In June 2017, Music Ally interviewed Stephen Phillips of Australian startup Popgun about his company’s development of an artificial intelligence (AI) called ‘Alice’ that was capable of playing piano with humans.
Alice would ‘listen’ to the notes played by a human, then respond with what she thought might come next. Even at that early stage of the project’s development, it was impressive.
Nearly a year and a half on, Popgun is ready to show off how its technology has developed, and talk about what it might mean for the startup and its plans for commercial AI-music products.
The best place to start is this video, shared with Music Ally by Popgun, which shows the evolution of its technology from January 2017 to July 2018:
The video starts with Popgun’s AI showing its prediction skills: an engineer plays short piano sequences, and the AI responds with what it thinks they’ll play next.
By August 2017, Alice was capable of improvisation: listening to a sequence played by a human and then modifying it while preserving its musicality. By the end of the year, the AI was creating original piano compositions without any human input, in a variety of styles.
Early in 2018, Popgun began developing an AI to play drums, then bass, and then to have the two play together, as well as with the piano. And then between March and July this year, the work began on supporting human vocals, with an AI capable of composing an original backing track when given a human vocal loop.
Along the way, Popgun raised seed funding in February 2018 from Silicon Valley VC Khosla Ventures, as well as Techstars Ventures – it was part of the first Techstars music/tech accelerator in 2017 – while the company has also added former Pandora CEO Tim Westergren to its board in recent months, and grown its team to 16 people.
On the technology, Phillips now says that “the first year was pure piano: and at one point we were thinking ‘holy hell, we can’t do this’ before we had a series of breakthroughs. The piano got very good, very quickly towards the end of 2017.”
Even when Music Ally first interviewed Phillips, Popgun had already switched strategy. At the outset of Techstars Music its hope was creating an AI that could learn from popular tracks on Spotify, and then compose its own hits. By the end of the accelerator program, the emphasis was more on developing Alice, potentially as an educational tool for new musicians, as well as a creative foil for professionals.
And now? “We are back on ‘we’re gonna make some hit pop songs over the next couple of years’ now,” says Phillips. “It’s going to be a new instrument that producers will use. A tool: a more abstracted drum machine. It can play the guitar, the bass and the piano, and each one of those AIs can listen to one another and play together. That’s very new.”
Phillips is very clear about what Popgun’s AIs can and can’t do, however.
“It doesn’t have taste. It has no conception of culture. It sees a whole lot of music, and from that it can model the distribution about how to play that instrument. It has lots of concept about key and timing and phrasing, but no concept of ‘good’ and ‘bad’,” is how he puts it. “Humans are integral to that. That bit’s not in the AI: it has to be done by humans.”
At this point, Popgun has an AI (or a group of AIs) that can play piano, bass and drums together, and which if provided with a human vocal line, can compose music to back it – while also mixing and mastering to a releasable quality. By the end of the demo video above – the footage from June 2018 – the only thing not being done by an AI is the vocal.
“We’re now thinking about how we can bring that technology to market so that people can start to enjoy it,” says Phillips, while happily admitting that there is plenty more improvement to be made to the technology itself.
It struggles with jazz, for example, as well as certain kinds of dance beats, which is why Popgun is training its sights on the pop genre, as well as a sprinkling of trap, hip-hop, R&B and country-pop. Philips is confident that as the AI continues to improve, it will prove its mettle even to sceptics.
“We will have pop hits!” he says. “Well, either us or someone like us will build software that every producer on the planet is using. When we show this to producers, they’re like ‘I need this!’. Press a button and something creative comes out. It moves them from being the sole source of creativity to being the curator of this experience. ‘No, no, no’ a thousand times until something resonates, then they’ll take that and polish it… I can’t see any serious musician or producer who won’t adore that.”
This illustrates one of the important strategic decisions that Popgun – and any startup focusing on this kind of technology – for example Jukedeck, Amper Music and AIVA – will have to take.
It’s this: once you have an AI capable of composing, playing and mixing/mastering music, do you commercialise it as a tool for professional musicians, producers and songwriters? Or do you go in another direction and try to get it into the hands of non-musicians, and try to democratise music-making in (although the parallel isn’t exact) the way that Instagram democratised photography?
Phillips is wary, at this point, of telegraphing Popgun’s intentions. He prefers to talk about the opportunities of both these paths.
“The fact that these things are creative engines is what’s exciting. Painting’s easier, music’s coming,. And soon AI is going to be writing stories and doing films,” he says.
“And that’s not just ‘here’s an Oscar-winning film’. It will be a tool for scriptwriters to start generating content. This idea that given enough content, it can model all of the possible distributions of that, and then there are a thousand gold nuggets of ideas that you can harvest.”
“Creativity has always been regarded as such a human function, and maybe this threatens that. But it doesn’t take creativity away from us: it just augments it. It will let people who can’t tell stories, tell stories. That joy of making something other people like is going to be available to many more people. We think that’s cool stuff.”
People do feel threatened by this, including musicians and labels. Yet part of the ethos of Techstars Music was to put its startups in front of music rightsholders regularly: and to see them as mentors, partners and potential investors. According to Phillips, there has been a marked lack of defensiveness from those rightsholders when shown Popgun’s technology.
“I have demoed to all of the major labels, and the response has always been super-engaged. They want to understand it. The first question is always the same: ‘Who owns the music?’ That’s their business, but when my answer is ‘We don’t know! No one knows!’,” he says.
“There’s real excitement about what it’s going to do: across the board everyone’s engaged. A little bit scared around the control part of it, but super-excited.”
He also addresses the question of whether AI threatens the livelihood of artists. “Music is not just a technical skill. If the technical barrier to playing something that sounds like music is dropped, then the artist is all that’s left. And the artist will still be better,” says Phillips.
“They’re artists! They’re engaging, they’re charismatic, they’re talented. I’m sure if we gave this tech to David Bowie, he would do it better than anyone else on the planet. He’s Bowie! This is another instrument.”
In the near future, Popgun will release its first products that let more people get their hands on this instrument. “We really feel we’ve reached a point where we want to release product over the next year,” says Phillips. “I couldn’t be more excited about seeing it get out there.”