Appeared in Music Business Worldwide

In a world where exponentially advancing technology is created to solve various human basic needs, could the next frontier be the need for creative ideas? What’s the value of a computer-generated idea? But why buy an idea from a machine?

It’s a bizarre question at first glance. It reads like absurdist poetry. Yet we hold that our ideas are discrete entities with definable intrinsic value. Is it possible a machine could generate such a thing?

As AI slowly seeps into business, culture, everywhere, we are being forced to answer this question. If the results of the data ingesting, pattern recognizing, and predictions have some validity, they may qualify as ideas worthy of the same consideration as human-created ideas. After all, determining worthiness or value is a deeply human-inflected system, and as is AI.　Without human input, machine learning cannot happen. The machine is merely the surface layer of the ideation process. We are buying the distillation of human experience and perspective, processed on a scale unheard of before our time, using methods that feel alien and opaque--for now.

Accepting this possible answer--yes, I’ll buy a machine’s idea--somehow upends some of our notions of creativity. They shake their very core. The machine is not a person, not conscious, has no awareness or context. It has nothing to say. It has merely generated something. We are used to considering human artists as the driving force behind value. In the traditional definitions of artistic merit, the value of an object, utterance, or performance depends on the artist’s unique abilities and perspective. A machine’s idea is perceived as less valuable. After all, it didn’t really put anything into its creation.

Or did it? Our relationship with machines has been relatively one directional.  We’ve created technology to solve basic problems and only when said problems were solved, was technology paid for.  AI too is suited for this but, assistive creative AI, particularly in music and other artforms, introduces a new paradigm because the problems it attempts to solve is our limited creative nature itself. Many are conflicted, sometimes even offended by the demand pay for such technology, as the transaction would be an admission of defeat.

We are being introduced a new paradigm, a new relationship with machines. Creative AI by design is a combination of both human intuition and machine intelligence. This newly shared control principle frees us humans to imagine new creative processes introduced by the machine, not possible independently by either human or machine.    

We’re still struggling to understand the relationship between human and machine, just as we’re struggling to think of all humans as equals. A similar tension arose with the advent of photography. Was it really art, if an image did not have to be manually produced and could be reproduced fairly easily? It continued with film, which felt even more divorced from the “authenticity” and “aura” (to use terms critics kicked around in the early 20th century) of painting or sculpture.

When machines get involved in making art, it makes creativity more accessible, lowers the time between intention and execution (as limiting as it may be) and that democratization of creativity ultimately has shifted the center of commercial value. It moves from the prizing of a unique object or the restricted access to a certain live performance to the audience. This is a very natural source of value:  We love to give a number or value to objects and even more so to the people around us. We yearn to be valued, and we also yearn to value.

In a modern post-internet society, content value is post-creation. Machines can take an inhuman number of human hours and produce novel ideas in a very non-human way. That still doesn’t feel right to us. “You’re taking a millenium of work and giving me a random idea!” we want to counter. No matter how good that idea sounds, I may struggle to say that it is worth purchasing.

However, that is not where value is made now. In a world where every piece of music is equally accessible, the worth of a piece of music isn’t associated with the music itself, but its ability to attract listeners’ attention, the amount of time that people listen, share, talk about it. There are dozens of examples of mediocre artists who have large, passionate fan bases. Even if it’s a masterpiece, if they don’t share it or devote time to a track or album or video, it doesn’t have much value. It’s not about the artistic merit, the file type, or any other aspect of consumption. This dynamic is present, with or without machine learning, of course. AI is merely forcing us to reckon even more explicitly with the tension between originality and value, collaboration and consumption.

According to this theory, the hours put into consumption are more determinate of value compared to the endless hours put into production. I may have practiced longer but I may not be more skilled--or more able to produce something that attracts sufficient attention. Advances in tech can allow people to skip those long production hours and start creating, as these hours are not really rewarded (though they can be truly rewarding to the creator). The value of a piece of clothing or artwork is only quantifiable by the consumer, if they want to see it or take people to see it, if they want to wear it.

As AI becomes as normal to us as the drum machine and vocoder, as photos and film, we may change our answer. We may be happy to spend a few bucks on a machine’s idea, if it might pay our bills for a month, when developed into a hit. We have come to a new age of reckoning with machines in art. It’s a time that may completely reframe our understanding of artistry and value.

Amadeus Code is now open to all users. To celebrate, it is rolling out a new way to construct songs for composers unafraid to explore the possibilities of AI-assisted songwriting. “Our AI has the ability to find really unexpected, but yet compelling melodies and match them to the harmonies suggested by chord progressions,” explains the co-founder of the Tokyo-based tech startup Taishi Fukuyama. 

Harmony Library gives users direct access to the chord progressions that power Amadeus Code’s AI songwriting assistant.  These chord progressions can be searched by specific parameters, including genre, mood, tempo, and key, or by title or artist. Amadeus Code then generates an infinite number of melodies on top of the chords found in the selected progressions, melodies a user can tweak, dissect, regenerate, and eventually export to their favorite DAW. Composers can then share their creations with other Amadeus Code users and to the world. By providing the chords and tools that help shape the melodies granularly, users can proactively collaborate with Amadeus Code to create original ideas.   

“Radio, MTV, and streaming have all made an enormous impact on how creators were exposed to the music of their time. That has inherently shaped how and what they produced. Now that we’ve reached maximum capacity to consume the content we actually have access to, it’s only inevitable that we’re now getting creative and taking control over what ultimately influences our work,” Fukuyama notes.  The founding team of Amadeus Code are active professional music producers with experience producing some of the most famous artists of Japan and Korea, and a technologist and serial entrepreneur in music tech.

“Amadeus Code, unlike existing music AI on the market today, is not intended to create finished works to put against a home video, for example,” Fukuyama adds. “We hope that users will complete the work started by Amadeus Code by telling their own unique stories, which will continue to be what makes music truly irreplaceable by artificial intelligence.”  

This is not a copy-and-paste operation for hit making, however. This is about more than robo-songwriting. It’s about using algorithms to suggest unfamiliar melodies and expanding one’s imagination efficiently. “AI has this peculiar ability to find novel solutions--some successful, some not so much. These are suggestions which a composer can take or leave,” says Fukuyama. “Its decisions can spark a new idea for a composer, getting her into new creative territory.”

Appeared in Performer Mag

When you give musicians new technology, they find new ways to congregate around it and create new music--and new business opportunities. The next technology poised to do this is AI.

The rock band as we know it wouldn’t have sprang into being, were it not for the electrifying of guitars and basses. The 2 MCs and DJ with the turntable format gave birth to what we now call hip hop. The 808 and the mass accessibility of production software sparked EDM, as well as a whole slew of new genres and scenes. Musician-focused cutting-edge tech is what keeps music evolving.

AI is about to do something similar to the drum machine, to the DAW, to turntables, to name just a few recent innovations impacting the sound of popular music writ large. Just imagine: looking at history and what musicians have done with these creative technologies, we’re likely to be on the brink of another moment in musical history parallel to the birth of rock or hip hop.

Like previous revolutionary tools, AI is not, let us emphasize, not about to replace all songwriters, composers, audio engineers, or any other number of humans creating music. In fact, with the advent of the next world-dominating popular genre or culture comes a net positive economic impact for the industry and creators. AI may create new lucrative opportunities, just as technologies like sampling launched a whole new sub-industry for beat makers and licensors. The tools just need to be made and harnessed by the right creators. At least, this is what history tells us.

A carefully structured collection of algorithms and data, AI can perform specific tasks at lightning speed and digest large amounts of information in ways extremely challenging to individual humans. Because of their speed and scale, AI processes can uncover complex patterns and come to unexpected conclusions.

In creative industries, the pattern-finding side of AI is important, in that machines are learning to perceive things we grasp intuitively, but not consciously. These patterns can be employed to generate novel iterations of set forms, like a sequence of tones that make a melody (what we have worked to get our AI to do at Amadeus Code). They can also find what deviates in set ways from a pattern, something truly random and perhaps inspiring in its unexpectedness.

The peculiar answers AI programs can generate are sometimes the key to their success in beating humans at their own games. Alpha Go, the DeepMind-powered machine that beat the world champion at the complex game, did one small thing in every game it won: It made a move that the master player thought was ridiculous. It was so out of the blue, so out of the usual context and rhythm and practice of the game, he thought it was a mistake. Instead it won the computer the game. Yet humans do this all the time: How many chess, say, or gymnastics moves new to the ancient field are named for people who suddenly, inexplicably come up with them--and perfect them until they work?

This is in fact the often overlooked wonder of AI: It can come up with things no person would come up with, over and over and over. It can change our perspective through these wild suggestions, these seemingly ridiculous moves. It promises to do this very thing in popular culture, especially in music.

Yet it is an intimately human endeavor, even as melodies come to us from the interplay of algorithms and the rich melodic data of the past. We’ve discovered this, as we’ve taken our tools and used them to create finished songs.

We generate melodies. But then we have to take the next steps, and here is where things get interesting. As we’ve created  demo songs, we’ve gotten questions from singers: How do you want me to sing this? A human composer will have the answer ready, as she wrote with a motive or purpose. She can describe the topic, the emotion, the story of the song. But there’s nothing behind a computer’s melody. That’s why directing the vocals for demos was an epiphany for us. In the end, we had to ask, what was this song actually about?

AI cannot automate this connection. My AI, for example, will not tell you about its struggles. It will not illuminate your inner worlds. It can only expand your memory and give you some ideas that might allow you to express these things in a novel way. That may become your next song or an entire new genre of popular music. It can give you an infinite string of notes; only you can interpret them.

This is the right role for AI, the most inspiring and gently disrupting place for this new technology to push culture in a new direction. This could be a very welcome thing in a time when many are complaining about the narrowing of sonic ideas in pop, perhaps due to the data-driven nature of labels’ approaches to artists and tracks. Having computers generate whole tracks, as some AI engineers propose, or write whole albums or programatically switch a track from one genre to another… these are cool gimmicks. They aren’t going to give musicians that new cutting edge that leads to radical departures from what we hear today. AI can and will do so much more--and will benefit many creators in the process.

Imagine you have thousands upon thousands of melodies, going back to the 17th century, to create a vocabulary of song to help inspire your own music making. Press a button, and a new tune springs out of this rich set of ideas.

That’s how AI-powered app Amadeus Code works to spark creativity. Taking centuries of songs, it divides them into their smallest units, the atoms that make up a groove or lick. Then it uses algorithms to generate new tunes and chords based on era, rhythm, and range. The approach that powers the algorithms sets AC apart, finding novel ways to understand melody and chord structure that have nothing to do with scores or audio file analysis. The generation engine then flows seamlessly into an easy-to-use mobile interface for creative prompts on the go.

“We all have those moments when we’re looking for a little push, that nudge that gets us to explore fresh territory when we’re writing or working on a new track,” explains Taishi Fukuyama, co-founder and COO of Amadeus Code. “We imagine musicians and producers using Amadeus Code the way designers use idea boards, as a way to keep the creative juices flowing.”

CEO and founder Jun Inoue got the idea for Amadeus Code in a flash of insight in the shower, in a period of his professional career as an electronic music artist and producer when he had to come up with what felt like an endless series of melodies for clients. “About seven or eight years ago during my busiest period as a music producer, I was having trouble coping with an increased demand for hit songs from my clients. Continually churning out songs that are not only better but also different to the previous ones is a really hard thing to do. At that time, I was also unable to take sufficient breaks or enough time to recharge my creative batteries. Then I was struck with an idea that AI could come up with various ideas for melodies by analyzing all the hit songs of the past and thus assist me in the songwriting process. I actually came up with an image of a system that would enable this.”

Amadeus Code built this system with methodical attention to what makes a good song. First, the researchers who joined Inoue established a lab that carefully laid out what songs could be considered hits, taking into consideration several centuries of composing. They then determined the smallest possible unit of music that could still hold meaning--the phonemes of musical language--and that could combine to create larger melodic “sentences.”

There are distinct challenges to generating melodies and chord sequences to make music. Though Inoue and his colleagues put themselves into composers’ and musicians’ shoes, AI isn’t necessarily perfectly suited to the ultimate goal of Amadeus Code, to give songwriters viable ideas. “AI is proficient at searching for a single correct answer from among countless candidates. However, our biggest challenge was to provide an infinite number of different products to an infinite number of people,” says Inoue. “It was difficult to come up with an algorithm capable of doing this with hardly any precedents available.”

By perfecting an interlocking series of algorithms and creating a proprietary system for building melodies from the basic “licks” of selected works, Amadeus Code has solved this challenge, opening up AI-generated song ideas to anyone with a smartphone.

The app lets users adjust the AI-generated songs in a wide variety of ways. You can limit the era of the melodies it draws on--even projecting possible future hits by setting the parameter to 20xx. You can try more sustained notes or go for rapid-fire staccato. You can play around with lines that sound familiar, using the most common pop song patterns, or that veer off into wild and unexpected places, drawing on less popular approaches.

Once you hear something you like, you can save the entire song or pop out a section, asking the app to regenerate a new song based on that piece. “You can put together a collection of melodies that you connect with in your Song Library,” explains Fukuyama. “Then you can export them as audio and MIDI files to your favorite recording or production software and play around with them a bit more.”

These nuggets may inspire a new twist or turn in a song when you’re stuck. Or they may get you writing something in a whole new style. “We see this as a way musicians, songwriters, and producers can seek inspiration from some of the best songs in history,” says Fukuyama.

“This experience taught me that there was a lot in common between songwriting processes and genetic evolutionary processes,” reflects Inoue. “It convinced me that the feeling that encourages musicians to write songs was essentially the same as humankind’s urges to share and change, and thus ensure high-quality evolution. This astounding fact is one of the things that I want to share with everyone.”

Music composition and songwriting are often described the way other types of creativity and artistic processes are: mysterious, specialized activities that spring magically from the depths of the artist’s mind. Yet writing music is also a craft, and seemingly effortless breakthroughs are courted by skilled writers via practice, listening, analysis, and other methods.

Amadeus Code was built to be the next tool in this trade, the next supporting element to push the craft that leads to sonic art. By harnessing the fast calculations of computers and the rapid pace that algorithms can use to find novel combinations according to specific rule sets, AC is a prompt, generating new melodies and chords composers can use or discard, combine or arrange.

This is merely an extension of techniques creative people have used for eons. Creativity does not leap from nothing, but is the process of collection and selection, of building snippets and snatches into fully developed musical statements. From a folk song that becomes a symphony, to notes jotted down on a scrap of staff paper, to a hasty voice memo recorded at a bus stop, composers have been collecting ideas and using them to build masterpieces since formal composition became regular practice. AC simply adds another source of ideas to the mix.

The rules that guide our algorithm were distilled from data pulled from several centuries of popular works, from the Baroque through the top hits of the moment. This data had to be translated into units an algorithm could work with, which meant determining the smallest possible piece that still contained sufficient information to be recombined into a novel melody. In our machine learning model, we called this unit the “lick,” the atom of our song-generating system. We then assigned each lick four values that would allow us to evaluate their potential sequence in a melody, thereby generating convincing melodies successfully. These melodies are matched with chords.

Effectively, the great hit songs yet to be written are the great hits of the past--with a twist. For this reason, we took the so-called “Noisy Channel” approach, in which we told the algorithm to switch out a random lick in a hit song and replace it with another, random lick from a selection of likely replacement candidates, defined by another algorithm. We set a beginning and end for the melody; otherwise, the algorithm would generate melodies into infinity. And we made the results conform to Western notions of harmony, a key part of the recursive, self-improving side of Amadeus Code.

Recursion allows AC to expand its database with limited knowledge resources, accruing a collection of excellent, unique melody fragments to draw on. Human curation of these elements--which are kept by a user and which discarded--determines what is good. No machine can do that, of course. We are using the AC algorithm to gather the basic tendencies of melodies and chords, based on a wide range of musical styles and eras, and put this power in the hands of composers, who have always relied on outside inspiration to refine their visions. (See more details of our model here for those interested in the full complexity of song idea generation.)

The melodies Amadeus Code creates are helpmates, not replacements. Humans are the heart of creation and curation, and there is no machine that can change that. However, we feel strongly that AI has a thrilling place in the composition process, one that we hope producers, writers, and composers will adopt and adapt to their own needs and dreams.

About Jun Inoue
CEO, Amadeus Code - Music Producer and inventor of Amadeus Code. Studied at Berklee College of Music and J.D. from Keio Law School. Doctor of Jurisprudence (professional).