no need for copyright.
This ease of creation could even end up being a serious threat to human artists. “You could get an AI to produce 20 million different songs and copyright them all and put them on some shelf and just wait until somebody else writes the song and sue them,” says Klingemann. “I don’t know if that’s the future that we want to have.”
The bottom line, then, is that copyright law is not yet equipped to handle the enormous amounts of data these sorts of AIs require and generate. “Right now, AI is not so sophisticated, so it doesn’t have much commercial value,” says Guadamuz. “When it crosses a certain threshold where we’re going to have more sophisticated art or music or poetry, we’re going to have to make a decision.”
Yet serious money has already been paid for AI-created works, as demonstrated by the sale of an AI-generated portrait last year (see “AI creations for you to enjoy”).
It is likely that the question of AI copyright will be answered by some future lawsuit that sets a precedent and trickles through the courts. If the ruling is that AI art cannot be copyrighted, it could kill the genre entirely as artists refocus on work that can pay their rent. Yet if the law decides that such art can be protected, it could damage other methods and industries as AIs flood copyright offices with millions of applications and simply wait for someone else to infringe them.
Regardless, this difficult decision is coming up fast. The key issue is whether a machine can truly create art on its own. If we decide it can, that opens the door to even larger consequences: if an algorithm can own copyright, what other rights should it have? It is a question that could barely be imagined in 1923.
AI creations for you to enjoy
Portrait of Edmond de Belamy
This blurred and unfinished portrait bears a mathematical equation in place of the artist’s signature – a hint that it was created using an algorithm. The trio of students behind the work trained borrowed code on 15,000 existing portraits, sparking questions over who really produced the work. It was auction house in New York last year for $432,500.
Zone Out
Director Oscar Sharp and AI researcher Ross Goodwin tasked their AI, Benjamin, with producing an entire short film in 48 hours. They fed it footage from public domain films, screenplays and footage of human actors. The AI generated a script, pasted the actors on to existing clips, and produced its own voiced dialogue and musical score. The result (above) is .
Harry Potter and the Portrait of What Looked Like a Large Pile of Ash
Botnik Studios uses a customised, predictive text keyboard to produce parodies of books, TV shows and other media. By feeding its software with the Harry Potter novels, Botnik created a , which includes Ron eating Hermione’s family.
This article appeared in print under the headline “Creative differences”
