AI & Music: How AI Is Changing Music Creation & Creativity Transcript, AI Summary & Key Points

00:00 Quick experiment. I'm going to play a short musical idea. Don't judge whether it's good or bad, just ask yourself this question. Where do you think it came from? Okay, was that something that was composed? In other words, I wrote that years ago. Was it something that improvised? I made it up just right there in the moment. while we were doing the recording, or was it something derivative where it was heavily influenced by other artists that I've essentially absorbed and copied or was it co-created where basically AI

00:45 proposed the chord sequence and I just adapted it. Well those categories sound distinct, but in practice the boundaries blur. And here's the thing, from listening alone, you probably can't tell. But here's a real question, does it really even matter? Either you like it or you don't, and music is so subjective with so many different tastes that it's hard to get a universal consensus on any particular song.

01:15 We've traditionally treated music as if originality was visible in the output, but musicians know that's rarely true. For instance, blues music, it's had a formula of 12 plus three equals blues. What do I mean by that? Well, it is a 12-bar progression with three chords played in a very similar sequence. So it's all just variations on themes, at least a lot of it is.

01:42 And we look at folk music. A lot of this is also variations as one artist will essentially do a cover of another artist's song, maybe add their own flavor to it as they do that, and then we've got jazz and jazz is all about improvisation right you just keep on hearing a theme and then taking that and adding to it, and then, we've had producers who essentially borrow textures and conventions from other areas and work those into their work as well.

02:14 So AI didn't invent this idea of recombination. What it changed is the speed, the scale, and the opacity. Before you start shouting at your screen that this is all just going to result in AI slop dribbling out of your speakers, I want you to consider this. When have we not had AI slop, or some form of slop, passing for music? Have you listened to some of the streams these days?

02:38 Or the radio in years past? Well, so that again is a matter of taste. And the second thing I want to consider is we've actually already had an AI artist appear on the billboard charts. Some of this stuff is really good. It's only gonna get better. It's not gonna get worse. The bottom line is this, the toothpaste is not going back into the tube. This technology exists and people will use it.

03:03 So pretending otherwise is just living in a state of denial. I think instead it's better to take a look at some of the possibilities of AI in music from the perspective of an AI nerd with a guitar collection and way too much money invested in stereo equipment. First off, we need to acknowledge the fact that music has always been shaped by technology, from the invention of the piano to the electric guitar to synthesizers, digital recording and playback equipment.

03:30 All of this stuff contributed. Even live performances use amplifiers and audio effects like reverb and distortion to change the nature of original sound. Today, in another leap, technology isn't just helping people make music, it can now create music with us. Let's take a look at how this has evolved over the decades. We're gonna start with the 50s and 60s with algorithmic music.

03:53 Now what does that mean? Well, computers started making music long before synthesizers or MP3s even existed. In the 50's, researchers used early mainframes to generate algorithmic compositions. They were creating music with mathematical rules. Computer scientists were essentially teaching machines to play melodies using punch cards. Yeah, imagine that.

04:16 I even tried my hand at this back when I was in high school in the late 70s. Yeah, I'm that old. These pieces though, sounded random and mechanical, but they introduced a key idea. That is that music can be represented as data. So we carry that idea forward. Then in the 70s and 80s, we got digital synthesis. We had synthesizers and things of that sort that could play sounds that sounded like real instruments.

04:46 And traditional instruments that we knew, as well as they could invent sounds from instruments that didn't even exist. Then around that same era, we got this thing called MIDI, the Musical Instrument Digital Interface. And that became really important. It gave us a universal language for notes, for timing, pitch, velocity, everything you need to control music events.

05:10 This was huge. MIDI lit computers, instruments, and software all talk together. It's the backbone of digital music production even today. Then in the 90s, we had digital audio workstations. This is where we really started bringing computers into this. Computers were now full creative partners for editing, sequencing effects and sampling. Then in 2000s, we introduced machine learning into the picture.

05:39 And in this case, auto tuning, beat matching and smart audio cleanup tools came onto the scene. Then in the 2020s, we got deep learning and more recently, generative AI. And these things really took off. Researchers realized you could train models on thousands of songs to generate new patterns, harmonies, or even beats. This is the foundation of modern AI music systems, which can compose and perform with more artistry than we've ever seen before from a computer.

06:11 Today's AI models can compose full songs, melody, harmony, baselines, even multi-instrument arrangements. You can ask AI for a cinematic score in the style of Hans Zimmer, or a lyrical expressive Chopin-style piano nocturne. And it will generate the material these guys never actually got around to writing. The best part, AI can iterate instantly, giving you unlimited ideas.

06:37 So if you don't like the results, just run it again until you do. And then language models can, not surprisingly, write lyrics. That's what the middle L in large language model stands for is language, of course. So and they can write this in any style. It could be rap, pop, metal, country, even multilingual blends. They're especially good at brainstorming phrases and exploring themes.

07:02 So if you're writing a song and you're stuck on the next line and what it should be, well, AI can give you 50 alternatives in two seconds and let you decide which is good and which is garbage. AI can also model singing voices. So imagine the idea here where you have actual backing vocals without a human singer. There's also a voice style transfer, letting one person singing sound like a completely different voice.

07:32 Now here, we're talking next level karaoke when you can pick not only your song, but also who you wanna sound like. AI tools can clean audio, separate vocals. They can also help auto-master tracks, suggest chord progressions, or even write string arrangements. Instead of replacing engineers, the best tools act like super smart assistants. We could even use an AI filter to take out the sound of a train that loves to ride by our studio every time I'm trying to record, based on a true story.

08:05 AI can even generate music in real time, reacting to your movements, your voice, or even your playing. Imagine a human and a machine improvising together. Could be great, could be awful, but when has that not been the case with improv sessions, with only humans manning the instruments? I'm reminded of some garage band sessions I was in back in the day, and I'll just say thank goodness for tolerant neighbors.

08:30 Models learn by training on large data sets of recorded music, and guess what? People do too, but more on that in a minute. AI can learn patterns, things like common chord progressions, rhythm structures, and instrument combinations. During generation, the model predicts, then, what should come next? The next note, the next beat, the sound sample, and it continues building those from those probabilities.

08:58 This is why AI is so good at style imitation, at blending genres and exploring variations. Are people really that different? Ask any successful musician, and they will tell you that they had major influences on their music as well. In fact, they'll give you a long eclectic list of influences, bands, solo artists, that come from a wide variety of styles and genres.

09:25 They learned from each other and created new music from unique combinations of all their influences and dare we say, patterns. Look, all the notes have already been invented. There's no more new notes to create. What we're doing now is just rearranging them in new ways and new patterns to create new results and new songs. Saying that a certain artist was a major influence on your music sounds a lot more acceptable than, I sampled all your data, generated a model, and now predict your output when I compose.

10:00 But the analogy is still there. The influences of what we've heard inevitably work their way into what we produce next. We learn from what we hear. It's a form of imitation, but with variation. The songs you hear influence the music you write. If I go to listen to the whole catalog of a particular artist and then compose a song in that style, no one accuses me of copyright violation.

10:26 It's how music has been composed since the very beginning. AI just does it on a different scale that's shocking to our human sensibilities. We need to develop a new frame of reference to reflect the current reality. Let's take a look at some different variations on how AI can work with a human musician as a partner. So we'll consider writing the music, revising it and performing it.

10:50 So we could have a human write, revise and perform. That's just traditional. That's how we've always done music and that's fine. It still works. We could have variation where we have the human write and revise the music But AI helps in the performance. So this is perfecting the performance and it's going to do it consistently the same way every single time.

11:13 Now we could look at another variation where we've got a human who's involved in writing. AI helps with revising and gives suggestions. The human decides what to keep and what to throw out and then we have the human involved in doing the actual live performance. So here we have an AI assist that's occurred. Another variation to consider, we have AI write the composition initially, but the human does the revisions on it and decides, okay, that was a nice first draft, but here's what I really want it to do.

11:44 Give me some more options. No, I'm gonna pick from those, that sort of thing. And then the human actually ends up performing the thing in the end. So it's AI inspired. It's like having a collaboration partner, brainstorming partner that gave you the initial idea, the spark, but then the humans carried it the rest of the way. And there's a lot of different variations, as you can see how this matrix would play out, all the way down to where AI is writing, revising, and performing.

12:09 This is the pure AI case. But my point with this is that there's a hybrid set of cases that we could consider. And there are a lot these. And you can decide where you want and how much involvement you want from AI in the music. As with any collaboration, you need to know the strengths and weaknesses of each member of the team. It's no different when we're talking about musical collaboration with AI.

12:35 So who's best at what here? We've got AI, we've got humans. Well, what is AI really good at? Well, it's good at doing things over and over and over again in infinity. It's got infinite patience, infinite variations. It can generate songs as long as we ask it to, compose new music. It can do variations on those and blend them with other things. It can do augmentation, so clean up music.

13:04 It can write transcriptions for us. It could even do a modulation where we put it into a different key, things like that. We can personalize the music more by telling it to do this again, but now do it in this style or things of that sort. So the bottom line is what AI is really good at here as a collaborative partner is scale. It can this with no fatigue.

13:27 It will just keep doing it over and over and over again and we can decide what we like and what we don't like. Now, the human over here, what are humans particularly good at when it comes to music? Well, I'm going to suggest to you it's connection. It's finding the expression of the music. It's the lived experience that falls out in the words and in the melodies.

13:50 It's what does the thing mean? Live performances also are a thing that people are particularly good that. And... There's just no substitute sometimes for watching somebody go up there and smash a guitar on stage, or you have a jam session where all kinds of things are happening and you've got a million moving parts and nobody really knows what's about to happen next.

14:10 Or you could take the same song and play it again with a very different feeling to give a very different meaning out of that same thing. So, where are humans particularly good? Well, there's a lot of things, but one of them I'll say is feeling. And if we don't have feeling, then the music is just a bunch of noise. I think the future of AI in music is less about AI replacing musicians and more about new collaboration models.

14:37 Sorry, dude, but howl with a six string is just no substitute for the energy of a packed concert arena or the experience of a more intimate live music session. You want a performance by someone with flesh and blood. No offense. You want to feel a connection to the music and to the artist. Just like the synthesizer didn't eliminate violinists, AI won't be the end of musicians either.

15:01 We're gonna see things like AI-driven instruments. We're going to see hyper-personalized playlists of music created just for you in real time. It'll be like having your favorite artists available to give you new music of the sort you like best to fit your mood at that particular moment and in unlimited supply. And it's gonna learn your tastes and get better over time.

15:23 We'll have things like real-time adaptive game music to enhance that experience. And tools that let anyone, regardless of skill, express musical ideas, which is great news for wannabes like me trying to up their game. Look, I don't play the guitar because I'm hoping to make it big one day. That ship sailed long ago. In the same way, I won't run every day because I still think I have a chance at the Olympics.

15:47 That's another ship that's far out to sea. But both of these activities give me a chance to express a part of myself and AI can actually help me do better with both of them. In the end of the day, music is subjective. We don't all like the same thing. AI didn't invent lousy music. Humans have been really good at making that for ages, but we still listen to human musicians, don't we?

16:10 So why not give AI the same grace? Give that the opportunity to have an honest hearing rather than just simply dismiss it as AI slop before you even heard it. As I said before, look, the toothpaste is not going back into the tube. This technology is here. We might as well get out in front of it and figure out how we can use it to make music better. Because music is more than just sound waves bouncing off your eardrums.

16:37 What really matters are the same questions we have with all art and with all music. The questions of how does it make you feel? Does it move you? Does it resonate with your life experiences? Does it inspire you in ways that you never even imagined? Good music does that. Music has always transformed with technology and AI is simply the next instrument in that toolkit. But like every instrument, the magic comes from the human playing it.