INTELLIGENCE Tue, 12/21 2:10PM • 21:29 SUMMARY KEYWORDS ai, music, sound, song, human, create, musicians, max, song contest, computer, artists, people, artificial intelligence, lyrics, produced, brody, game, music producer, creativity, masih SPEAKERS Afua Mfodwo, Dr. Seguin, Max Savage, Clip, Rachael Masih Afua Mfodwo 00:05 In this episode producer Rachael Masih unpacks how music created with artificial intelligence will become more popular in the future. My name is frm photo and this is unraveled. Rachael Masih 00:32 The song you're hearing was not written by a person. This song, as human as it may sound was composed by a computer and her name is Ava. audio engineers and artificial intelligence experts are releasing tools like compu Tozer and Muse net that help musicians, filmmakers, podcasters, and basically anyone who needs music, create music through machine learning. Grimes, a Canadian singer songwriter and music producer said in an interview with Times magazine that she believes we're coming to the end of human art that machines are going to do it better. But contestants in the AI Song Contest are proving that notion wrong. They're using AI to help push the boundaries of their own human creativity. Clip 01:40 During favorite, listen to your body acquire from Team EVO Jeyes. John, get it Max Savage, Maxim's Brady Jenkins. Rachael Masih 01:59 Hi, I'm Rachael Masih. And on this episode, we're taking a look at how the use of artificial intelligence is going to open up a new genres and avenues for creativity in the music industry. This is Max Savage, a music producer and filmmaker based out of San Francisco. He produced the song listen to your body choir for the winning team, emoji ized. Max Savage 02:32 Originally, I had gone to school with John Gillick at a school called pure mind, San Francisco. And we had done a couple of tracks together that he had written and I have produced a mixed mix engineer and he just hit me up randomly was like, Hey, I'm gonna mentor this AI Song Contest. The last one was a Eurovision contest. Are you interested? And my buddy Matt Sims is also going to join us and I was like, hell yeah, let's do it. This is weird. Let's try. When we answered I really didn't think we were gonna win it all. Rachael Masih 03:04 Emoji arised and 37. Other teams including Roboto lab and sound obsessed, submitted songs for the second ever AI Song Contest. And even though this contest and the use of AI and music seems extremely New Age, the concept of randomly generated music is old, like Mozart old. In the 1700s, Johann Philippe Kern Berger, invented a popular new game, musical Licious verbal spill, which translates to musical dice game. This game, kind of like aI promised that anyone could be a composer. Here's how the game went, you would have a pair of dice, what is essentially a bingo card full of numbers, and a key. Every number on your bingo card matches up with a bar of music on your key. All you had to do was roll the dice. Find the number you rolled at the top of your bingo card and choose a number from the column below it. Then, based on that number, you'd find the bar of music it represents in the key and add it to your composition. You do that enough times and boom, you would have a completely original piece of music. These musical dice games caught on big time. The next version, ludicrous. malapa Atticus was similar to the first game, you'd roll a number, follow the rules on your guide and build a composition one note at a time. This game had a more complicated set of rules. So The outcome was much more random. And pretty similar to the way AI works today. Dr. Seguin 05:06 There is a kind of a rolling of the dice that happens during generation, because it's just choosing randomly between like, which note to turn on, etc. And it just keeps keeps rolling. It's except it's like a, you can think of it as a 400 sided loaded dye that it's rolling over and over and over. Rachael Masih 05:27 That's Dr. Seguin, or a researcher at the vector Institute for artificial intelligence. He's talking in a recent panel for the University of Toronto, called AI as foil Siggy have trained his AI on hundreds of classical piano performances by Chopin and similar artists. He said the process is like a loaded dye, because each instruction is coded with a number. And depending on what number you just rolled, different instructions become more likely. For example, if the A I just played a C, then notes in the C major chord, E and G become more likely to come next. After the machine was fully trained, Steve told it to generate something statistically similar to Chopin. Here's what he got back. When he told the computer to do it again, he got this. These generated songs sound thoughtfully composed for a few seconds. But the longer they go on, the more disembodied they sound. This happens for a couple of reasons. One is that there is no sense of direction. Humans write music that builds and wanes and builds and wanes. Computers don't. It's random. Another reason that the music sounds a bit inhuman is that well, sometimes they're just too damn hard to play. Dr. Seguin 07:56 You can do one jump, but then to do a jump, and then a few seconds later, and other jump, and then a few seconds later, another jump, you could feel as a player, I could feel the disembodied pneus of it, because no person would have done that. Rachael Masih 08:10 But for artists in the fields today, it may not matter if it sounds human or not. The beauty may lie in the organic quality of the sound that an AI can produce. Paired with the inhuman sounds that it thinks to create participant of the AI Song Contest, Max savage said that making his AI song ended up not being about humanity at all. Max Savage 08:43 I actually found out that the AI would just create the most organic creating weird sounds weirder than the human sounds, so actually had to like tone and kind of calm AI down our direction for a change where originally I thought that we were going to let the AI kind of take whatever we had, and turn it into something new. And the way we ended up doing it was that we treated the AI artists. And so we I don't do it. I was like, Okay, what did they I want to say, what are the lyrics you want to say, Rachael Masih 09:21 to take it back to the beginning of the songwriting process. We're taking it back all the way to 1961. Max Savage 09:32 To start with the idea that we base it on something kind of historical because like what's the root can be like, what's the what's the nutrients, we're going to eat it? And so we found the moon first. Recording with a computer voice, essentially was the song Computer like, it was an IBM computer in like 1961 or something like that. So that is the route. And so initially we covered it and I had Brody who I was the vocalist on it, human vocalist. I was like, oh, let's have you sing it. And then when we realized that we weren't allowed to use anything that already existed. I was like, Oh, no. Rachael Masih 10:37 Since their team wasn't allowed to use pre existing songs, they thought the lyrics of Daisy Bell into an AI and asked it to tell them what it would write if it were up to the AI to continue writing verses to Daisy Bell, Max and the other members sifted through the output and chose the lyrics. Listen to your body choir, a lyric that makes just enough sense to work but not enough sense to be written by a person. They did the same thing with the melody and the rhythm. And they ended up with what is essentially a spacey robotic, part two of Daisy Bell, written by a computer. Their vocalist, Brody Jenkins, recorded her voice in the bathtub. They ran it into the AI. And here's Max's interpretation of what that sounded like. Max Savage 11:38 For example, he kept on making this weird challenge cover. And so I use that as my root note and built a piano that would play using that as its source. And so then I could play essentially could play the AI, the AI version of singing as Brody. Rachael Masih 11:58 After that, they let the AI interpret itself over and over again, until they felt done. Then they had to make it sound Poppy and palatable. Max Savage 12:11 And what I noticed was when the AI would kind of create these human sounds, or emulate them, it would sound like it would sound like a distant ghost trying to be human. And so I think that some people, early versions of that kind of sack, and then I went to that, especially for a Song Contest, which is, you know, a popularity contest to be honest. And I was like, it needs to be more pop. And since they had done Eurovision last time, I was like, There's no way to make this more popular. Just put like a kick snare on Rachael Masih 12:50 the process of creation is not going to be stifled by AI, the AI Song Contest, and all its competing musicians prove that humans have the capacity to dig deeper, and use AI as inspiration to create something that can't be artificially replicated. Max even said that if AI were to start pumping out top hits, no one would care about them. If every song on the radio sounds the same, there's nothing special about it. The music industry will always need humans to make even more unique music than before. Max Savage 13:33 I think I've learned a lot about using chaos, to create art and to create music, because so much of it was a chaotic profit, just going okay, here comes a bunch of random stuff, you know, it will give you the weirdest ideas like lyrics, like cars, doing push ups, was one of our lyrics on a song. We'll love that was, hey, I tried to describe my writers. And it's just such a strange thing that like no one could ever none of us could ever. Rachael Masih 14:06 And even though cars doing push ups is a silly example, it's still a good example of the kind of prompt that might lead someone to create a song with the essence of cars doing push ups. And that song might be original and well loved, who knows. The world of AI and music is opening up new possibilities, new genres, and new ideas for the future. The band yacht used AI in their album change tripping. They fed an artificial intelligence their entire back catalogue of music to produce an album that reflected on their careers as artists. The band shadow planet used jukebox An AI that autocompletes any audio you give it as a third collaborator on their album. This made their songs a strange mix of folk and electronica, Holly Herndon, musician with a PhD from Stanford, made an AI musical collaborator that she calls spa, an AI trained on her own voice that can create music that reflects on the AI's huge capacity to learn. And its limitations. None of these artists have seen AI is anything but a means to push music into new creative avenues, and genres. Max hopes to see AI add to the creativity of humans by being kind of an assistant. Max Savage 15:59 You know, after after afterwards, I have ideas as to how I would love to use AI in the future. But the tools aren't there yet. And I think AI could be very, very helpful for a lot of musicians, especially for musicians that are not technically minded. Unfortunately, right now, the musicians that are able to use it, other ones that are very technically minded. And that's kind of cool. But that's sort of, I don't know, that's very, I actually feel like both technical people could already dive in technical problems with music and stuff like that. But the people that I know, that are not technically minded, I feel like would benefit most from having some sort of an AI or some sort of an assistant, I can help them translate their ideas. Like for example, when an artist just sends an iPhone recording, just going, I want the base to go, boom, boom, boom, boom, boom, boom, boom, boom, boom, you know, it takes technical skills, and then translate that into an actual bass doing that. And some musicians are so part of the real world that they don't want to be around a computer, they don't want to dive into that, like, it just sucks the life of all their creativity. And I think if you have a system in place, that's really wonderful, and would allow a lot more people to create new types of music we haven't heard yet. Rachael Masih 17:18 Like Max touched on, AI may make music more accessible to the average person, so that you don't need to be a mix engineer, or know how to sample music to make a song. You wouldn't even have to know how to play an instrument necessarily. This would open up doors for lots of new creatives with new ideas, who can use AI as a tool to fill in the gaps in their own knowledge. AI has the potential to bring new people into the game. People who never thought they would be making music. And those people could achieve things that they never thought possible. Kind of like Max and the rest of Team mo just Max Savage 18:07 I'm sitting on the Zoom, you know, and I didn't know it was this big of an event. And so then they're like, okay, okay, you know, this is all you're getting, like, sent around to the different rooms. And people have been nice backdrops are all dressed up. For me. It's like 7am, but for them, it's like, I don't even know what time it was, like 3pm Everybody looks great, has professional cameras and everything. I'm just sitting there, you know, without a shower and like sweatpants. And I was like, oh god, oh, what am I doing? Oh, God. They started announcing the different voting. And they do like the audience votes. And we had a decent amount, but we definitely weren't winning. And then the judges voted. And they all gave us a perfect 10. And I was just like what she told me like, What, again, the winner is Hey, and then they suddenly I'm on screen, and I'm sitting in my sweatpants. And they're like, Get up and dance. Because you want Oh God. So I had to get up and show the world that I was in sweatpants. But it was it was it was really cool. And validating and all those things. And I don't know, maybe maybe the cool part of it is that, you know, while these teams of 20 people from larger corporations and stuff made awesome content and do really cool songs. Maybe the cool thing is that the new this new world of online content and AI and all that stuff allows for people that aren't necessarily organized aren't necessarily super established to create something awesome. Also, it has to resonate in until maybe that's starting already. Rachael Masih 19:53 So there you have it. The world of AI is new and strange and ever growing The possibilities at this point are endless. People can take AI wherever they want to. We are on the verge of tons of new genres and new tools that will help make anyone in artists and I for one can't wait to see what comes out of it. Afua Mfodwo 20:49 This episode of Unravelled was produced by Rachael Masih. Our associate producer is Talha Hashmani and our executive producer is Elena De Luigi. Special thanks to John Powers for composing our theme music, and Ben Shelley for creating our podcast artwork. Our professor is Amanda Cupido and special thanks to Lindsay Hanna and Angela Glover. - 1 - 00Transcribed by https://otter.ai