Can AI replace a singing teacher?

Lately, I’ve been getting a lot of signals that singing teachers are worried AI might replace them. I’ve received emails, I’ve spoken privately with a couple of teachers, and I see a lot of comments about it on our social media posts.

Last week, I received an unusually long email asking whether our company is a threat to the singing teacher community. The person who wrote to me also asked whether Singing Carrots Studio, which we give teachers for free, is safe to use or whether teachers should avoid it. Specifically, they wanted to know whether their lesson recordings might be used to train AI models.

I have to say, the email was constructive. It wasn’t aggressive or anything. It was thoughtfully written, and I felt it deserved a dedicated reply. So I spent quite some time putting my thoughts down and sharing them with the person who wrote to me.

Then I realized I’d basically outlined how I see this as the person running Singing Carrots. So I thought I’d share what I said here, without naming the person or making it obvious who they were. Here are my thoughts.

Why I think singing teachers are hard to replace

How many people come to a singing lesson mainly to sing with another person, to have a good time, to be heard, to experience singing together? How many would still come even if they weren’t trying to learn a particular technique?

Of course, it varies from teacher to teacher. In my experience, more than half come for exactly that reason. And I think that sets singing teachers apart from people in a lot of other professions. Think about a clerk working at a counter. As a customer, you usually go there because you need to get something done. If you can get it done automatically tomorrow, you probably will. When people come to spend an hour singing with their teacher, that teacher is in a very different position.

I think that’s true for music teachers generally, but for singing teachers in particular. Of all the jobs out there, it’s one where the human connection is such a big part of what people come for. From that angle, I wouldn’t worry much about AI replacing singing teachers. The person those students came to sing with is their teacher.

Second, there’s the technical side. I come from a software engineering background. I build products and I write code. Writing software is one of the first things the AI labs have gone after, and you can see why. The output you’re trying to produce is text. You can run the code and check whether it does what you asked. We in the technical community are still figuring it out, but we’ve already made a lot of progress in automating that kind of work.

Now think about what a singing teacher does in a lesson. They listen to the sound. They look at the student. They sing something for them so they have a reference. They draw on their training and experience to decide what to try next. And how do they check whether that choice worked? They might hear a difference in the room, but it can take months of practice before they know whether what they did actually helped that person improve. From the perspective of somebody trying to train an AI, that’s a long time to run even one feedback loop. Just technically, a singing teacher’s job is an extremely hard thing to automate.

Third, there’s the money. A lot of people have fun singing. Relatively few are willing to spend a lot of money on it. Singing education is a fairly small market for a big corporation to go after, especially given how hard the problem is. That’s one of the reasons small companies like ours are working in this field.

So imagine you literally made a list of all the jobs in the economy and ranked them twice. How hard is each job to replace? And how much money is there in replacing it? I think singing teachers would end up near the top in terms of how hard they are to replace and near the bottom in terms of the money to be made by doing it.

Now imagine that half the jobs on that list have been automated and singing teachers are still some way off. By that point, the world around us will already be a very different place. My guess is that singing teachers are more likely to feel the effect through students whose own jobs have changed and who can no longer afford lessons. And even then, I think we could still be a long way from an AI that can conduct a lesson as effectively as a singing teacher does in the room.

What Singing Carrots is actually for

As for what Singing Carrots and our potential competitors are doing, I can speak for our company. Our goal has never been to get rid of singing teachers and have everybody learn from AI. The problem we set out to solve is that a huge number of people don’t have singing in their day-to-day lives. I started this company because, in my own experience, singing is simply an amazing thing to have in your life. I find it therapeutic. It helps me calm down and regulate how I feel. It’s also a way to spend time with other people. We’re social animals. We need community, and singing gives us a reason to get together.

My rough sense is that fewer than a third of people actively sing as part of their lives. Of the other two-thirds, I think roughly half are what I’d call latent singers. They’d enjoy singing, but somewhere along the way they came to believe they can’t do it. Those are the people I have in mind when I think about our customers. I want to convince them that they can sing and that they can enjoy it. We give them somewhere to start without a big financial commitment and without having to sing in front of somebody they’re afraid might judge them.

Now, if you ask me how I measure success, how I know we’ve actually helped that person, my answer is that they’ve started singing in real life. And one of the most natural places they can do that is a singing studio, taking lessons with a teacher. To me, that’s a natural graduation step. They start with AI because it’s low pressure. They spend a month singing, build a little confidence, and then one day they feel ready to take a lesson with a teacher.

I expect some people who might otherwise have booked a lesson will use an app instead. My bet is we’ll also help people get to a first lesson who would never have considered one. In essence, we’re trying to make the market for singing education bigger. And I think the studios that become that next step will be the ones where teachers recognize how much people value having a place to sing with other people and a community they can belong to.

Of course, we also have people who already sing and use us for mechanical and technical practice. Singing teachers know how much regular work that takes, three or four times a week. Most teachers don’t have the capacity to see each student that often, and most students don’t have the money to pay for it. I think automating some of what I call the gym part of singing is a good thing for everyone. It’s routine work that needs doing between lessons, and it shouldn’t have to cost a lot of money.

What happens to lesson recordings?

For context, Singing Carrots Studio is a free product from Singing Carrots. It’s basically a Zoom alternative where singing teachers can host their live lessons. Once the lesson’s over, we store the recording and transcribe it. We give the transcript to an AI model, which writes a short summary of what the teacher and student worked on, what the teacher assigned, and what to focus on next. So, if the student also happens to be a Singing Carrots customer, the next time they practice with our AI coach, the coach already has the context of what happened during their singing lesson.

Now, an obvious question is whether we also use those recordings to train AI models. We don’t today. We might someday, but we don’t have any particular plans to do so. I can see how the data could be useful, and if a teacher records a lesson with us today, I can’t promise that recording would never be used for training.

We give Studio away for free because we see it as a marketing channel for Singing Carrots. Say a teacher uses it with ten students. One of those ten might want to do some extra pitch training at home between lessons. That’s a potential customer for us. If that student buys a subscription, it can cover the cost of video hosting for the teacher and the whole class.

When we tested it, we almost broke even. So we figured, okay, let’s keep offering Studio to singing teachers on a pro bono basis. Teachers get a free place to host lessons, and some of their students discover Singing Carrots that way. That’s what’s in it for us today. If we ever start using recordings for training, that would be another benefit for us. I understand why that possibility matters to teachers deciding where to host their lessons.

From my own side of this

I think my job as a software engineer is much higher on both of those lists. It’s easier to automate, and there’s more money in doing it. Even knowing that, I don’t think fighting it makes much sense. There are two reasons I feel that way.

First, the way I think about the economy, you have a bunch of people and companies with problems, and another bunch of people who make a living solving them. Now imagine AI solves some of those problems, including the ones I’m working on. That might happen. Taken on its own, I think that’s a good thing. There are fewer problems left to solve.

Of course, if solving one of those problems was how I paid my bills, I still have a very real question. Where do I fit now, and how do I make a living? I think that’s a question for our political institutions, for the people running the big AI labs, and for the people lobbying our politicians. It’s a question about how we run our society and how we share the benefits. For me, part of the answer is taking voting seriously, choosing our politicians wisely, protecting democracy, and figuring out redistribution.

Second, I have to decide what I personally do about it. This could affect me even faster than it affects singing teachers, even though I’m the one building an AI application. There’s nothing protecting a little company like Singing Carrots from a big American corporation like OpenAI or Anthropic. Imagine two more model releases and suddenly the models can do everything our product does and write all the code our employees write. Maybe we all lose our jobs, and the people who want to learn singing online just go and study with ChatGPT.

That might happen. To me, that’s much more plausible than Singing Carrots replacing singing teachers. I don’t think that’s happening tomorrow. And I’m not especially worried, because I think I’ll keep doing what I can as a responsible citizen, I’ll keep helping solve problems, and I’ll adapt. I’ll figure out how to plug myself into a new economy. That’s why I’m working on my entrepreneurial skills and learning how to put these AI capabilities to work.

So, can AI replace a singing teacher? I’m building a company around an AI singing coach, and honestly, I think AI is more likely to replace me than it is to replace the teacher my customers go on to take lessons with. For a lot of people, getting to spend an hour singing with another person is the whole point.

What I worry about more is what happens to people whose jobs are easier to automate, mine included. If AI takes over the work someone does to pay their bills, how do we protect their livelihood? How do we make sure the profits the big AI labs make are shared with the people whose work is being automated? Those are political questions, and I think how we answer them will matter far more than what a company like mine does in music education.

As for what to do in the meantime, I’d give a singing teacher the same advice I’d give myself. Try not to spend too much time worrying about being replaced, do your best, keep exploring the world, and take care of the people around you.