- 29 Aug 19
Named alongside Cardi B and Lizzo in Forbes’ 30 Under 30 class of 2018 – and joining all-star alumni including Drake and The Weeknd – Drew Silverstein keeps good company. Although sitting comfortably with the brightest musicians, songwriters, DJs, agents, managers and entrepreneurs in the industry, Silverstein is first in class when it comes music’s next frontier – the use of artificial intelligence to create songs.
After a career as a film composer in L.A., Drew Silverstein moved to New York where he co-founded Amper Music. His mission? To combine the highest levels of artistry with groundbreaking artificial intelligence technology to empower anyone to create unique music, instantly. In 2017, Amper raised $4 million in seed funding and is now at the cutting edge of the race to crack AI music. Hot Press caught up with the Amper CEO in Rome, after his recent TEDx Talk, to find out what AI means for our future music consumption, how it will affect our approach to songwriting, how to collaborate with a digital version of yourself, the exploitation of intellectual property by Facebook and Google, the probability of robot composers with feelings, and why he predicts AI to be the greatest creative revolution in the history of music.
MARK HOGAN: What is Amper?
DREW SILVERSTEIN: Amper is an AI composer, performer and producer that creates unique and professional music in a matter of seconds. The music can be tailored to content or it can be standalone. Our mission is to enable anyone around the world to express themselves creatively through music regardless of their background, expertise or access to resources. Because fundamentally, every person is creative – by the fact that we’re people. But just being creative doesn’t mean we have the ability to express our creativity. Singing in the shower is easy. Painting a landscape is hard. Composing an orchestral piece of music is probably harder. So the challenge is not creativity, the challenge is expressing it.
Who will use AI in musical composition?
Ultimately, it provides even the most non-musical individuals – oftentimes video editors, podcast creators and other content creators who utilise functional music, or music that’s valued for its use-case more than its artistic creativity – the ability to take their idea and turn it directly into music. It democratises that expressive ability. On the other end of the spectrum there are creators of what we call artistic music; music that’s valued for the collaboration and creativity that goes into making it, much more so than just its use-case. These artists, musicians and composers already use technology in their process; Amper is the next generation of technology to help further their music-making, by fleshing out new ideas, enabling quick and easy experimentation, working through writer’s block, collaborating with digital versions of themselves…
How does it work?
Amper is designed to require a minimum amount of input to create a unique piece of music. Every piece of music is created note by note from scratch; there’s no pre-created material, no licenced music, no loops. It really is composed out of thin air, then performed, produced and recorded. So if all you know is the style of music you want to create, the mood you want to convey and the length of your piece of music, you can create a piece of music in a matter of seconds. If you want to talk instrumentation, timing, spotting, harmony, rhythm, you can give as much input as you want. And so the hurdle to create is very low. If you know more about music, the ability to utilise it more powerfully is massive.
Tell me about collaborating with a digital version of yourself…
Amper has the ability to learn how to make music like anyone. We’re very careful about not doing this regularly, because of legal issues, but have worked directly with musicians who wanted us to create music that sounds like them, in order to create a digital version of their own creative artistry. You basically have you as Amper make music that then you get to collaborate with. I would imagine in the future it will become more the norm until everyone has their own AI version of themselves, and can make music with themselves as a computer.
None of Amper’s music is extracted from a bank of songs?
No, the songs don’t exist ahead of time. We build data sets that define everything about music, emotion and genre. So you’ve got music theory, orchestration, instrumentation, performance, the foundational objective things of music … and then we define genre and emotion. Ultimately, the technology derives from both of these data sets how to compose a piece of music that will make someone feel a certain way. So if you say ‘Amper, we need you to create a piece of orchestral hip hop that feels determined’, it will figure out how to write the sheet music that will feel determined, in the genre of orchestral hip hop.*
And then it’s ‘performed’…
Yes, we’ve written the ‘sheet music’; the notes are written down in code. The next step is the performance; turning it into audio. To do that we’ve got one of the world’s largest audio sample libraries. We record every note of every instrument thousands and thousands of times, every possible way you would ever play it. So you take a violin, every note, every dynamic, articulation, velocity, anything you could ever imagine. We capture every way this instrument could ever be played….
That was a big investment…
A huge investment, millions of dollars over five years. It’s now one of the world’s largest libraries and it is the only sample library created as a set of data first, as opposed to a library meant to be utilised by humans first. Because of that, there are massive advantages to being able to understand and manipulate the data but ultimately we’ve got this composed piece of sheet music, which Amper then looks at and ‘performs’. It puts thousands of snippets of audio files in a row and stitches them together so it sounds like one audio performance that came out of a studio. These are some of the world’s finest musicians recording their instruments in the world’s best studios.
Chord progressions are one thing, is creating a good melody the most difficult part?
The melody, chord progressions, rhythm and performance each have their own challenges. What’s wonderfully exciting, and also frustrating, is how to combine all of those elements into a piece of music. Ultimately, music – like everything else in the creative world – is subjective. There’s no right answer. You and I can listen to the same piece of music, have different opinions and both be correct. We can change our opinions, and we’re still correct. Obviously, we believe from an objective perspective that the music we create is quite good but then it’s critical that what Amper creates can be collaborative and editable. No-one will ever get it right 100% of the time; we give you the ability to make changes.
There are copyright issues for AI platforms building their music bank from Spotify, for example, even if they’re breaking it down to a granular level. If an AI song is made up of one hundred pieces of music, and if my song on Spotify is used, I want 100th of the split, if it’s a hit, yeah?
Totally. It’s a great question. There’s a philosophical argument and then there’s the reality; my philosophical argument is much more liberal than the business reality. With human composition, if you use one song as inspiration to create a piece of music, and it sounds exactly like that song, no-one’s going to argue that it’s a not copyright infringement. If you use a hundred songs as inspiration, and you use one-hundredth of each song, who knows? Maybe it’s infringement, maybe it’s not. If you use a million songs for inspiration, and one-millionth of each song… at some point you would say it’s just being creatively informed by the work. And so there is somewhat subjective delineation in terms of what the sample size of your inspiration would need to be before it’s considered just the creative zeitgeist. That’s never been defined and Amper certainly doesn’t want to be the one to go through a lawsuit to create that.
Tell me about your legal journey to protect against potential copyright infringement.
We build our data sets and our sample library because we want to be 100% clear of any potential copyright issues even to the extent that we can indemnify our users. Speaking to US copyright law, the two things that go into copyright infringement are access and intent. Did you have access to the material you’re accused of infringing? And / or did you intend to infringe on it? The only sure-fire way to ensure that there is no infringement is to have zero access and zero intent. Anything else is a grey area. That’s why we don’t train it on the world, we build our own internal data sets, so there’s no access (to already-recorded music). When you use Amper, you’re describing music with genres, moods and emotions, and not with artists and song names – so the user has no intent. So you can’t say ‘I’d like Beyoncé, Single Ladies’. Because of that, there’s no access and no intent. And so every piece of music created is not only unique but is free of any potential copyright issues.
What’s the business model?
We have an API (Application Programming Interface) that integrates into content creation tools and content distribution platforms. It allows you to control anything about the music. With our web app, designed for those less-musical individuals, you pay for monthly or annual access to create as much music as you like. It’s typically used by video editors, game developers and podcast creators because they want to use the music wholesale.
Who owns the copyright?
You made creative decisions in the creation of a novel piece of work, so you’ve created a copyright. By the terms of your agreement with Amper, you assign the copyright and all the rights back to Amper, and we provide you a royalty-free global perpetual license to use that music however you like. So, if you create a piece for an advert or a video using Amper’s music front-to-back, that music is owned by Amper.
What’s the deal for musicians?
When musicians use it as an ingredient in a newly created work – and to be clear, while they sometimes use it, the platform is not necessarily designed for musicians – we relinquish rights. If you use Amper as you use sample libraries, as an ingredient in your new creative work but not as the core of the whole thing, then we will grant you a license and not claim any ownership over the resulting music.
Have you done deals with labels?
We’ve done deals with artists. We’ve had a lot of good conversations with labels, and I think at some point it will just be commonplace.
Has there been push-back from the music industry?
Not anymore. Five years ago there certainly was. AI music is not a new idea. You can say that computer music has been around since the 1950s, since computers. Algorithm music has been around since at least the days of Bach, the 1700s. But Amper is the first company that’s been able to use that on a commercially successful level.
Bowie used the cut-up technique and subsequently developed a digital lyric randomiser for "igniting anything" that might be in his imagination. What are your thoughts on creating lyrics with AI?
We don’t do lyrics right now. Music is a hard challenge by itself! They’re getting better and there’s a parallel development. In some ways it’s a similar problem; it’s translating creative ideas into reality. In other ways, there are very different problems; with lyrics, you’ve got the actual written words, which can have all sorts of subjective meanings – and then you’ve got the performance of those lyrics. In music, you’ve got the writing and the performance of the notes. So the path is similar, but the details are different enough that we want to stay focused on our core creative abilities now, and we’ll save lyrics for a later day.
If AI could discern the formula for writing a hit, surely everything should be a hit? And If everything is a hit, then nothing is…
No, just because every person can create a piece of music, it doesn’t mean that every piece of music will be a hit. What makes music a hit is in itself a very separate thing. There are people whose careers are built around trying to recognise it; some people say you can predict a hit because of the things that people respond to, others say it’s almost impossible,
Should we leave songwriting to the talented songwriters? Will the democratisation of music creation muddy the water in terms of what we listen to, and are we adding clutter to an already-crowded environment?
I would say very much no. If we think about writing… just because anyone today can create a blog or can self-publish a book, it doesn’t mean that we don’t value great authors any less. In fact, we probably value them more, because we recognise how much better they are than everything else. They’re using a lot of similar tools but they use them to further their artistry, rather than just to enable their baseline expression, and the parallel to music is very similar.
Isn’t there a valid aesthetic position that the imperfections in a performance are what make a piece of music? You’ve created original sounds, but can you feel the push and pull of a drummer playing in the moment?
Totally, yeah. One of the most wonderful things about humanity is that we make art. It’s inevitable that AI music will be indistinguishable from human-created music. Because even the imperfections, the push and the pull, will be solved for; whether it’s today or in a thousand years, provided that humanity is around. It’s a matter of when, not if. But the differentiation is on the value that we place on the music; are we valuing it for the artistry of the creation, of the performance, of watching a person make a thing? Or are we valuing the music because we have a video that we need music behind? Take coffee as an example. I live in New York, where you can get coffee in McDonald’s for a dollar; you get caffeine. Or you can go to an artisan coffee shop in Brooklyn and spend $15 for an artistically created drink from Rainforest Alliance growers, that’s conflict-free, organic and non-GMO. I’m still getting caffeine, right? But what I value about it is very different. I value the process that was used to make it: the collaboration and creativity. Even though the end result is that both give me the same amount of caffeine, the value I place on them is very different.
Will robots become sentient?
I think the answer is probably yes.
Between now and 10,000 years from now. I mean that seriously. I don’t know when but I think provided that humanity still exists, it’s an inevitability. Much like almost any technological evolution probably is, provided it doesn’t break the laws of physics. Given enough time. I wouldn’t be the one to say it’ll be next month or next decade or next century but given enough time I think it’ll probably happen.
So a robot will be able to feel the music?
Yes. That might be 300 years from now but who knows?
Will AI ever be able to write a line like ‘I can almost smell your T.B. sheets, on your sick bed’, as Van Morrison did?
And deliver the line with the same emotion?
Probably, yes. Which, again, no time soon, to be clear…
He broke down in tears after recording it, such was the feeling that went into the session…
Yeah, totally. One of the tropes used to describe AI or leading edge technology is that it seems like magic… until it’s not. A hundred years ago, people could have been talking about this electricity thing, ‘will it be able to do this manual job or that manual job?’. In that time period, it was almost inconceivable – now it’s just commonplace. So from a philosophical perspective, if pushed to give a binary answer on ‘can a thing happen?’ – again, provided it doesn’t break fundamental rules of physics and existence – probably ‘yes’. But even when that day comes, there will always be a differentiation and distinction with something made by a robot – or even something made by a person, that’s valued functionally, versus artistically. We can listen to Van Morrison on Spotify and it’s cool. But there’s something different if you see him play live. There’s an emotional connection with his creativity and that is something that’s just innately human. So even when robots can do things themselves, it doesn’t mean the value we place on it will be the same.
Do you see a time when people will stay tuned to an AI-only radio station?
Sure. Amper’s music has already been proven to be indistinguishable from human-created music. That is to say, almost all of the time, if you didn’t know you were listening to AI music, you wouldn’t know. Even a musician might not know. Or they might say, ‘it’s not an Academy Award-winning song yet’ but almost 100% of the time people say ‘oh, it sounds like a person has made the music.’
Where will AI music figure in the Top 40 within the next five years?
I would say 80% of the Top 40 will use AI music technology in its creation – as a tool in the creative process – within five years. I think very little of Top 40 will be made solely and 100% by AI.
People fall in love with musicians. Can they fall in love with AI-created music and is that something you’re trying to encourage?
Yes, with an asterisk. Empowering musicians to help make music is what we love to do. People fall in love with Bono, right? And there’s a relationship. In Japan and China, there are artists – digital avatars – with huge fan bases. They’re completely ones and zeros! It’s not necessarily for us to decide where people invest emotionally. We just want to make sure that whoever wants to be that creator has the ability to do so.
Should Facebook and Google be defined as publishers and be made to take responsibility for what is published on their platforms?
Yeah, I think so. The DMCA cover they’ve had for a long time is very enabling to creators but also provides a lot of opportunity for unfair exploitation. One might argue that that was necessary to create value and the opportunity to make these things work in the nascent stages of social media and online sharing. But in the mature stage that we’re at now, it’s critical that those who enable the exploitation of IPs – Facebook, Google and others – take responsibility and be held responsible for what they do. Because they’re very powerful concepts… technology platforms that enable the world to connect globally. And oftentimes the content creator gets the short end of the stick. There’s an argument of fairness and equality. I’m not the one to decide where that line is but I would certainly say that right now much more of the value goes to the platforms, publishers and distributors than to the creators. And the pendulum can definitely move back towards those who really create the value.
There is a growing realisation that the Internet has been used to colonise individuals, creating a new kind of digital serfdom...
AI music is a great democratising force. Typically, lordship and serfdom come about when there are inequalities in access and ability. When technology advances, it ultimately lowers the bar from the time, cost and resource perspective of what it takes to be able to do a ‘thing’. Then more people can do that thing, it becomes more democratised, resulting in less disparity between the haves and have nots, which ultimately diminishes the gap between those who previously were the lords and the serfs. So technology and music AI are great democratisers and enablers of expression. How we utilise that is up to any individual. There will continue to be artists we revere and those we don’t. They’ll all have access to the same technology, so it comes down to what do you do with it?
How can the malign influence of social media be halted?
One of the biggest societal changes with the advent of social media and global connectivity is that our interactions have become less personal. When that happens many people are emboldened. It gets harder to vet material. Ideas spread more quickly and easily. And when the barriers to malevolent behaviour are diminished, and when the cost and commitment of sharing something is as simple as a click – rather than reading, discussing and debating – it creates an environment where the safeguards that we’ve historically had as a human race are largely eroded. The teams in all the large social companies are working on figuring out, in essence, how to put safeguards back in, without unfairly limiting and undermining the core value proposition of their platforms in the first place. And that’s an imperfect part that certainly is very important both to the consumers and creators of a platform. I think everyone agrees it hasn’t been perfectly solved yet.
You said in your TEDx Talk that AI will be the greatest creative revolution in music history. Do you have concerns as to how experienced musicians will embrace it?
Many may see it as scary thing at first, and again largely it’s because there’s fear around its creative opaqueness in terms of the consequences. But I hope everyone sees it as a massively enabling tool. AI is the next generation of technology in music-making. It should allow less-musical individuals to all of a sudden express themselves through music. And it should allow more-musical individuals to harness a new superpower. Think about how much creative expression benefited from the digital revolution over tape. And how before that, creative expression benefited from tape over non-recorded music. Every step of the way, things – the world – changed.
What advice do you have for Irish songwriters?
I would say to Irish musicians and songwriters that how we make music will continue to evolve, it always has. Our career – or our ability to help others accomplish their goals – might never change. In the same light, new tools should enhance rather than undermine our ability to help others achieve their goals. So if the music we create today is valued as functional music, that value is being democratised and so you can move up the value chain with your music. If the music you create is valued as artistic music, your career can be as prosperous as ever.
How do Irish musicians get their music into movies in the US?
To get your music into production, typically you either compose the score, which is the path that I took, or get placements for your songs. Each has a very different route. Composers build relationships with directors and producers in studios to become a part of the core creative team. Artists seeking placements need to build relationships with music supervisors – and sometimes producers as well – to hopefully be top of mind for new projects.
How do you crack the music supervisor nut?
Half the game is persistence, a quarter is talent and the other quarter is being a good person and working hard. It’s true now more than ever that people want to work with people they can get along with. And so if you put your head down, you make good music and you’re a good person, I’m confident that most people can crack it.
What are your thoughts on Irish music?
I love Irish music! As a film composer, it’s obviously a huge inspiration for a whole genre of projects. I started out writing music for movies, TV shows and video games in Los Angeles. U2 is one of the bands I grew up listening to; the Edge’s guitar is part of the foundational sound I grew up with. It certainly was an influence.
To finish, will AI help to create a song better than The Beatles or Van? Songs better than we’ve heard to date?
Better is subjective. It will help make songs more efficiently, it will help expand the creative boundaries. It will help us to experiment more easily and cheaply. It will help flesh out new ideas. Whether they’re better songs? My grandparents would say the best music was made in the fifties, someone else might say the best music is made today. It’s such a subjective thing. But we can say it will help further enable the expression of creativity through music. For certain. What we do with that capability is up to us.
Drew Silverstein appeared at TEDx Roma. Go to ampermusic.com to compose using artificial intelligence. Go to hotpress.com/aimusic for more.
*(A)I COMPOSED ORCHESTRAL HIP HOP
Intrigued by the idea of an AI-created piece of orchestral hip hop as described by Drew Silverstein, Mark decided to put Amper to the test. He tasked Amper to write and perform a piece of music that, in Drew’s words, “will feel determined, in the genre of orchestral hip hop”. Listen to the result at hotpress.com/aimusic