Music business and technology expert David Hughes takes us on a deep dive through the pros and pitfalls of AI

David Hughes knows a thing or two about paradigm shifts in the music industry. Formerly CTO at the Recording Industry Association of America (RIAA) and Vice President of Technology Strategies and Digital Policy for Sony Music, Hughes was responsible for helping the latter pivot from distributing their catalogue physically to digitally in the late ‘90s.

Over the last three decades, he has also been active in the promotion of Hi-Resolution and Immersive music formats, having also contributed to developing some particularly notable formats and standards, such as MPEG, DVD-Audio, Blu-ray, the International Recording Code (ISRC) and the International Standard Name Identifier (ISNI).

In this wide-ranging discussion, he discusses the transformative potential of AI, the disruption we are already seeing, and what we as an industry can do to prepare for what’s ahead.

What are the big issues the music industry needs to solve in 2023? 

 In 2023, I think there are four issues and the first three of them are: AI, AI and AI. The first one is AI music creation and how that affects the marketplace. The second one is AI doing work, such as performing tasks that formally require significant experience or creativity. If you assume that most jobs in the music industry require either creativity or area expertise, that means that pretty much everybody’s job is going to be impacted by AI. And then the last one – which is really the big one – is AI and copyright. The fourth issue is the economics of music streaming, but perhaps that is for another day!

 It’s like there are three legs holding up the table that is the music industry. The first leg is the one most people think of; that is the creative process of making music – everything from songwriting to recording, producing, mixing and mastering to marketing will be impacted by AI. But there’s always been two other equally important aspects to the music industry and one of them is technology. This is not new: the music industry has always been driven by technology since the very beginning. The sound recording business was invented by Thomas Edison and then others such as Victor and RCA who figured out better ways to record music. Then they had to sell players and something for the players to play. As they came up with new formats they had to figure out how to create better quality recordings and then stereo and all these things. That technology extends to the present day, with streaming technology, high resolution and immersive as well as the disruptive aspects of everything from peer-to-peer technology 25 years ago to AI today.

 The last table leg is copyright. Without copyright law, we simply don’t have a business in the music industry. And people often forget that if you can’t enforce your copyrights, then people can distribute your music without paying you. At that point it’s no longer a business, it becomes a hobby. And that’s fine if your passion is making music and giving it away. But as a commercial industry, we have to care about collecting money.

“People often forget that if you can’t enforce your copyrights, then people can distribute your music without paying you. At that point it’s no longer a business, it becomes a hobby”

AI must come with some benefits though. Song Sleuth, for one, uses AI to solve problems that were previously not possible by human intervention alone. 

 Yes, absolutely, AI can be used to solve problems. I was just guest lecturing at the university here in Washington D.C. and one of the students, quite wisely I suppose, asked, “okay, so if AI is going to impact most jobs in the music industry, what job should I do?” I said, “That’s not quite the right question. It’s going to impact all the jobs. What you want to do is figure out how you can leverage AI to do your job better. Don’t be left behind. Don’t be the one that’s trying to ignore that these new technologies exist.”

 We’re using these new technologies to solve problems that were otherwise… I don’t know if I want to say unsolvable, but it was simply not practical to attack some of these problems. For example, the thing that Song Sleuth is doing, trawling over massive amounts of content and making sense of it and identifying the unclaimed music on YouTube. You just can’t do certain things manually. Of course, Song Sleuth does the final step by hand, and hopefully as the AI gets better and better, it becomes more accurate and the manual process becomes less and less labour intensive on a per track basis, per dollar basis, however you want to calculate it.

 So yes, it’s not all bad, but it is a big deal. The other thing I will say is AI is going to be quite disruptive, not just for music but across most industries. We’ve seen disruptive technologies in the past. For example, in the 1990s, we had a perfect storm of technologies. We had the ability to compress digital files and make them smaller. We had the increasing availability and speed of digital networks to distribute them, and then we had portable devices to play the music on. All of these things happened, and it created a perfect storm for a shift from physical distribution on CDs to digital distribution. And it did create a lot of opportunities – eventually, but from 1999 until about 2012 or so, sales effectively went down every year for well over ten years, because we were disrupted, primarily by peer to peer technology. Even though download and streaming services were available, our sales were still going down every year until it got to the point where the average person realised it was simply more convenient to pay $10 or £10 a month for Spotify.

What happened at that point? 

 Then we started to recover. But my point is that when we have a disruptive technology, the disruption usually happens quickly, and the opportunities often come more slowly. So, I think we might be going into another period of disruption before we come out the other end of the tunnel and can make the most of all these new technologies.

 One of the things about artificial intelligence, for example, is the ability to take a sound recording that was recorded in stereo and analyse it, and break it down into stems. So you can break out the vocal track and the drums and the bass and the guitar and the piano, which is something that we just never imagined would be possible because it was recorded with two microphones and everything was mixed on the tape. By being able to do that, a whole new set of businesses become possible. Whether you take out the vocal track and have a karaoke business from it, or you take out the drum track and you have a learn how to drum business from it, or you give the stems to DJs and let them remix the song themselves or whatever it is.

Where do conversations around copyright happen in all of this? 

 Well, a lot of these new opportunities are complicated by copyright. Who’s allowed to do what? What permission? Who has the right to give the permission to do that? These are complicated issues, but we will get it if there’s enough business opportunity, people will figure it out. 

 Conversations around copyright will no doubt follow, but when we have these disruptions it seems different elements move at different speeds – the technology moves at one speed, the legislation and business practices and infrastructure moves at a separate speed.

 We had a business that was based on shipping CDs and vinyl for many years and then suddenly we had to change to digital distribution. That’s what I did when I went to Sony Music. I started the department in the late ‘90s to sell music on the internet. My job was to replicate our physical infrastructure for digital and figure out how to push the music out digitally to download and streaming services instead of out of a factory to Tower Records. But as I said, things like Napster made it much more complicated. Some of these new  technologies will be beneficial much quicker. So, as you say, Song Sleuth is literally solving a problem that humans using traditional computers can’t really solve.

“A lot of these new opportunities are complicated by copyright. Who’s allowed to do what? What permission? Who has the right to give the permission to do that? These are complicated issues, but we will get there if there’s enough business opportunity, people will figure it out.”

There are other technologies out there like Boomy, which is this AI that’s creating millions of new tracks. I’ve met the CEO, Alex, and I think they have both a creative and a disruptive side. Anybody can go on Boomy and create a track. If you pay a little money, you could own that track, then you can put it on Spotify and you can monetize it and get money. Provided someone listens to it of course.

Coming back to AI’s involvement in the creative process. What did you think of Nick Cave calling out AI as ‘a grotesque mockery of what it is to be human’?

 Sadly, in five years it might be perceptually indistinguishable. There are predictions that people say that even musicologists won’t be able to tell the difference. Fairly soon AI will be able to listen to all of Mozart’s music and create a new work that if you were to say to a musicologist, “oh, we found this undiscovered work in the back of a desk in Vienna and we had the symphony record it. What do you think?”, they won’t be able to tell if it was composed by Mozart or not, because the AI is going to learn all the ways to detect Mozart the same way that the expert does. And once it mimics all of those, it will be incredibly difficult to tell them apart. It’s just like that project where they fed a whole bunch of paintings by Rembrandt into an AI and it created a new “Rembrandt painting.” It may not have been as good as Rembrandt’s best paintings, but the real question is, could it pass as a Rembrandt? And the fact that Rembrandt probably didn’t even paint everything because he had people in his workshop doing things in the background, it gets really complicated. 

 So, Nick Cave had a visceral reaction, which is that this is an abomination – and the music is not that good. But soon it may still be an abomination, but the music might not be that bad. In terms of the threat of replacing humans, I was on a panel for the Copyright Office and WIPO three years ago. This was just before COVID in January 2020. On that panel, my job was to explain all the ways that AI is going to impact the music industry. I went through a list of all the jobs, pretty much everything from coming up with ideas for lyrics, writing lyrics, coming up with ideas for melodies, writing melodies, writing lyrics and melodies, and creating sound recordings, to mixing, mastering, whatever.

“It’s just like that project where they fed a whole bunch of paintings by Rembrandt into an AI and it created a new “Rembrandt painting”. It may not have been as good as Rembrandt’s best paintings, but the real question is, could it pass as a Rembrandt?”

This was three years ago, mind you, and even at that time, it was already happening. There was a label executive who told me, “pretty soon there’s going to be music created by AI on Spotify.” And I said, “yes, there’s already millions of tracks on Spotify created by AI.” At that point, they were mostly meditation, yoga, and ambient music. Now, we see maybe 10s of  millions of tracks on the streaming services, and in addition we see much of the production music is generated by AI. 

 So you tell the AI either I want something that sounds like X, or you give it parameters. It has to be ‘spooky but upbeat’ etcetera and then it creates your stuff. It’s starting with the low hanging fruit and working its way up. And it’s just going to keep working up. The more formulaic the music is, the easier it will be for AI. We think that formulaic is really bad. But again, I remember this interview with Chuck Berry that I love, where they asked him, “how did you know what to write about when you’re writing all your hit songs, how did you know what topics to write about?” And he said, “oh, that was easy. I just went to the ice cream parlour and I sat by the soda jerk, and I listened to the high school students talking. And they only talked about four things: music, falling in love, school and cars. So he said, why would I write about anything else? That’s what my demographic cares about.” So that was formulaic. 

 Then he knew that his song had to be three minutes long, and he knew that the hook had to come in 45 or 55 seconds in because of the radio and everything. He had a whole bunch of formulas. And then on top of that, he did his magic. His competitors were probably using formulas too, just not as cleverly. My point is, the AI is going to learn those formulas. And in some music, for example country music, the melodies are very formulaic. So writing the music for country music in many cases will be very easy for an AI. But writing the lyrics for country music  will be much, significantly more complicated because those lyrics are all based on human experience. Three years ago, I said it will be quite a while before an AI is able to do that. But now that we realize that an AI could go out and read the 5000 most popular novels ever written in English, pretty soon the AI starts to get a pretty good feeling for what humans care about. And they understand that when a child or an animal dies, people are sadder than when an adult dies. And when an old person dies, it’s generally less sad than a young person. And it’s going to figure all these things out and write a song about them. Just think of the reaction to Eric Clapton’s wrote Tears in Heaven.  So the AI figures out what moves people emotionally. It doesn’t need to have emotions to manipulate human emotions. And I think that’s an important thing we’re going to see with AI. The AI doesn’t need to feel sad. It just knows that if it’s told to write a sad song, that breakups and divorce and being physically distant from the one that you love and people dying, those are sad things.

Is it possible to predict how this kind of nascent AI technology is going to change not just the way things are made, but then the consumer and listening habits? For example, the TikTok-ification of people’s attention spans means people are becoming accustomed to listening to clips of songs.

 People are used to shorter attention. YouTube Shorts and TikTok have got them compressed. But what I think is interesting is it’s changing the way that people make music. Now people are thinking “wow they’re composing and recording the music, how will this sound in a TikTok?” All in the hopes that one of their releases goes viral.

 What do you think we as an industry can do to prepare ourselves for what’s ahead?

 Well, I think the first thing is that people have to be prepared for a bumpy road. It’s going to be quite disruptive, much more than people think. The major labels, for example, can’t decide if AI is a threat, an opportunity or both. And, of course, the answer is: it’s both. They could use an AI company to strip out the audio on all their old recordings and then monetize them in karaoke. Great, that’s an opportunity.

“People have to be prepared for a bumpy road. It’s going to be quite disruptive, much more than people think. The major labels, for example, can’t decide if AI is a threat, an opportunity or both. And of course the answer is it’s both.”

But if people are putting up songs that sound like Bad Bunny, then that’s an existential threat. So in this whole threat versus opportunity discussion, the reality is we just don’t know. Except that, as I said, my experience is that threats happen faster because the threats are outside the existing business infrastructure. And to leverage the opportunities, you actually have to change the business from the inside. For example, you might have a great virtual reality or augmented reality business opportunity with your artists, but that might mean that you have to negotiate new contracts with them so that you can put them in the Pokemon-Go game when people walk around collecting their songs or whatever. The newest artists might have that in their contract, but we know that the ones 5-10 years ago are not going to because that technology didn’t exist.