Creative AI
Artificial Intelligence (AI) is no longer a thing of the future; it’s now a part of our everyday lives. From self-driving cars to voice-activated assistants, like Alexa and Siri, AI is becoming increasingly integrated into our daily routines. Creative AI has recently exploded onto the scene, cultivating huge interest in how AI can democratise creative processes. 2021 saw the launch of DALL-E which made waves as a text to image generator. Developers Open AI took DALL-E one step futher in 2022, releasing an updated version capable of generating more realistic images with a higher resolution – DALL-E 2. But 2022 ended in a blaze of glory for creative AI with the launch of ChatGPT, also developed by Open AI. Major tech industry players were left playing catch up at the start of 2023 with mixed results. Google’s answer to ChatGPT, Bard, went down in flames when it was revealed that the system was answering questions incorrectly, during it’s launch demo – ouch! And Microsoft followed suit with a new version of Bing (Bing Conversational Experiences), claiming it to be more powerful than ChatGPT. None are completely faultless however, with many reports of the sharing of incorrect information, argumentative responses and areas of ethical concern. (If you haven’t heard about New York Times journalist Kevin Roose’s interaction with the new look Bing, you can read all about it here! And here!)
One area where AI has been making huge waves recently is in the music industry. With AI-generated music gaining more and more attention, many people are asking: What is all the fuss about AI music?
What is AI music?
AI music refers to music that’s composed or generated using artificial intelligence technology. Rather than being created by a human composer, AI music is created by algorithms that analyse data and use machine learning and deep learning to generate musical compositions. AI music can range from simple background music to complex compositions that mimic the style of a particular artist or genre. AI systems are trained on big data – huge datasets which contain large quantities of quality, labeled data. In the case of AI music systems, we are talking about large quantities of existing music, composed by humans, labeled by genre, tempo, melody, harmony, instrumentation, and so on. AI systems will use these labels to learn about how music is constructed and how to generate its own tempos, melodies, harmonies, and so on.
It may feel like AI music is only really starting to make a splash now, but it has been around a lot longer than most people think, with a UK startup leading the way! Ed Newton-Rex and Patrick Stobbs met as choirboys at the age of 8, then reunited as students at Cambridge University, from whom they secured £120,000 of funding to develop a computer that could write music. Newton-Rex started devloping this idea in 2010, and in 2014 convinced his long-time pal to leave his job at Google and join him in founding Jukedeck, an artificially intelligent music composer ‘that writes original music completely on its own‘. Jukedeck’s platform allowed users to set a few parameters such as genre, tempo, length of track and instrumentation before clicking a button to generate a track faster than a human could ever hope to compose one. Added to this was the bonus that the track could be downloaded and used for free. Jukedeck’s AI-generated music has been used by popular YouTubers and businesses such as Google, Coca-Cola and the Natural Histroy Museum. Jukedeck was acquired by Bytedance (TikTok) in 2019, but it really paved the way for similar businesses to find success all over the world: Amper Music was acquired by Shutterstock in 2020; Boomy launched in 2019 allowing users to generate AI music and upload their “creations” to Spotify in one swift movement; launched in 2016, AIVA has the capability to generate full orchestral scores for audiovisual content; and dozens more companies have emerged offering similar services. This is, of course, not to mention the various AI-powered tools that are now available, such as LANDR, Starmony, Audialab and WaveAI, intended to ease and speed up a music creator’s processes.
What are the advantages of AI music?
One of the main advantages of AI music is its efficiency. Unlike human composers who may take days or even weeks to create a single piece of music, AI music can be generated in a matter of hours or even minutes (in some cases, seconds). This speed and efficiency make it easier to produce a large volume of music quickly and at a lower cost. Additionally, AI music can help automate the music creation process, freeing up time for human composers to focus on more complex musical arrangements.
Another advantage of AI music is its versatility. AI-generated music can be tailored to fit a specific mood or genre, making it ideal for use in film, television, advertising and even gaming. AI-generated music can be tailored to specific projects and can be adaptive; adaptive AI music is becoming more popular in gaming where music can be generated in real time to match game play, creating a more immersive experience. Another use for adaptive music is functional music – music designed to help you complete a function, such as running or sleeping – some AI music generators in this field even create personalised soundscapes which adapt to your heart beat or the weather. AI music can also help to fill the gap in the market for customised music, which can be expensive to create using traditional methods. Your YouTube video is 36 seconds long and climaxes as your feet hit the water after jumping off a cliff? Sure, AI music can be generated to the exact length of your video, and can provide a climactic musical moment as you hit the water. Best of all, in many cases, the music is royalty free! In other cases you can pay a small fee to own the copyright!
AI music can be extremely cost effective and time efficient for businesses, audiovisual creators and content creators as AI-generated music is produced at a lower cost compared to traditional music production methods, which typically involve hiring human composers, musicians, and producers. AI music can also be generated in a matter of seconds; human created music could take days or even weeks to materialise. On the flip side, from a music creator point of view AI music products can help to automate parts of the music creation process, freeing up time and resources for human composers to focus on more complex musical arrangements.
Perhaps most importantly, AI music products and services democratise the music creation processes, allowing absolutely anyone to unleash their creative side, regardless of their musical knowledge or training. AI music is accessible to non-musicians, making the music making process more inclusive and open to a wider range of people. AI music also has the potential to create entirely new genres of music and push the boundaries of what’s possible within traditional music creation methods. With AI, it’s possible to create music that combines elements from different genres and styles, resulting in unique and innovative compositions.
So, why am I only hearing about this now?
Well, yes AI-generated music has been around for a while, but to generate music from a machine is a far more complex process than say, creating a blog post with ChatGPT, or even creating an image of a starry night over Chelmsford or a painting of a woman and her baby in space. The results of AI-generated music have not always been that great because of these complexities.
The most complex issue with AI-generated music, when compared to AI-generated images or text, is the lack of data. There is a large amount of data available for AI models to learn from when training to generate images or text… an entire Internet of data, in fact. And while discussions around the legality of this are ongoing, access to music is far more difficult. Music is a highly complex art form, with many layers to be considered. As such, music files are also far more complex than an image file or text file and elements of a music track need to be cleaned and labeled before an AI system will recognise its parameters. AI systems cannot be trained on the song or piece of music as a whole, as you might do with an image, music must be broken down into melody, harmony, rhythm, dynamics, timbre, articulation, instrumentation… and the list goes on. The music industry’s copyright framework, in theory, is also pretty robust – the rules are supposedly pretty clear – and while some AI music companies may be using copyright protected music, either confused about the rules or flat out breaking them (more on that another time), this means that quality and well labeled popular music can be pretty inaccessible. AI systems can learn as much as they want from classical music which is no longer protected by copyright and sits in the public domain, but a system trying to learn how to generate various genres of pop music will struggle due to the lack of accessible data available.
There is also wide discussion around the meaning of creativity. Can a machine be creative? Music is considered to be a form of creative expression, and it can be challenging to teach a machine to be creative. Human composers draw on their experiences, emotions, and cultural backgrounds to create music, making it difficult to replicate this creativity in an AI model. Humans create music to evoke emotion in listeners, be that joy, sadness, nostalgia, excitement, anger… it can be challenging for AI systems to capture a mood or emotion and then evoke those moods or emotions through new music. Not to mention, music is entirely subjective. What one person likes about a song or piece of music can be hated by another. Some music is pleasing to the ear, some really is not. How can an AI system predict what will be pleasing? How can an AI system know what is going to be popular? As with all creative AI outputs, the results can be pretty questionable – perhaps more so with AI music.
What are the concerns around AI music?
Despite its many advantages, there are some concerns around AI music that are causing some people to be cautious. One of the primary concerns is the potential loss of jobs for human musicians. As AI music becomes more popular, cheaper, and faster to produce, there’s a risk that it may replace the need for human composers, arrangers, and session musicians. This may be particularly true for those just starting out in their careers, where the creation of library music and music for online content is not only their bread and butter, but also where they hone their skills and build their reputations.
However, while this may be a concern for some, it’s important to note that AI music can also create new opportunities for musicians to collaborate with AI algorithms in the creation of new music. AI-assisted music creation has been around for some time: Taryn Southern released an album called I Am AI back in 2018, a collaboration with Amper Music among others; Benoit Carré produced Beatles sound-a-like Daddy’s Car in collaboration with Sony Flow Machines in 2016; more recently breakout star d4dv began his ever more successful career producing music using BandLab; more and more artists are dabbling with AI-powered music creation tools.
Another concern is the potential for AI-generated music to infringe on copyright laws. This will be discussed in far more detail in subsequent posts, but the problem is three-fold:
Firstly, with AI music it can be difficult to determine who owns the rights to the music that’s created. There is currently a lack of consistent understanding among AI music companies, and among creators, as to: (a) whether copyright subsists in AI-generated music at all; and (b) if copyright does subsist in AI-generated works, who is the author of that work and is the owner of that copyright. This is partly due to a lack of harmonisation of copyright laws internationally, and partly due to the ambiguous nature of copyright law.
Secondly, there’s a risk that AI-generated music may sound too similar to existing compositions, leading to accusations of plagiarism and copyright infringement, particularly if the AI system has been trained using those existing compositions. Google recently released a paper called MusicLM: Generating Music From Text introducing its latest AI music model which generates high fidelity music from text descriptions and hummed or whistled melodies. It promises to yield some of the best results we’ve seen thus far from AI music systems. However, Google is yet to release MusicLM to the public as results show that at least 1% of its musical outputs infringe on music used to train the model.
Thirdly, whether AI music companies can legally use existing (human-created) and copyright protected music to train their AI systems is highly debatable. Again, differing approaches to copyright internationally and ambiguous laws make this a confusing concept. In theory, any reproduction or use of a copyright protected song or piece of music needs a licence – permission from the rightsholder and/or the payment of a licence fee. Some AI companies respect this possibility and create their own training datasets by employing human composers, some only use music in the public domain to train their AIs, some claim that it’s perfectly acceptable to use the music of others so long as the output does not plagiarise, and some simply do not disclose their processes (the AI black box).
What does the future of AI music look like?
It is extremely difficult to predict what this technology will look like in a year’s time, or in five year’s time. Who knows how good AI music could be in the future? Who knows how far away we are from autonomous AI music generation systems? However, despite the many concerns around AI music, it’s clear that this technology is here to stay and will continue to develop and improve. I expect we will see more AI-generated music being used in a range of contexts, from online content and social media, to film and television, and advertising and video games. It’s also possible that we may see AI music being used to create entirely new genres of music, pushing the boundaries of what’s possible with traditional music creation methods. We have much more to discover and much more to discuss, but one thing is undeniable – the future looks bright for AI.
Concluding remarks
AI music is generating a lot of buzz in the music industry and beyond, particularly in 2023 following the success of AI-generated images and text. While there are concerns around the impact of AI music on human musicians and on copyright laws, there’s no denying the efficiency and versatility that this technology brings to music creation. As AI music continues to evolve, it’s important for policymakers, industry leaders, and musicians to work together to ensure that AI music is used responsibly and ethically within the context of the music industry. With the right approach, AI music has the potential to revolutionise the way we create, consume, and enjoy music in the years to come.
Please share your thoughts in the comments!