With the rapid development of artificial intelligence technology, we are facing an unprecedented change. As the UK Parliament debates artificial intelligence and copyright, Paul McCartney, a member of the Beatles, has sounded the alarm: if we are not careful, artificial intelligence may squeeze young musicians out of the stage.
McCartney expressed new concerns about generative AI and its impact on the music industry. He warned that if artificial intelligence can continuously produce music that imitates human artists without giving the creators due credit or paying royalties, it will undoubtedly pose a serious challenge for young musicians. They may find it difficult to stand out, which is undoubtedly an issue we need to pay close attention to.
Currently, the UK government is considering amending data laws, and the government is considering allowing artists to opt out of the program where their works are used to train AI. If this bill is passed, artists will be able to refuse AI to be trained based on their works, thereby limiting AI’s ability to imitate these works to a certain extent. This is a positive move that will help protect the rights of artists and reduce the negative impact of AI on the music industry.
The Music Industry Landscape Has Changed Because of AI
The conflict between AI and music copyright is not only in parliament, but also in many social fields. Many applications that use artificial intelligence to create music, such as Tad.AI, Suno and UdiO, are facing lawsuits filed by major music companies. This has undoubtedly exacerbated this conflict and triggered widespread discussion about the boundary between AI and music creation.
Since this year, a large number of AI generative music tools have emerged around the world, and the industry landscape has changed dramatically overnight. Previously, AI music projects such as OpenAI’s MuseNet, Google’s MusicLM and Meta’s MusicGen have attracted widespread attention in the industry. The music model can complete the entire process of music creation, including lyrics, singing, composing, and soundtracking, at one time, so it is hailed as “eliminating” the threshold for music creation. Although you can find music inspiration with the help of AI, you will feel that they are basically pop songs after listening to them for a long time. In the future, the songs on Douyin may become more and more similar, and they will be produced like an assembly line. If you have to sit in front of a computer and input keywords to generate songs, the process may seem a bit boring.
At the same time, the case of the Beatles shows us the possibility of AI in music creation. By training on the sounds of John Lennon rehearsing, AI algorithms were able to complete unfinished works such as “Now and Then”. This undoubtedly proves the potential and value of AI in music creation. However, it also raises questions about copyright and royalties. If artists choose to allow AI to train on their works, should they receive corresponding royalties?
AI Lowers The Threshold for Creating Music
Another important reason why AI music has caused a boom is the reduction in the threshold and price of use. Take a certain AI music tool as an example. The basic version is free, the PRO version costs about $10 a month, and can generate 500 songs; the more advanced version only costs $30 a month, and can generate about 2,000 songs. This means that the cost of an AI song is only about 0.1 yuan. AI music is not perfect. AI can create, but it is not good at modification. Every modification is completely different, so more complex songs will still be done by humans. Despite this, AI music is technically superior to many pop songs on the Internet.
On April 2 this year, more than 200 well-known international musicians signed an open letter calling on AI developers, technology companies, platforms and digital music service providers to stop using AI to infringe and devalue the rights of human artists, and asked them to promise not to develop related AI music generation technology, and not to refuse to provide artists with reasonable compensation. There are 246 artists who signed the letter, most of whom are from the European and American music scene, including Billie Eilish, Katy Perry, Nick Minaj, etc.
The impact of AI on the music industry is two-sided. It can lower the threshold for creation and allow more people to try music creation; it may also cause some people’s jobs to be replaced by AI, especially artists who rely on simple creation and cover songs may lose job opportunities, such as some singers who sing demos have been replaced. With the continuous increase in music production, the number of songs that people can hear each year has far exceeded their digestion capacity. The emergence of AI has further intensified competition in the music market and seized the original music space. This may affect some musicians who are serious about making music libraries and content, and their works may be submerged in the vast amount of music due to the popularity of AI.
Music Big Models Face Higher Technical Barriers
Unlike language big models, which are widely used in multiple scenarios, music big models have been aimed at specific application scenarios since their inception – lowering the threshold for music creation, so that ordinary people can now get high-quality songs through natural language descriptions with the help of big models, which used to be work that only professional musicians could complete.
Compared with text and video models, music big models face higher technical barriers. Music, like video, is a long-time technical form. If video can be divided into 24 frames per second, music contains tens of thousands of sampling points per second, and each sampling point has a strong correlation, which makes music one of the most complex modes. Some products on the market that are currently named “music big models” do not actually meet the standards of big models in the strict sense. The generation of some works relies on the technology of music structure and rules, or is achieved by converting music into symbolic languages such as MIDI and then entering the model. Although this method can produce music at a near-human level, it cannot touch the essence of music – the understanding and creative expression of musical emotions, connotations and overall structure will never surpass the level of existing music.
A truly large music model should have end-to-end learning capabilities, with training data directly derived from complete music works listened to daily rather than translated MIDI, and be able to create an integrated process from conceiving lyrics, designing melodies, arranging accompaniment to simulating vocal performance. Unfortunately, most current models have only made progress in one link of music creation, such as melody and accompaniment, and have not yet achieved full-scale music generation. Moreover, users currently cannot see through the internal logic of the model, but can only see its external performance. The natural “black box” effect of AI has caused many cases of fakes in the music model.
