Ed Newton-Rex, Chief Executive Officer, Fairly Trained Carol E. Reiley, Co-founder and CEO, DeepMusic.ai Moderator: Verne Kopytoff, Fortune
Category
🤖
TechTranscript
00:00And a big thank you to our guests. We're here to talk about music. And before we
00:05ask, you know, what your organizations do, I wanted to set the stage and ask, can
00:13AI music be groundbreaking? Can it be as unique as the Beatles or Jimi Hendrix or
00:22Beethoven? That's a big start. I'll start with you, Carol. Start off with an easy question.
00:31You know, I think when I think about AI, it was like through the full entire stack. I
00:36think the obvious one is like music generation and helping artists compose and write songs.
00:42Then I started like digging into like music production crowd. Then you looked at like
00:46the labels and then like the front of house marketing tools. And I think AI is we agree
00:54is like the superpower. So how do you use AI to become super creative? So for me, I
01:00was really looking at how do artists become super creative? Because I viewed them as like
01:04the most creative humans on earth who have made this their profession. How would they
01:08do it? And then see how that impacts society and humanity later on. I think it has great
01:15potential. And I think for the last few years, I've been digging into exactly that question
01:19on where's the value and how can we make the pie bigger? And, you know, recent surveys
01:25have also showed that the number one job that young people want to do is to be an influencer.
01:33So how does that type of creativity also utilize AI to create content and put that out?
01:40What about you, Ed? What's your perspective?
01:41I think I've seen innovative things be done with AI music. Absolutely. When I was at Stability
01:49AI, our audio team released a product called Stable Audio, and it was basically a sample
01:54generator. And, you know, you could get it to do some pretty wacky things. Like one of
01:58my favorite things that one of our team produced was I think the prompt he used was hammering
02:03wood at 128 beats per minute. And it came out with this weird combination of a kind
02:09of hammering sound effect, but in a drumbeat form. And I think sample generation in general
02:15is something where this tech is already being adopted, and you can get some innovative stuff
02:19out of it. I do, you know, I think unfortunately, though, a lot of what is produced with a lot
02:24of the systems out there at the moment is incredibly derivative, right? People call
02:27it slop. And I think that's like a pretty good term for it. So I think that the majority
02:33is not going to be, you know, musically innovative in that way, I think.
02:38I might also follow up and just also push that like AI could, the same thing can be
02:44said about humans, right? I think a lot of the stuff that we also produce, at least to
02:47start with, is also incredibly derivative or trying to copy chords. So I agree. I think
02:53that maybe AI could be used as a challenger to push us to be better. Could be one lens
03:00to also look at. But I do feel like it's an inflection point where it could be a lot of
03:05slop and derivative music, and AI might just streamline everything to be the same derivative
03:12music. So I'm looking for us to really push that boundary.
03:16Right. And well, tell us what your organization does, Deep Music AI.
03:20Yeah. So, you know, I think five years ago, I had just sold my company, which was in self-driving
03:31cars, which had a large impact on society. And I was looking for what could be the next
03:35big disruption. And I went through my own existential crisis. I'm like, what if AI reached
03:40the point where it could replace human creativities? Because we've sat on panels, and I always
03:46had said, if your job is in the creative, you're going to be safe. And I was like, what
03:50happens if that assumption isn't true? What would that look like? And what if we were
03:54to let AI replace the creative side, the stuff that I enjoy doing? What would that look like?
03:59Or how could I find a way to utilize and lean into AI so that that could become a new area?
04:10So I really wanted to partner with artists and people who were more creative than me.
04:15I'm an entrepreneur, and I think that's, like, one set of creativity. But I couldn't
04:18figure out how to, like, systematically break down how I went through my creative process.
04:24I know that sometimes I just get inspired and a light went on my head. But I started
04:27talking to different types of creatives, trying to figure out how they went through their
04:30process and had landed with the San Francisco Symphony. I was working with them and composers
04:38and musicians, trying to understand that. And so I created this nonprofit with one of
04:43my dear friends from grad school, who is an artist, Hilary Hahn. It was during COVID,
04:48so she was not touring. And so we had, like, a three-hour beautiful conversation about
04:54what happens if AI was to disrupt her career. Because I had seen that in the self-driving
04:59car space, where eight graduate students from Stanford, we were able to put a car out in
05:04three months on the streets driving around Mountain View. So I could see the acceleration
05:09of AI, and I knew that it was coming for creatives very soon, and wanted to help the
05:15artists not be reactionary, but if they were to lean into it and band together and understand
05:19the tools and be part of the companies that were building these tools, how would that
05:24change, like, the course of humanity? That was my 2018 revelation and what I've been
05:30trying to unpack since then.
05:31Fantastic. And tell us about Fairly Trained.
05:35Well, Fairly Trained is a nonprofit I set up about a year ago. I've worked in generative
05:40AI since 2010, and for about 12 years since I joined generative AI to about 2022, people
05:49didn't train their models on copyrighted work without a license. Everyone knew it was illegal.
05:55No one did it. They trained research models, but they would never commercialize it. But
05:59then in 2022, a few big companies you've all heard of started releasing commercial models
06:04and threw caution to the wind. And I ended up working at another generative AI company,
06:11and eventually I quit that company because I thoroughly object to this unlicensed training.
06:17I think it's illegal, and I think it's unethical. And so I set up Fairly Trained to, you know,
06:22we're a nonprofit. What we do is we certify generative AI companies that don't take that
06:25approach. We certify generative AI companies that don't claim that they can scrape the
06:29life's work of the world's creators and train vast commercial models on it.
06:34The good guys.
06:35Well, I think so, but everyone thinks they're the good guys, you know.
06:38And then you've also supported or had your petition signed by 12,000 creatives, including
06:48The Cure's Robert Smith and Radiohead singer Tom York, calling for compensation for artists.
06:55Yeah, it's tripled. We have 36,000 signatures.
06:58Oh, 36,000. Okay.
07:00This was a statement we put out like a month ago, and it's a 29-word statement. It's still
07:05available for signatures. If you happen to agree with this, we're in San Francisco, so
07:09you probably don't. But if you happen to agree with this, the statement is, the unlicensed
07:13use of creative works for generative AI training is a major unjust threat to the livelihoods
07:19of the people behind those works and must not be permitted. That's the entire thing.
07:23It's been signed by a lot of creators, including, you know, a lot of well-known creators. But
07:28more importantly, like thousands upon thousands of working creative professionals who you
07:33probably haven't heard of and whose lives are being impacted by this right now. So if
07:39you do agree, please go and sign it. It's still open. It's still rising.
07:43And now the music industry has sued a couple AI music engines. What should we think about
07:48that, Carol?
07:51I agree that they need to be compensated, acknowledged. I do think that technology has
07:59disrupted the music industry many times over, from like sampling electronic music. And this
08:04is another new tool that is coming, and they need to be, I think, fairly compensated. I
08:11think it's great that they're standing up and doing something about it. Another approach
08:18is I think it's also interesting that some artists have openly modeled, like created
08:25models of their voice so that the public can utilize it and they can rev share things.
08:29So I feel like there's new business models that need to be created. We need to be experimenting
08:34with them. But I do 100% agree that the rev needs to be split. They need to feel fairly,
08:45not just monetarily compensated, but I think acknowledged for their work. There need to
08:50be ways to spot if that has been not upheld. I think what Ed is doing is fantastic. And
08:59I think really banding the artists together to have a community and a place to talk and
09:03have conferences like this is important.
09:07What kind of compensation scheme do you envision? Is there a right one?
09:11I don't think there's a right one, actually. I think often AI companies will turn around
09:16and say, well, come on then, rights holders, creators. What's the silver bullet for how
09:21this should work? And I don't think there has to be one. The beauty of licensing, and
09:24licensing is clearly the solution, I think, here. The beauty of licensing is that two
09:27parties can come together and figure out a deal and figure out what works for them. That's
09:31not going to be the same for everyone. One person is going to want an upfront payment.
09:34One person is going to want a revenue share. Stability AI, so I did licensing deals like
09:39this with creators, and we put in place a revenue share with music creators whose music
09:46we trained on. And I think that kind of model can be great, because no one really knows
09:51how much money a lot of these things are going to make, but a revenue share can work really
09:53well.
09:55There is this argument that AI companies often give in legal filings as well, where they
10:01will say, well, it's not feasible that we could pay everyone. These payments would be
10:07so small. What's the point of paying people? And to that I say, you've obviously not been
10:12a creator yourself and got small royalty checks from Spotify or whatever it is. Just because
10:19a payment is small doesn't mean a creator doesn't want it, and it doesn't give you the
10:22right to take their work without permission. So that's my view. I think you've got to start
10:26from licensing and consent, and you work from there. And if people don't want to be involved
10:30because you're not offering them enough money, then you don't use their work, and it's as
10:34simple as that.
10:35Carol, what will AI ultimately mean for the jobs of musicians, do you think?
10:43I do think a lot about the future of work and where people go and how we utilize our
10:48time. It's undeniable that it will be disrupted. There will be a lot of jobs replaced, particularly
10:58things related to maybe like music for commercials or royalty-free music, that type of thing.
11:05I don't know how much joy those artists have creating music, but there is a paycheck there.
11:10I do think that those type of jobs might be more easily replaced because it will be cheaper
11:18using AI-generated music, or you could have personalized music and your own lyrics. But
11:25I do think there's always room for true creators. I do think that more people will be jumping
11:32in. I feel like music seems... I've worked in many different applications, and music
11:36seems to be really one that resonates with the majority of the public. I feel like music
11:40touches people, it gets them through hard times, or they themselves want to express
11:46themselves a certain way. So I think there'll be more artists than ever before, and people
11:52would be able to use the tools. And I think what we need to work on now is really the
11:56interface to these tools. If you guys have played around with a lot of the tools that
12:03are available, I think they're fun for a while, but the stickiness level doesn't seem to quite
12:07be there. And what I'm interested in is how do we get people to come back to the tools?
12:13I think there is something with personalization. People love AI bias for things that are personal
12:19for their life or their album, photo album, and they want to personalize music, but it doesn't
12:26seem to be something that has any stickiness right now. So I think a lot of work needs to be done on
12:30the interface side. And Ed, what percent of the music industry do you think is doing it right
12:36now, following copyright that you believe they should? Well, I mean, in the AI world,
12:44there are lots of companies who are licensing all their training data. So Fairly Trained has
12:49certified 16 AI music companies involved in AI music who don't use copyrighted work without a
12:56license. It's the absolute minority of companies in the music space who are training on the world's
13:02music without permission. And the two most prominent are Suno and Udio, and they've both
13:07been sued for huge amounts of money by the record labels. And I'm sure if other people follow their
13:11footsteps, they'll be sued as well. So yeah, a lot of AI music companies are doing the right thing.
13:15I encourage you to check them out. They're on our Fairly Trained website. I agree with Carol. I
13:23think that there are good use cases for some AI music. One thing we saw recently was like Randy
13:31Travis, the country singer who had a stroke, using AI voice models to recreate his voice and be able
13:38to release new music. And there are these sort of empowering stories, but unfortunately, I think
13:43right now it's pretty unavoidable that the main use case of AI music will be essentially the
13:47commodification of music. There isn't text-to-music models. I agree with Carol. I mean, that retention,
13:54that real use case. For a long time, people were thinking, where's the mid-journey for music?
14:00Loads of people use that, and then a couple came along. And how many songs about your trip to the
14:05coffee shop do you really want? Honestly, text-to-music is not the answer. So I think there's
14:13a lot of work to do in the AI music world to find the really good application. And unfortunately,
14:18as I say, I think commodification is going to be the most scalable one, and that's something I'm not
14:24thrilled about personally. Right. So let's look ahead five years from now. Do you think we'll
14:30have an AI Taylor Swift? Is that possible? Carol? Yeah. I mean, I would say that the goal is never
14:42to mimic the artist. I think Taylor Swift is a powerhouse, and her job is...
14:47I'm saying more of a A-star on her caliber, on her level.
14:50Yeah. I think what is so amazing about her is that she is the entire full stack, it seems like,
14:55from songwriter to CEO to paying those bonuses, and to all the crew and everyone running this
15:03company. So she's like an empire, and I think that is interesting to see how she markets herself,
15:09how she connects with the community. And I think AI can help an artist that doesn't have her
15:16efficiency productivity, really use her as a business model to learn all the different aspects
15:21of her job, and maybe utilize AI, if you don't have as powerful of a team, to be able to do some
15:27of that. So while that is interesting to have... We've seen AI avatars, you can go to hologram
15:36concerts. I think Kim Kardashian, for instance, is interesting because she puts out so much
15:43content. So can you imagine a superstar like Marilyn Monroe being... The team behind her
15:50constantly putting out 24-7 content with her singing things and doing deep fakes, and now
15:55all of the eyeballs of our world are watching only five characters. So I see that future
16:02happening where there's only a few superstars, and it becomes harder for independent artists
16:06to break into that level and get the average consumer's attention. And I think what Deep
16:16Music is working on, besides the machine of the artist, is what are new ways to utilize AI that
16:24you can't do by yourself? So we are looking at neuroscience and brain activity, live performances,
16:31and getting the mood of a crowd to affect the performance during the stage. Live composition,
16:39visuals for music, robots dancing with ballerinas. I feel like new types of art forms
16:46and trying to figure out what is interesting that isn't a gimmick, that has staying power,
16:51that could be a new way to appreciate art, as well as behind the scenes for dynamic pricing
16:59and helping some of these art houses thrive in this phase.
17:07Great. I want to thank our panelists for joining us and talking about the future of music. I'm
17:13looking forward to how it turns out. I hope the musicians get their due. Thank you very much.