I understand both sides of this. I have spent my career in engineering and my life in music. Here is what AI can and cannot do, and why it does not scare me.

I am probably the wrong person to ask about this and also exactly the right person to ask about it.

I have spent years as an electrical engineer. I understand how these systems work at a technical level that most musicians do not have access to. I also make music from the most human place I know how to reach, songs written through physical pain, through watching people I care about head somewhere bad, through the specific kind of fear that comes from not knowing if your body is going to cooperate with your life anymore.

So when someone asks me whether AI is going to replace musicians, I have two answers. The engineering answer and the music answer. They are different answers and both of them are true.

Let me give you both.

THE ENGINEERING ANSWER

AI music generation has gotten genuinely impressive and it is going to keep getting more impressive. That is not a opinion. It is a description of how these systems develop.

The current generation of AI music tools can produce audio that is technically proficient. They can generate a track in a specified genre, at a specified tempo, with a specified mood, in seconds. For certain commercial applications, background music for a corporate video, ambient tracks for a meditation app, functional music that needs to exist and not draw attention to itself, these tools are already competitive with human-produced music on a cost-per-result basis.

The underlying technology is large language and audio models trained on enormous datasets of existing music. They learn patterns. Chord progressions, melodic structures, rhythmic relationships, production conventions. They get very good at recombining those patterns in ways that are plausible and sometimes genuinely interesting.

What they do not do is understand why any of it matters.

A language model generating lyrics does not know what it feels like to watch someone you grew up with slowly disappear into addiction. It has been trained on the linguistic patterns of songs written by people who knew that feeling, and it can produce text that resembles what those people wrote. But the resemblance is pattern recognition, not experience. There is no inside to the output. There is only the outside.

This distinction matters more than people on both sides of the AI debate tend to acknowledge. The people who are most afraid of AI replacing musicians often underestimate what AI can already do. The people most dismissive of the concern often underestimate what music is actually for.

THE MUSIC ANSWER

Music does not exist to produce audio. Audio is just the delivery mechanism.

Music exists to create a connection between one human being's interior experience and another's. The reason a song can make you cry when you hear it is not because the frequencies are hitting your ears in a particular pattern. It is because somewhere underneath your conscious awareness, you recognize that what you are hearing came from inside a person. That the person who made it went through something and found a way to put it into sound. And that experience they went through is something you have also been through or are going through right now or are afraid of going through someday.

That recognition is not mystical. It is how human brains are wired to respond to communication from other humans. We are a deeply social species and we have been making music for as long as we have existed because music is one of the most efficient ways we have ever developed to transmit internal experience across the gap between one person and another.

An AI does not have an internal experience. It has a training set and a generation algorithm. It can produce audio that superficially resembles music made by humans who did have internal experiences. But it cannot make music that carries that transmission because there is nothing on the inside to transmit.

I am not saying this to comfort musicians who are scared. I am saying it because I believe it is true and because I think it is actually the important claim. Not that AI is bad or that its outputs are always inferior, in measurable technical terms they sometimes are not, but that the thing music does at its deepest level is something AI cannot do because the thing music does requires a person.

WHERE AI ACTUALLY THREATENS MUSICIANS AND WHERE IT DOES NOT

This is where I want to be practical rather than philosophical, because the threat is real in some places even if it is not total.

AI will displace human musicians in functional music contexts. Background music, elevator music, stock music libraries, low-budget commercial work, gaming ambient tracks, any context where music is primarily serving a utilitarian purpose and the listener is not attending closely to it. In these markets AI is already competitive and it will become more so. Musicians who have been earning in these markets are going to feel real economic pressure and that is a legitimate thing to be concerned about.

AI will not displace human musicians in contexts where the human origin of the music is the point. Concert performance is not going away. The experience of being in a room with a person who is making music in front of you, with a body that can fail and a voice that carries the weight of specific lived history, is an experience that AI cannot replicate because the human body is part of what is being offered. The 4-hour shows I play are not valuable because of the audio output. They are valuable because something is happening between the people in the room and the person on the stage.

Recorded music from artists with genuine audiences is also less threatened than people assume. The listeners who care about Tony Oso's music care because of what is behind it. The spinal condition. The songs written in pain. The years of engineering followed by the decision to put all of that aside and say something real with a guitar. You cannot replicate that backstory with a model because the backstory is not a style. It is a life.

The middle ground is more complicated. Artists whose recorded output is stylistically accomplished but who have not yet built a genuine connection with an audience, artists who are skilled but not yet irreplaceable from the listener's point of view, are going to face increased competition from AI tools that can produce technically similar results at a fraction of the cost. This is real. It is already happening in some genres. The answer is not to compete on technical output. The answer is to be more human, not less.

WHAT THIS MEANS FOR HOW I MAKE MUSIC

I have not changed how I make music because of AI. But I have become clearer about why I make it the way I do.

Every decision I make in the studio is a human decision. The choice to leave a performance on tape that has a slight imperfection because the emotional truth of the take outweighs the technical precision. The choice to use a guitar I have played for fifteen years because the relationship between my hands and that instrument is embedded in how it sounds. The choice to write about specific people in specific situations rather than reaching for the universal language that might appeal to more listeners but would mean less to me and less to the people those songs are for.

These are not decisions an AI makes because they require a self. They require someone who has stakes in the outcome. Who has something to lose by being honest and chooses honesty anyway. Who has a history that shapes what notes feel right and what words feel true.

That self is what I am putting into the music. And that is what I am asking listeners to connect with when they hear it. Not the technical proficiency of the recording, though I care about that too. But the specific human being underneath it, with the engineering background and the spinal injury and the years of 4-hour shows and the two hundred records and the songs written at 2am when something needed to come out.

No model is going to reproduce that because no model lived it.

WHAT MUSICIANS SHOULD ACTUALLY DO RIGHT NOW

Register your works with a PRO if you have not. ASCAP, BMI, and SESAC in the United States. Your performing rights are your intellectual property and protecting them formally matters more as AI use of training data becomes a legal and legislative battleground.

Keep detailed records of your creative process. Drafts, voice memos, dated notes, session files. Documentation of human creative origin may become increasingly relevant as legal frameworks around AI and copyright develop. Nobody knows exactly how this resolves yet. Being able to demonstrate the human process behind your work is a reasonable form of protection.

Build direct relationships with your audience. Email lists, not just social media. Direct-to-fan platforms. Physical music. The closer your listeners are to you as a person rather than to your audio output as a product, the less substitutable you are. A listener who is there because of you is not going to replace you with an AI that sounds similar. A listener who found you through an algorithm because your audio matched a preference profile might.

Keep making music that only you could make. This sounds like generic advice but I mean it specifically. The more your music is tied to your specific history, your specific relationships, your specific emotional territory, the harder it is to replicate. Vague is easy to approximate. Specific is much harder.

The artists who are going to be most affected by AI are the ones who have been most interchangeable. The artists who will be least affected are the ones who have been most irreplaceably themselves. That was true before AI and it is more true now.

Be irreplaceable. That was always the job.

Tony Oso

Stream Tony Oso at tonyosomusic.com/music

Subscribe at tonyosomusic.com to stay connected to music made the hard way, by a person, from the inside out.