How artificial intelligence will do to music what Instagram did to photography
|MC_PREVIEW_TEXT|
View this email in your browser (|ARCHIVE|) http://hotpodnews.com/presents.... A Newsletter about Big Ideas in Music and Technology, by Cherie Hu This is issue #58, published on August 1, 2019 Happy August everyone!
This is the first email update focused on the Water & Music podcast (http://waterandmusic.transistor.fm) . In case this is new to you: like this newsletter, my podcast focuses on the fine print of big ideas in music and technology, through interviews with leading innovators, artists and thinkers from across the music business. It’s available on most audio platforms (https://waterandmusic.transistor.fm/subscribe) , including but not limited to Spotify, Apple Podcasts, Google Play, Stitcher, TuneIn, Overcast and Pocket Casts.
For the sake of widening accessibility to information, I’ll be publishing full, edited and hyperlinked transcripts for all future episodes (and, eventually, for all previous ones too). On alternating Thursdays, I’ll summarize the most interesting takeaways from those episodes here.
As I mentioned in the previous issue, that means this newsletter will now come out weekly rather than biweekly, every Thursday at 12pm ET — switching between a podcast episode synthesis and the more fleshed-out, exclusive essays you’re already familiar with. I’m also in the process of working with a developer on building a new website for Water & Music, where you’ll be able to access both newsletter articles and podcast episodes/transcripts in one central place. More news on that in the coming weeks. :)
Thanks for reading!! - Cherie How artificial intelligence will do to music what Instagram did to photography
For the tenth (!) episode of the Water & Music podcast, I had the pleasure of speaking with Alex Mitchell, co-founder and CEO of a startup called Boomy (http://boomy.com) .
You can listen to this episode on Apple Podcasts (https://podcasts.apple.com/us/podcast/episode-10-ft-alex-mitchell-how-artificial-intelligence/id1454221845?i=1000445574177) , Spotify (https://open.spotify.com/episode/5ec6EJBKaxUbqgXj5Dhjxp) , Overcast (https://overcast.fm/itunes1454221845/water-music) , Pocket Casts (https://pca.st/d1aN) , Stitcher (https://www.stitcher.com/podcast/water-music?refid=stpr) and most other major audio listening platforms (https://waterandmusic.transistor.fm/subscribe) .
Boomy just launched out of beta last week, and is building accessible tools to help users not only make and edit “instant music” on the fly using artificial intelligence, but also distribute and monetize that music directly on paid streaming services like Spotify and Apple Music.
Our conversation spans a ton of interesting terrain, both philosophical and practical, about how A.I. will transform music creation and the industry at large. Below is a summary of Alex’s most interesting takes; you may not agree with all of them, but such is the beauty and excitement of breaking new technological and creative ground in any field.
You can read the full transcript of this interview on Medium (https://medium.com/@cheriehu42/how-artificial-intelligence-will-do-to-music-what-instagram-did-to-photography-b69268a2d318?source=activity—post_recommended) . Alex also just did another great interview with Music Ally (https://musically.com/2019/07/31/boomy-talks-ai-music-we-want-to-make-music-thats-meaningful/) that is worth a read! 1. The term “A.I.-generated music” is misleading.
“You wouldn’t call what Slash does ‘guitar-generated music.’ And you wouldn’t call what Drake does, like, ‘autotune-generated music.’ I think this notion of ‘A.I.-generated’ — it puts kind of a weird emphasis on the tool.” 2. It makes more sense to optimize music composition algorithms for creativity, rather than for accuracy — even if the results don’t sound as nice.
“When we talk about artificial intelligence, you’re really talking about a set of data processing technologies … it’s an advanced way of analyzing a bunch of data very quickly, and allowing the algorithm to figure out what the best way to process that data is. This is technology that you see applied to photos pretty extensively. So, you know, here’s a picture of a person, or here’s a bunch of pictures of people, here’s a bunch of pictures of dogs; you, the algorithm, go figure out what’s a person and what’s a dog; and then now here’s this new photo, is it a person or is it a dog?
These systems have been designed for accuracy. In a system like that, you want to be able to distinguish accurately what is a dog or a person … one of the issues that you get with that is what I call the ‘over-optimization problem’ — where if you’re asking a trained algorithm that’s been trained on a bunch of MIDI data [to] write a song that’s ‘accurate,’ it’s trying to find what’s called a local minimum, so it’s going to write the same song every time, or very similar-sounding songs. If you’re giving it a reference and you’re saying, here’s what a ‘good’ song sounds like and here’s what a ‘bad’ song sounds like, you can get into this issue where there isn’t really enough variability and it isn’t super creative.
And so one of the things that we set out to do … [is] instead of trying to make an accurate song, try different creative ideas very rapidly … If you do it fast enough, what you can get is something that sounds totally stupid … like a bunch of garbled nonsense. But, that garbled nonsense might actually sound cool to someone. And then over time as we start refining the algorithms and working with the data that we’re getting back from our users, we’re able to refine this workflow into something where, now, we can start training individual creative profiles of our users.” 3. Apps designed by musicians often unintentionally alienate users without musical training.
“When we first launched the beta … we had four options you could pick: intensity, genre influence, tempo, and what you wanted in the song, so drums, bass, melody. And we were like, ‘This is so simple and intuitive. Everyone’s going to get it.’ People were so confused by just those four things. ‘What is intensity?’ was one thing that came up. ‘What’s EDM?’ was another question. ‘I know what hip-hop is, but, like, what’s trap?’
… [For] people who have music education, and who had been lucky enough to have music education, these are just concepts that we understand intuitively. But to the everyday user, it becomes very like, ‘I don’t know what that is. I’m not a music person, I can’t do that.’
… And so we sort of went back to the drawing board, and we said, how do we make this even simpler? And we launched what we call filters, because people get that. It’s like a photo filter. And the filter doesn’t have any musical thing in it at all. You can preview it, and you can kind of see what it sounds like, but we’re not enforcing any sort of genre on it. And we came up with silly names for them — one of them is ‘maximum thump,’ which is in place of high-intensity EDM.
All of our stats went up by four for the next group: four times as many songs, four times as many saves, four times as many people who kind of got through the process and understood it. And so that was a really important lesson for us … [for] the person walking around on the street who couldn’t care less about the things you and I care about in the music world, you’ve already lost them when you say the word ‘melody,’ or when you say the word ‘tempo.’ I think music people like us, we can be very esoteric with the way that we approach these things. And so it’s been a learning experience for us … you’ve just got to make this, super, super simple.” 4. In turn: to an extent, an anti-A.I. argument is an anti-accessibility argument.
“To be somebody who’s really ‘anti-A.I.,’ you sort of have to be someone who has the opinion that music is something that should be reserved for an elite class of person, for somebody who has gone through the training and taken the lessons and has the equipment and the time to create music. And anybody who doesn’t have that, or doesn’t have sort of this internal talent, doesn’t deserve to create music. I think that’s an argument you could make; I don’t think it’s one that’s going to age well.” 5. What’s the equivalent of “AlphaGo (https://deepmind.com/research/alphago/) ” for music — i.e. an algorithm that can measurably beat a human being at its own game? The answer is complicated, and personal.
”… The important thing to note with something like AlphaGo, or some of the great research that OpenAI (http://openai.com) has done around gaming, is that there’s a win. There’s a win condition. You won the game, or you didn’t win the game. The way they did that is, they made that algorithm play [Go] a bazillion times, some crazy number of times on a server, and let it kind of figure out every possible outcome.
For us, we have to stay really honest about: what is a ‘win condition’ in music? … I think the win condition for us is meaning. If we can help you make a song that’s meaningful to you, in some way or in any way, then we’ve won.
… You get into this [debate], like, what’s a ‘good’ song and what’s a ‘bad’ song? I think it’s really simple for us: What is the lower bound of effort that it takes for somebody to get from never having created a song in their life — or maybe [never] even thought about creating a song in their life — to having something that means something to them, and that they can go use?” 6. Songs can act as sonic emojis.
“So we had a whole bunch of gamers sign up for the beta — like sort of younger gamer kids — and they were making these EDM tracks. And we were looking at these song title names that they were giving their songs, they didn’t make any sense to us. So we had to Google it. And as it turns out, what they were doing was they were creating whole EDM tracks just to make fun of another player on Discord for losing at the game. And another player would send them this whole EDM song back being like, ‘haha, no, I beat you this time.’ That’s what we were able to discern from that whole thing and from talking to those users, is that they’re taking this and they’re using it to communicate positivity or negativity in a way that we never really expected.
We set out and we were like, you know, DJs are using [boomy] to create samples that they can play … If you told me six months ago that most of the usage would be gamer kids, like, making fun of each other, I would have not been able to guess that. And that’s not something that ever would’ve made sense before you could make a song in five seconds.” 7. The amount of music being created today might seem like a lot — but it’s only 0.04% of what’s happening in other industries like photography.
“40,000 tracks a day … I think people in our world — you know, the music world — we hear that, and we think it’s a big number. We’re, like, oh my God, 40,000 tracks.
Do you know how many photos go to Instagram every day? Off the top of your head? It’s 95 million. 95 million photos go to Instagram every day (https://www.reuters.com/article/us-facebook-instagram-users/instagrams-user-base-grows-to-more-than-500-million-idUSKCN0Z71LN?source=post_page---------------------------) . And there’s something like 350 million photos that go to Facebook every day (https://www.businessinsider.com/facebook-350-million-photos-each-day-2013-9?source=post_page---------------------------) .
You’ll hear me make references to photos a lot, because that’s really how I think about this market. I think for a long time, if you wanted to have a photo, you needed all this equipment — you needed all this time and expense, you needed the right lighting. It was this hard thing to do. And then instant photos came along, and all of a sudden it happened in two seconds.
We call what we do ‘instant music’ — we don’t even necessarily call it ‘music’ — and the metric that we track in terms of percentage of the world’s recorded music is there to inspire exactly the thought that you just had, which is getting ready for a world where there will be a million songs a day. There will be ten million songs a day. Of this I’m certain. It’s going to happen. The only question is on what timeline is that going to happen, what’s the role of automation, and what are things like Boomy in the market going to have on that. But we’re going to get there no matter what.
If you think of it as data, if you think of it as individual contributions of creativity, [40,000] is a tiny number. And I think, frankly, it’s so small … that in the real world, it just doesn’t register quite as much as other types of media.”
🌊 ✨ If you’d like to support even more thoughts and conversations on music and tech, I encourage you to become a paying member of the Water & Music ecosystem on Patreon (http://patreon.com/cheriehu?utm_campaign=Water%20%26%20Music&utm_medium=email&utm_source=Revue%20newsletter) . For as little as $1/month or as much as $40+/month, you can access a wide range of perks including: * A closed, members-only Discord server * Exclusive essays * Bonus material for already-published freelance articles * Monthly editorial meetings and video hangouts with me
…and much more! Thanks so much for reading! ✨
============================================================ Twitter (http://www.twitter.com/cheriehu42) Instagram (http://instagram.com/cheriehu42) LinkedIn (https://www.linkedin.com/in/cheriehu/) Website (http://www.cheriehu.com) Email (mailto:cherie@cheriehu.com) Medium (https://medium.com/@cheriehu42) Copyright © |CURRENT_YEAR| |LIST:COMPANY|, All rights reserved. |IFNOT:ARCHIVE_PAGE| |LIST:DESCRIPTION|
Our mailing address is: |LIST_ADDRESS| |END:IF|
Want to change how you receive these emails? You can update your preferences (|UPDATE_PROFILE|) or unsubscribe from this list (|UNSUB|) .
|IF:REWARDS| |REWARDS_TEXT| |END:IF|