Committing to Change: A Conversation with Mark Ethier

This is a discussion with Mark Ethier, who currently serves as Executive Director the Berklee Emerging Artistic Technology Lab (BEATL) at Berklee College of Music. Mark is also well-known in the music technology industry as the co-founder and former CEO of iZotope. He's been working with machine-learning since his days at MIT and those techniques have found their way into many of iZotope's products.

Berklee has been out in front of music technology changes since its inception. Berklee graduates have not just gone onto be Grammy winning artists, but also educators, entrepreneurs and product builders in many fields in and outside of music Bringing in a leader with Mark's background help design a curriculum around artificial intelligence is yet another example of the institution's commitment. Here's an interview we did with Mark several years ago.

The opportunity to have another chat with Mark came after an announcement of Berklee College of Music’s upcoming AI and Music Summit (AIM) June 3-5 in Boston. Although tickets may be available for the companion concert featuring Jordan Rudess and L'Rain, the actual summit has already sold out, another illustration of the effect AI is having on all aspects of music making. The interview explores how AI is already reshaping music education, from students experimenting with generative coding tools to assistive AI technologies that restore creative capabilities for disabled artists.

Mark stresses the importance of human performance and artistic communication. He describes Berklee’s AI summit as an attempt to bring musicians, educators, developers, lawyers, and entrepreneurs into the same conversation, so artists can have a stronger voice in shaping the future of AI in music. Is there a Moore's Law or something equivalent for the growth of AI?

Because it is improving at an exponential rate. A semi-log plot of transistor counts for microprocessors against dates of introduction, nearly doubling every two years I will admit I was led down a little research project about this. You go back to Moore's Law, which is that compute would double every two years at half the cost.

These days, it’s actually gone more to three years because we’re now at the level where there are physical barriers to our ability to keep that same trajectory. What I discovered is that currently, the speed of AI capability processing has been doubling every three and a half months. Wow!

So, that's like an order of magnitude every year. It’s both the compute growing, there’s more compute, and better algorithmic efficiency, sometimes driven by the AIs themselves improving efficiency, which is still kind of mind-blowing. Then the data scaling, both real data and synthetic data scaling, and all of these things are sort of combining together.

Investors are pouring money into AI start-ups. I'm wondering about a bubble. Do think this could affect the music industry in any way?

I think in some ways, by definition, we're in a bubble. We are at a time where there are a lot of people competing for the same market, a "winner takes all" sort of approach. You end up with a dozen players investing huge amounts of resources to go after the same prize, knowing that nine out of ten of those are not going to make it.

When that’s happened before, often there is infrastructure left behind that people still find value in. It’s funny to talk about the dot-com bubble because I remember I was an undergrad at MIT and there was a service where you could get anything delivered within an hour at the same cost it was at the store. You stop and think, "This shouldn't be possible."

Turns out it wasn't; it was just an inflated service that existed for a moment. I don't know if there's a bubble in music as much as there is in parts of music tech. I think a lot of it in the music products industry is certainly related—and I’m talking hardware here.

During the pandemic online retailers were growing incredibly fast, and then people could get out again, so they weren't home playing music as much. That's right. I heard a good description that in the hardware and software side, it was almost a double problem because you had all these people who leaned into their hobby, and now they're going back to their regular day job and throwing their hardware or software up on eBay.

Now these companies are competing against the equipment they sold two years ago. That's just rough. In the research you've done so far, what’s the most surprising thing you’ve learned and what is the scariest?

iZotope Ozone The speed and the breadth of the change that is happening. I can look back and say there was a creation revolution that happened when the digital audio workstation (DAW) became affordable. That was a pretty big shift in the democratization of music.

Then you had Napster and LimeWire catalyzing the changes that led to Apple's new music business model and Spotify. That was monumental in the way the economics of music happened. It seems like this is the moment where both of those things are happening at the same time, but in the period of months, not years.

Even a year ago, I would have said, "Oh, but it can't do this yet." And I've been proven wrong so many times. I want to pull apart the distinction between Generative AI (trained on licensed material vs.

scraped data) and Assistive AI. iZotope was using assistive AI more than a decade ago for things like RX, which among other things, removes "mic rustle" (the sound when a lavalier mic scratches against clothing). That was AI, but it wasn't scary.

We had virtual instruments that were pretty darn good, and now there are generative tools that will render as a virtual instrument using generative AI. It’s enhancing a workflow that had already been changed. What effect do you see AI already having on enrolled Berklee students?

Are you seeing new students coming in who could teach classes? Berklee's a very diverse place. There are students that are very engrossed in generative AI, and sometimes I have to tell them, "Hold on, slow down.

Do you know how it's trained? Do you own the output? Is the output copyrightable?"

And there are students who came to Berklee because they're passionate about guitar and jazz and haven't really engaged with technology in that way. We have seniors graduating this year who started at Berklee when generative music was not really a thing. At the same time I've spoken to a couple of high school music teachers who are starting to use some generative music tools in their classes because otherwise they didn't have access to music making.

We may see more students show up who have made it part of their creative process, but I still think this is still very, very new. Is the value of all this going to be musicians adding to their music, or building tools that can possibly make their music better or more unique? We are in the very beginning of figuring this out.

You have people like Jordan Rudess from Dream Theater, who has been collaborating with MIT to train his own personal model that can do live improvisation with him on stage. That's additive; that's not replacement. Berklee just hosted a conference called ABLE Assembly, and there are examples of people using generative AI to regain the ability to make music that they've lost.

If they've lost their voice for some reason, they're actually able to train these models on their own previously recorded voice. They can still bend the character and the emotion. I want to get beyond the place where there are tools where you type in a couple of words to get a song out.

I think that's a novelty. At some point people will want to see somebody play those arrangements. Having the skills to learn a song or a set quickly and perform consistently...

maybe there’s more work for a Berklee-trained musician…? A machine can run faster than a human, but we still go to track meets to watch who can run the 100-meter dash fastest. It’s deeper than that with music and performance.

Even though you can get a machine to produce an audio output, it doesn't mean that it's a real human communication. Let's talk about the Berklee AI summit. It’s in Boston June 3rd through 5th?

June 3rd through 5th is the summit, and then on the 6th and the 7th, we’re having a hackathon that’s being put on by Music Hackspace. The summit is sold out, but there may still be tickets available for the concert, which will feature Jordan Rudess and L'Rain. What are the most important things you want people to know about the summit beforehand, and what are the most important things you would like attendees to know afterwards?

We see a hole in the public discourse around music and AI, the artist's perspective is often missing. The summit is a chance to bring together musicians, educators, developers, lawyers, and ethicists. We're going to have people who are very skeptical and very against the integration of AI, and we are going to have companies where the integration of AI is their business model.

We want artists and educators to leave with a sense of agency. I feel a lot of places where people feel somewhat hopeless, like "this thing is all happening and we can't stop it." We want to show people that it's possible for us to go to these companies and researchers and have a voice in this conversation.

Berklee has always embraced change... David Mash teaching music software at Berklee in the 1980s That's right. It's shocking how many people in the music tech world are Berklee grads.

One other interesting story about the advent of generative coding at Berklee. Last semester, a faculty member taught a class on building business plans, and they decided to see if they could build a product using agentic coding tools. By the end of the semester, students with zero coding experience had built four or five fully functioning applications.

It really changes the calculus for what a student at Berklee can do in the world. And I just want to say thank you again for the opportunity to talk about the AIMS event.