It’s Kind of Scary but Exciting: Generative AI

It was around this time last year the company OpenAI launched its much ballyhooed protect: ChatGPT. Since then, AI has been a focal point in most discussions. The technology used in ChatGPT is called Large Language Model (‘LLM’) and chatboxes like ChatGPT generate content when exposed to large datasets, including books and websites. Only a couple of months ago, prominent authors filed a class-action against OpenAI for copyright infringement. However, copyright infringements related to generative AI is not limited to chat boxes. They extend beyond ChatGPT to music industry or image generation.

In light of the recent release of the Beatles’ song ‘Now and Then’ on 2 November 2023, in this blog we will discuss the tension between the use of generative AI and copyright infringements.

Written 45 Years Ago, Released Now: Now and Then

On 2 November 2023, a new Beatles song ‘Now and Then’ was released. The group, which publicly disbanded in 1970 following Paul McCartney’s announcement and tragically lost one of its member in 1980, has now released a new song featuring the voices of all its members. The question is, how? The answer lies in innovative technology: Artificial Intelligence (‘AI’).

John Lennon began writing this song 45 years ago and recorded a demo with piano and vocals at his home in New York two years before his assassination. Later, in 1995 the band members, Paul McCartney, Ringo Starr and George Harrison worked on the song with the intention to release it. However, at that time, technology limitations prevented them using Lennon’s recorded vocals.

Today, in 2023 artificial intelligence software has achieved what was impossible approximately 30 years ago. The AI technology, by removing background noise, improved the sound quality and isolated Lennon’s voice. Then, they mixed the record as it would be done in regular music production. Although this incredible technology opens up exciting possibilities to restore many old recordings, as Paul McCartney himself expressed, ‘it’s kind of scary’. In this instance, although the song reproduced was belonging to the group itself, the use of AI raises serious questions about intellectual property (‘IP’) infringements.

Copycat Songs

The use of AI, although in the case of Now and Then led to achievements beyond what technology in 1995 could accomplish, also poses the great risk of generating copycat songs. Recently in April 2023, Universal Music Group (‘UMG’), controlling more than 30% of the  global music market, instructed streaming platforms like Spotify and Apple Music to prevent AI services from extracting melodies and lyrics from their copyrighted songs. The Group represents artists such as Elton John, Taylor Swift, U2 and Harry Styles.

Generative AI is frequently trained on popular music. This, therefore, allows anyone to ‘compose’ a song with their preferred lyrics by blending the style of various artists. If we are to put this more into context, it allows you to compose a song using the lyrics of Frank Sinatra’s ‘New York, New York’ in the style of Mariah Carey’s ‘All I Want for Christmas Is You’ even if you hold no rights to either of these songs. Google’s recent development is a perfect example of this situation.

In the beginning of this year, Google announced its new experimental AI tool called MusicLM.  MusicLM is a model that was trained from a dataset comprising 280,000 hours of music and generates music from any text description. So, by putting a prompt like ‘upbeat pop for a workout session’ the model generates two versions the song and allows you to rate the versions which later contributes to the enhancement of the model itself. After its announcement however, Google decided not to release the product due to potential misappropriation of creative content. Alternatively, if we put it plainly violation of copyright. But what is the music industry’s standing against the generative AI.

The Industries Stance Point against Generative AI

It should be made clear. No one is opposing the development of AI. The issue lies in the unauthorised use of any kind of content, whether it’s books, music or even the information posted on websites.

Moreover, no one is disregarding or objecting the valuable creations that generative AI can produce, such as the recent Beatles’ song ‘Now and Then’ released three decades later due to technological advancements. Unfortunately, Now and Then is one in a million.

In all industries emphasis is placed on understanding the nexus between IP and AI as well as appreciating what amounts to unlawful use of recordings in any kind or form.

Prominent music corporations, such as Universal, will not hesitate to protect both their own and their artists’ rights in cases of copyright infringements. The firm has stated that it holds a ‘moral and commercial responsibility’ to its artists to prevent the unauthorised use of their works as well as to stop streaming platforms from incorporating content that is in violation of their artists’ rights.

What’s To Come

Globally, discussions on the regulation of AI are underway, notably in the US and the EU. While these efforts revolve more around establishing safeguards against fake content or election interference, the EU AI Act proposal introduces transparency requirements for generative AI. The proposed regulation mandates generative AI, so chatboxes likes ChatGPT or the way Now and Then was produced, to publish summaries of copyrighted data used for training. These requirements could potentially mitigate copyright infringements. Here in the United Kingdom, although the government does not currently plan to enact a specific AI related legislation, it acknowledges the need for AI systems to possess a sufficient level of transparency.

Quoting Paul McCartney who remarked, ‘AI is kind of scary, but exciting because it’s the future’ we are awaiting the unfolding impact of generative AI’s involvement on our lives and legal disputes.