European Union: AI Observer, Issue 4: What If Generative AI Does Infringe Copyright? A Lesson From History

Welcome to the fourth issue of the HLK Newsletter tracking the AI legal landscape. In this issue, we are taking a look back at peer-to-peer sharing, a massive industry brought swiftly to heel by copyright law and update you on discussions regarding the AI Act, US Executive Orders and reflect on what – if anything- we learnt from the AI safety summit at Bletchley Park. 

The trials of Napster: a model for generative AI?

On a late summer's day, right at the start of a brand-new millennium, Silicon Valley got some very bad news.

Back then, the Internet still seemed young and limitless, zinging off in a thousand directions, alive with unimagined possibilities. Google was 2, Amazon mostly sold books and broadband was a word very few people even understood. Against this background, a boom in sharing was taking place. Called 'peer-to-peer' sharing, it felt like lending a friend a book or an album. Of course, things were changing from the physical to the digital, so it was natural that sharing went that way too.

Back then, the biggest name in sharing was Napster. In just over a year, Napster had acquired 20 million users. It wasn't the first service of its type, but it was user friendly and loved by music enthusiasts. Suddenly, albums that you wanted to buy – that you absolutely would buy if you could get your hands on them – you could listen to, right away. And if you wanted to share your love for a band, you could do that too. The platform was a rocket ship. In early 2000, universities were banning Napster simply because it consumed their network, accounting for up to 60% of traffic.

But while some big names like Courtney Love and Public Enemy came out in support of Napster, the music industry wasn't universally happy. More specifically, Metallica wasn't happy, which is sad but true. Dr Dre wasn't happy either – no diggity on that one. Trade bodies and record companies were particularly cross, and on a Wednesday in late July 2000, Napster was hit with a preliminary injunction by a San Fransisco court, who noted that "a majority of Napster users use the service to download and upload copyrighted music". Napster fought on for a while and is still a brand, albeit with a very different business model. But the dream of sharing on a massive scale seemed dead.

So where to from there? File sharing over the internet did of course continue, but became shadier. Entities such as The Pirate Bay sprung up, with or without veneers of respectability, to allow sharing using entirely decentralised BitTorrent streams. Copyright holders continued to pursue these sites in relation to copyright infringement, meaning that the sites tend to relocate frequently and the founders of The Pirate Bay even served time in prison. For the most part, while these sites saw plenty of traffic, peer-to-peer sharing of music and films died down to the internet savvy, losing its casual userbase.

But user desire was still there for an above-board way of getting hold of music at the click of a button, and a solution quickly emerged. Streaming services such as Spotify saved music lovers from trawling the depths of the internet for songs they were interested, and record companies switched focus from physical media such as CDs to digital. Ultimately, for users, it wasn't about avoiding payment so much as it was about getting hold of the music they loved easily. Users have embraced these services and while artists may abhor the rates, royalty streams at least exist, and function efficiently.

So where to for Generative AI – assuming that the large language models sued by big name authors and image creation engines in the sights of right holders lose in court? What would a more "copyright first" business model look like? First, opting works out of training data for some systems is possible, and its prevalence will likely grow, but it is currently technically and practically complex. It also seems a little legally dubious on its own. As a further idea, draft legislation in France proposes taxing AI companies to fund collective rights societies – although this will require artists to sign up. A "micro licensing" model, similar to Spotify and its ilk, might work although again in that case metadata would have to be provided to track rights ownership.

Users who have had a taste of the delights of generative AI are unlikely to want to give that up, but that doesn't mean they want to infringe copyright. Ultimately however, if the court cases go against the current models, history shows that the market will likely find a way.

Legislation watch

Over the last two weeks in the UK, we have had the AI Safety summit at Bletchley Part, as well as the King's Speech (in which the British monarch reads a speech written by, and setting out the legislative agenda of, the government). Despite the heralded 'Bletchley Declaration' on AI safety, those hoping for a clear insight into the UK's legislative plans in relation to AI have been largely disappointed.

Work on the EU AI act continues, with some possible use cases for real time facial recognition being entertained, rather than a near outright ban, as was previously the case, and more detail emerging on the obligations which may be imposed on the providers of powerful "foundation models".

In the US, The White House issued an executive order to establish standards for AI safety and protection of privacy, equality and civil rights.

Elsewhere in the world, it seems that Norway is to create a ministerial position with responsibility for digitisation and artificial intelligence. Germany has pledged significant resources to strengthen "the AI value chain" in a 12 point action plan.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.