Consider two music industry approaches to artificial intelligence (AI). One is taken by Giles Martin, son of Sir George Martin, producer of the Beatles. Last year, he used AI to remix the Beetles’ famous 1966 album Revolver. It learned from the original mono recording what all the band members’ instruments sounded like (John Lennon’s guitar, for example), and Martin was then able to separate them and reverse-engineer them into stereo.

The result was stunning. The second approach isn’t bad either. It features the response of the moody Australian singer-songwriter Nick Cave to lyrics written in his style by ChatGPT, an AI-powered tool developed by the startup OpenAI. “These songs suck,” he said. “Songwriting is not imitation, replication or pastiche, it’s the complete opposite. It’s an act of suicide that destroys everything man has tried to create in the past.”

Remember the Napster

Cave is probably not impressed by the latest version of the algorithm hiding behind ChatGPT, called GPT-4, which OpenAI recently introduced. Martin may find it useful. Michael Nash, executive vice president and manager of digital products and services at Universal Music Group, one of the world’s largest music publishers, illustrates with these two examples the excitement and fear that is generated by the artificial intelligence that powers content-generating applications like ChatGPT (text) and Stable Diffusion (image).

AI can help in the creative process. But it can also destroy it or completely hijack it. When it comes to recorded music in general, the rise of bots is reminiscent of another historical event that shook the world: the meteoric rise and fall of Napster, the turn-of-the-millennium platform for sharing mostly pirated song recordings.

Copyright law ultimately took care of Napster’s demise. For aggressive bot providers who are accused of going over the dead bodies of intellectual property, Nash has a simple message that sounds like a threat from a music industry veteran who remembers the Napster era: “Don’t go to the market to beg for forgiveness later. That’s what Napster did.”

The main problem here isn’t AI-created Cave-themed songs or fake Shakespearean sonnets. It’s the oceans of copyrighted data that the bots sucked up in between learning to create anthropomorphic content. This information comes from everywhere: social media feeds, internet search engines, digital libraries, television, radio, statistics, and so on.

AI models are guilty of robbing databases, often without permission. Those responsible for the source material complain that their work is siphoned off without their permission, credit or compensation. AI platforms are simply doing to other media what Napster used to do to songs: completely ignoring copyright. Lawsuits are starting to swarm.

Define fair use

It’s a legal minefield with implications that reach beyond the creative industries into any business where machine learning plays a role, from self-driving cars to medical diagnostics and factory robotics to insurance and risk management. True to its bureaucratic nature, the European Union has issued a copyright directive that also talks about data mining (it was written before the current bot boom).

Experts say America lacks precedents specific to generative AI. Instead, it has conflicting theories about whether data mining without permission is permissible under the “fair use” doctrine. Napster tried to defend “fair use” in America, but failed. That doesn’t mean the result will be the same this time.

The main “fair use” arguments are fascinating. Let’s quote a lesson on the subject published in the Texas Law Review by Mark Lemley and Bryan Casey: a use of a copyrighted work is considered fair use when it serves a valuable social purpose, the source material is transformed from the original, and the act does not in any way affect the core business of the copyright owner.

Critics argue that AI does not transform the mined databases, it merely uses them. They argue that the companies behind machine learning abuse fair use to cash in on individuals’ creations for free. And they argue that it is not just the livelihood of individual creators that is at risk, but the whole of society if AI starts to promote, for example, mass surveillance and spread misinformation. Lemley and Casey weigh these arguments against the fact that the more access to training data AI has, the better it is, and that without such access, no AI would ever have come into existence.

In other words, that the industry could go into decline before it properly comes into existence. They see the situation as one of the most important legal questions of the century: “Will copyright law allow robots to learn?”

The Getty pioneers

One of the first lawsuits, which attracted widespread publicity, was filed by Getty Images. The photo agency accuses Stability AI, which owns Stable Diffusion, of infringing the copyright of millions of photos from its collection in order to create an image-generating AI model that will compete with Getty Images. Since the case is not being settled out of court, it could set a precedent for fair use.

Even more important could be the verdict soon to be delivered by the US Supreme Court in the case of copyrighted modifications of the late artist Andy Warhol’s paintings of pop idol Prince. Daniel Gervais, a copyright expert at Vanderbilt Law School in Nashville, believes the justices could provide long-awaited general guidance on fair use.

The accumulation of copyrighted data isn’t the only legal problem that generative AI has to deal with. In many jurisdictions, copyright only applies to human-generated works. Thus, we enter another grey area when we consider the extent to which bots (and their owners) can claim copyright protection for their works.

Outside of the courtroom, the main questions will be political – in particular, whether generative AI should be subject to the same protection against liability for damage caused by published content as social networks, and to what extent generative AI violates data protection.

The fight over copyright will be a big one. According to Nash, the art industry should quickly take a stand to ensure that artists’ work is licensed and used ethically when teaching AI models. He urges AI owners to “document and publish” their sources. It recognises that this is a delicate balance. Creative people do not want to appear as enemies of progress.

Many of them may start using AI for work. The lesson to take away from the Napsterian “reality cure”, as Nash calls it, is that it is better to engage with new technologies than hope they will just disappear.

Maybe this time it won’t take fifteen years of watching revenues collapse to understand it.